1. Overview — Complete Business Flow
Dynasty's workflow should start with one unified intake, then classify into business lines. Each line has its own timeline, but all share the same finance, approval, service and learning controls.
flowchart TD
A[Unified Intake
Email / WhatsApp / Website / Call / Walk-in / Partner / Staff] --> B[Create or match Client + Lead/Trip Workspace] B --> C[Capture minimum facts
contact, source, original request, destination/theme, date, pax, budget signal] C --> D{Classify business line} D --> BT[Bespoke / Private Tour] D --> M[MICE / Incentive] D --> G[Group Tour enquiry] D --> T[Ticketing] D --> A2[Ancillary service] D --> S[DMC / Supplier partnership] D --> O[Other / admin / irrelevant] BT --> BT1[Qualify Lead] --> BT2[Shape Direction] --> BT3[Source Quote] --> BT4[Price & Close] --> BT5[Operate Trip] --> L[Post-trip learning + database update] M --> M1[Corporate Qualification] --> M2[Scope Event/Trip] --> M3[Supplier/Tender RFQ] --> M4[Proposal + Contract + Payment] --> M5[Operate Group/Event] --> L G --> G1{Fixed group available?} G1 -- No / current default --> G2[Convert to Bespoke / Private if suitable] G1 -- Yes / future product --> G3[Check departure, seat, price, tour code] --> G4[Booking + payment + documents] --> G5[Tour operation] --> L T --> T1[Identify product] --> T2[Live availability + price source] --> T3[Collect payment first] --> T4[Issue / confirm] --> T5[Service changes] --> L A2 --> A21{Attached to active trip?} A21 -- Yes --> A22[Add ancillary task to Trip Workspace] A21 -- No --> A23[Usually decline / nurture / refer] S --> S1[Register supplier/DMC] --> S2[Collect profile + pricing + sample itineraries] --> S3[QC and score] --> S4[Update supplier / experience database] O --> O1[Close / route internally / archive] BT4 --> F[Finance controls
margin, invoice, payment, refund] M4 --> F G4 --> F T3 --> F F --> P[Approval Center] P --> BT5 P --> M5 P --> G5 P --> T4
Email / WhatsApp / Website / Call / Walk-in / Partner / Staff] --> B[Create or match Client + Lead/Trip Workspace] B --> C[Capture minimum facts
contact, source, original request, destination/theme, date, pax, budget signal] C --> D{Classify business line} D --> BT[Bespoke / Private Tour] D --> M[MICE / Incentive] D --> G[Group Tour enquiry] D --> T[Ticketing] D --> A2[Ancillary service] D --> S[DMC / Supplier partnership] D --> O[Other / admin / irrelevant] BT --> BT1[Qualify Lead] --> BT2[Shape Direction] --> BT3[Source Quote] --> BT4[Price & Close] --> BT5[Operate Trip] --> L[Post-trip learning + database update] M --> M1[Corporate Qualification] --> M2[Scope Event/Trip] --> M3[Supplier/Tender RFQ] --> M4[Proposal + Contract + Payment] --> M5[Operate Group/Event] --> L G --> G1{Fixed group available?} G1 -- No / current default --> G2[Convert to Bespoke / Private if suitable] G1 -- Yes / future product --> G3[Check departure, seat, price, tour code] --> G4[Booking + payment + documents] --> G5[Tour operation] --> L T --> T1[Identify product] --> T2[Live availability + price source] --> T3[Collect payment first] --> T4[Issue / confirm] --> T5[Service changes] --> L A2 --> A21{Attached to active trip?} A21 -- Yes --> A22[Add ancillary task to Trip Workspace] A21 -- No --> A23[Usually decline / nurture / refer] S --> S1[Register supplier/DMC] --> S2[Collect profile + pricing + sample itineraries] --> S3[QC and score] --> S4[Update supplier / experience database] O --> O1[Close / route internally / archive] BT4 --> F[Finance controls
margin, invoice, payment, refund] M4 --> F G4 --> F T3 --> F F --> P[Approval Center] P --> BT5 P --> M5 P --> G5 P --> T4
2. Business Lines at a Glance
| Business line | Status | What belongs here | Entry condition | Redirect / exit condition |
|---|---|---|---|---|
| Bespoke / Private Tour | Core focus | Customized private travel for individuals, families and affluent small groups. | Client wants custom travel; has destination/theme/date/pax/budget signal; likely Dynasty-fit. | Pure lowest-price shopping, no willingness to share basics, impossible request, or budget mismatch. |
| MICE / Incentive | Core | Corporate incentive, meetings, events, retreats and group movements. | Corporate entity, pax scale, objective, decision chain, target date and budget signal. | Unclear decision owner, unrealistic budget/timeline, or pure low-cost group travel. |
| Group Tour | Traditional model paused/rebuilding | Fixed departure group tours or future curated thematic groups. | Client asks for fixed departure or existing product. | If no active product, convert to bespoke/private if suitable, otherwise nurture. |
| Ticketing | Selective | Air ticketing and ticket-servicing requests. | Live source exists; client accepts service fee/payment terms; payment before confirmation. | Low-margin unsupported request, no service-fee acceptance, or credit risk. |
| Ancillary | Attached | Insurance, visa guidance, VIP airport service, fast track, restaurants, special access. | Usually attached to active trip or strategic client. | Standalone visa-only / low-value admin unless strategic exception. |
| DMC / Supplier | Infrastructure | Supplier onboarding, registration, pricing, sample itineraries, experience database. | Potential partner has useful destination capability/pricing/content. | Incomplete profile, poor quality, duplicate, or no commercial fit. |
1. Bespoke / Private Tour — Core Timeline
This is the primary workflow and should be the strongest visual anchor in the ERP. The main stages remain simple; complexity sits inside subtasks, revisions and approvals.
1Qualify Lead
Is this a real customer? Is it Dynasty-fit? Do we have enough information to proceed?
2Shape Direction
Match the client to the best reference itinerary first; if no match, either keep qualifying or ask DMC when intent is high. AI should not invent itineraries from domain knowledge.
3Source Quote
Once required brief fields are complete and intent is clear, source executable options from two DMCs / suppliers where possible.
4Price & Close
Run QC and margin checks, prepare client quote, revise, confirm and collect payment.
5Operate Trip
Confirm resources, prepare departure, support during travel, follow up and capture learning.
2. Bespoke / Private Tour Flowchart
This flowchart shows how a Bespoke case moves through the 5-stage timeline, with loops for missing information, budget mismatch, revised scope, supplier issues and payment/approval gates.
flowchart TD
A[New bespoke enquiry] --> B[1 Qualify Lead
create workspace, classify, capture required intake fields] B --> C{Required intake fields present?
contact + channel + original request + travel signal} C -- No --> C1[Ask missing mandatory facts once
or nurture/archive] --> B C -- Yes --> D[2 Shape Direction
search reference itinerary database first] D --> D0{Good reference itinerary match?} D0 -- Yes --> D1[Use best-matching reference itinerary customize to client needs] D0 -- No --> D2{High client intent?} D2 -- Yes --> D3[Go to DMC proposal/RFQ path
neutral, no booking commitment] D2 -- No --> D4[Continue client engagement
gather missing info and intent signals] D4 --> D5{Still low intent / poor fit?} D5 -- Yes --> D6[Politely decline / nurture
no DMC work] D5 -- No --> D D1 --> E{Required RFQ brief complete?
destination + dates/month/duration + pax/rooms + budget/hotel level + style/constraints} D3 --> E E -- No: missing mandatory facts --> E1[Ask targeted clarification
one message, highest-impact gaps first] --> D E -- No: budget gap --> E2[Expectation management
shorter trip, off-season, lower hotel, alternate destination] --> D E -- Yes --> F{Intent and commitment sufficient?} F -- Strong intent / deposit --> G[3 Source Quote
send same neutral RFQ to 2 DMCs where possible] F -- Weak intent --> F1[Nurture with direction options
no detailed supplier work] --> D G --> H{Any DMC reply meets mandatory completeness
and itinerary score threshold?} H -- No --> H1[Targeted follow-up for missing fields
hotel names, price basis, terms, inclusions] --> G H -- Yes --> H2[Use first qualified DMC proposal
second DMC remains comparison / backup] H2 --> I[4 Price & Close
QC, margin, client proposal] I --> J{Money / commitment / complaint / change risk?} J -- Risky --> K[Human approval required] J -- Not risky --> L[AI executes/logs with evidence] K --> M{Client confirms and payment verified?} L --> M M -- No --> M1[Payment reminder / revise scope
no irreversible supplier lock] --> I M -- Yes --> N[5 Operate Trip
confirm suppliers, pre-departure, support] N --> O{During-trip issue?} O -- Yes --> O1[AI contacts DMC, records evidence, escalates high-risk cost/complaint] --> N O -- No --> P[Post-trip feedback
supplier score, experience DB, SOP learning]
create workspace, classify, capture required intake fields] B --> C{Required intake fields present?
contact + channel + original request + travel signal} C -- No --> C1[Ask missing mandatory facts once
or nurture/archive] --> B C -- Yes --> D[2 Shape Direction
search reference itinerary database first] D --> D0{Good reference itinerary match?} D0 -- Yes --> D1[Use best-matching reference itinerary customize to client needs] D0 -- No --> D2{High client intent?} D2 -- Yes --> D3[Go to DMC proposal/RFQ path
neutral, no booking commitment] D2 -- No --> D4[Continue client engagement
gather missing info and intent signals] D4 --> D5{Still low intent / poor fit?} D5 -- Yes --> D6[Politely decline / nurture
no DMC work] D5 -- No --> D D1 --> E{Required RFQ brief complete?
destination + dates/month/duration + pax/rooms + budget/hotel level + style/constraints} D3 --> E E -- No: missing mandatory facts --> E1[Ask targeted clarification
one message, highest-impact gaps first] --> D E -- No: budget gap --> E2[Expectation management
shorter trip, off-season, lower hotel, alternate destination] --> D E -- Yes --> F{Intent and commitment sufficient?} F -- Strong intent / deposit --> G[3 Source Quote
send same neutral RFQ to 2 DMCs where possible] F -- Weak intent --> F1[Nurture with direction options
no detailed supplier work] --> D G --> H{Any DMC reply meets mandatory completeness
and itinerary score threshold?} H -- No --> H1[Targeted follow-up for missing fields
hotel names, price basis, terms, inclusions] --> G H -- Yes --> H2[Use first qualified DMC proposal
second DMC remains comparison / backup] H2 --> I[4 Price & Close
QC, margin, client proposal] I --> J{Money / commitment / complaint / change risk?} J -- Risky --> K[Human approval required] J -- Not risky --> L[AI executes/logs with evidence] K --> M{Client confirms and payment verified?} L --> M M -- No --> M1[Payment reminder / revise scope
no irreversible supplier lock] --> I M -- Yes --> N[5 Operate Trip
confirm suppliers, pre-departure, support] N --> O{During-trip issue?} O -- Yes --> O1[AI contacts DMC, records evidence, escalates high-risk cost/complaint] --> N O -- No --> P[Post-trip feedback
supplier score, experience DB, SOP learning]
3. Bespoke Stage Definitions and Required-Field Gates
The main gate should be required information first, score second. A case cannot move forward if mandatory fields are missing, even when the score looks high. Scores remain useful, but mainly as diagnostics, prioritisation and AI confidence.
| Stage | Objective | Mandatory fields to enter / operate | Condition to move forward | If not met | Example |
|---|---|---|---|---|---|
| 1 Qualify Lead | Identify whether this is a real travel enquiry, classify business line and create workspace. | Must have: contact/channel, original request text, enquiry source, at least one travel signal (destination/theme/date/pax/product). AI fields: suspected business line, urgency, owner, duplicate/thread match. | To Shape Direction: real contact + travel request + no hard reject + enough basics to ask useful next questions. Score helps prioritise, but does not replace mandatory fields. | Ask for the minimum missing mandatory facts. If spam, impossible request or no response after reasonable follow-up, archive/nurture. | “I want to go Japan” is enough to create workspace but not enough for RFQ; AI asks date/month, pax, duration, hotel level/budget and interests. |
| 2 Shape Direction | Turn a vague enquiry into a quotable brief. Search reference itinerary database first. | Must have before RFQ: destination/region/theme; date/month + duration; pax + adult/child/room assumptions; budget or hotel level/spending signal; travel style/interests; special constraints known or marked unknown; client intent signal. Nice-to-have: flights, meal style, guide language, room bedding, must-not-do list. | RFQ brief complete and client intent high enough. If no reference match: high intent → ask DMC; low intent → continue qualifying; poor-fit → decline/nurture. | Ask targeted missing questions; use budget bands if client refuses budget; do not consume DMC capacity for vague or poor-fit enquiries. | Switzerland with too-low budget → no formal RFQ; suggest shorter trip/off-season/lower hotel/alternative destination. |
| 3 Source Quote | Obtain real executable itinerary and quote from DMCs/suppliers. | RFQ package must include: client brief, dates/month/duration, pax/rooms, budget/hotel level, routing/theme, must-haves, constraints, response deadline, quote format required. DMC reply must include: day-by-day itinerary, specific hotels or clear hotel category + room assumptions, price/currency/pax basis, inclusions/exclusions, vehicle/guide details, terms, validity and cancellation. | At least one DMC reply passes mandatory completeness and itinerary quality threshold. Normally send to two DMCs; use the first qualified reply that reaches standard instead of waiting unnecessarily. | Send targeted DMC follow-up for missing fields; if first DMC is incomplete, chase while waiting for second; if none reaches standard by deadline, escalate or use backup DMC. | DMC gives total price but no hotel names/room assumptions → not quote-ready; AI asks specifically for hotel list, room type, tax/service basis and cancellation terms. |
| 4 Price & Close | Convert supplier proposal into client-facing quote with margin, evidence and terms. | Must have: selected DMC proposal, source-linked costs, margin calculation, FX/tax assumptions, client-facing inclusions/exclusions, quote validity, payment schedule, cancellation/change terms, approval card. | Client-facing quote requires approval. To Operate: client written confirmation + verified payment condition + supplier commitment approval. | Run QC, margin and risk gate. Any price, payment, discount, supplier lock-in or exception requires human approval. | DMC quote comes back much higher than initial estimate → approval card with reason, margin impact and alternatives before client quote. |
| 5 Operate Trip | Deliver the trip and feed learning back into the database. | Must have: client confirmation, payment status, supplier confirmations, final itinerary, traveller names/passport/flight details as needed, emergency/local contacts, internal owner and escalation path. | Closure-ready when trip completed and pending costs/complaints/refunds/invoices/feedback are resolved and supplier score updated. | Create departure checklist, monitor deadlines, handle urgent issues, archive feedback and update DMC/experience database. | Flight delay → contact DMC to adjust transfer, record cost/responsibility and escalate if money/complaint risk exists. |
4. Full Scoring Standards for Bespoke Gates
Yes — the proposed stage gates use scores. The score decides whether a case is ready to move to the next stage. It does not decide whether AI may execute externally; execution still depends on the confidence and risk gate.
4.1 Lead Qualification Score — move from Qualify Lead to Shape Direction, threshold ≥60/100
| Item | Points | Full-score standard | Partial / low-score example |
|---|---|---|---|
| Real contact and ownership | 15 | Name/contact/channel known; owner or sales responsible person identified. | Forwarded vague message = 5–8. |
| Business-line fit | 20 | Clearly bespoke/private, MICE or high-value travel service. | Unclear travel request = 8–12; non-travel/admin = 0. |
| Basic trip facts | 25 | At least 3 of destination/theme, date/month, pax, duration, hotel/preference known. | Only destination known = 8–10. |
| Commercial fit / budget signal | 20 | Budget, hotel level, client history or request suggests Dynasty-fit. | Budget likely too low = 0–8. |
| Intent / urgency | 10 | Client has real travel window, wants proposal/follow-up, or is referral/returning client. | Casual browsing = 2–4. |
| No hard reject | 10 | Not spam, not pure impossible request, not clearly outside service scope. | Pure lowest-price shopping with no flexibility = 0–3. |
4.2 RFQ-Ready Score — move from Shape Direction to Source Quote, threshold ≥75/100
| Item | Points | Full-score standard | Partial / low-score example |
|---|---|---|---|
| Destination / route scope | 20 | Main destination or route is narrowed enough for supplier to quote. | “Europe” only = 5; two possible countries = 12. |
| Date / duration | 15 | Exact dates or quotable month + duration. | “School holidays” = 5–8. |
| Pax / rooms | 15 | Pax, adults/children and room assumptions clear enough; variance ≤20%. | “Family” only = 5. |
| Budget and hotel level | 20 | Budget fit ≥0.8 or hotel/spending profile supports expected cost. | Budget mismatch = 0–8; no budget but luxury hotel cues = 12–15. |
| Travel style / must-haves | 15 | Interests, pace, guide/driver preference, food/activity needs known. | Generic “relaxing” = 5–7. |
| Special constraints | 10 | Mobility, diet, visa, children/elderly, flight constraints identified. | Unknown = 0–3. |
| Commitment level | 5 | Deposit paid or strong intent/deadline; otherwise soft RFQ only. | Casual enquiry = 1. |
4.3 Supplier Reply Completeness — move from Source Quote to Price & Close, threshold ≥80/100
| Item | Points | Full-score standard | Partial / low-score example |
|---|---|---|---|
| Itinerary substance | 20 | Day-by-day routing and experiences are usable, logical and client-fit. | Generic copy/paste itinerary = 8–12. |
| Hotel / room assumptions | 15 | Hotel names or level, room type assumptions and alternatives clear. | “4-star hotel” without details = 6–8. |
| Transport / guide details | 15 | Vehicle type, guide language, driver/guide hours and route logistics clear. | Transport mentioned but no vehicle/guide detail = 6–8. |
| Price and currency | 20 | Cost, currency, pax basis, validity and tax/service assumptions clear. | Total price only, no basis = 8–10. |
| Inclusions / exclusions | 10 | Meals, entrances, tips, flights, porterage, taxes clearly separated. | Ambiguous inclusions = 3–5. |
| Terms and cancellation | 10 | Payment deadline, cancellation/change terms and supplier validity clear. | No cancellation terms = 0–3. |
| Quality fit | 10 | Proposal meets Dynasty quality; no obvious routing/brand/feasibility issue. | Poor luxury fit or impractical routing = 0–5. |
4.4 Quote-Ready Score — prepare client-facing price, threshold ≥85/100
| Item | Points | Full-score standard | Partial / low-score example |
|---|---|---|---|
| Source evidence | 20 | Supplier quote/API fare/source links saved in evidence ledger. | Price copied without source = 0–8. |
| Cost and margin calculation | 25 | Internal cost, markup/margin, FX/tax assumptions and exceptions clear. | Cost known but margin not checked = 8–12. |
| Client assumptions | 15 | Pax, rooms, dates, inclusions/exclusions and validity shown. | Assumptions hidden or vague = 4–8. |
| QC / feasibility | 15 | Routing, seasonality, hotel, guide/vehicle, activity feasibility checked. | Supplier itinerary not QC’d = 0–6. |
| Payment and terms | 15 | Deposit/balance, due dates, cancellation/change terms ready. | Payment schedule missing = 3–6. |
| Approval readiness | 10 | Human approval card has price, margin, evidence, risk and recommendation. | Approval lacks evidence/rationale = 3–5. |
4.5 Execution-Ready Score — move from Price & Close to Operate Trip, threshold ≥90/100
| Item | Points | Full-score standard | Partial / low-score example |
|---|---|---|---|
| Client confirmation | 15 | Written confirmation of itinerary, price, terms and passenger assumptions. | Verbal/unclear confirmation = 5–8. |
| Payment evidence | 25 | Deposit/full payment received and matched to case. | Payment promised but not verified = 0–5. |
| Supplier confirmations | 20 | DMC/hotel/airline/cruise confirmations or booking references saved. | Supplier says “noted” but no confirmation = 5–8. |
| Traveller details | 15 | Names/passports/flight details/diet/mobility/emergency contacts captured as needed. | Missing passport or flight detail = 5–10. |
| Final itinerary / documents | 15 | Final itinerary, vouchers, contacts and briefing ready. | Draft itinerary only = 5–8. |
| Operational ownership | 10 | Internal owner, DMC local contact, escalation channel and deadlines clear. | No escalation contact = 3–5. |
Recommendation: use these scores as gates and diagnostics, not as blind automation. The score should show what is missing; the AI action should be “ask for the missing highest-impact field” or “escalate for approval”, not simply “fail”.
5. Shape Direction 100-Point Score
Shape Direction should be measurable, not based on “feels enough”. ERP should show the missing points and AI should ask for the highest-impact missing information first.
| Item | Points | Full-score standard | Partial-score example |
|---|---|---|---|
| Destination / theme clarity | 20 | Clear country/city/route theme, e.g. Hokkaido ski + onsen. | “Europe” = 5; “Switzerland or Italy” = 12. |
| Date / duration clarity | 15 | Exact date or ≤14-day window plus duration. | “Year end” = 5; “mid-Dec, around 10 days” = 12. |
| Pax / rooms / traveller structure | 15 | Pax, adults/children, rooms, beds, age/mobility needs known. | “Family trip” = 5; “2 adults + 2 children, 2 rooms” = 13. |
| Budget / spending fit | 20 | Budget, hotel level or history proves Dynasty bespoke fit. | No budget but Aman/Four Seasons = 15; clearly too low = 0–8. |
| Travel style / interests | 15 | Pace, hotel preference, food, culture/nature/shopping/family interests clear. | “Relaxed” = 6; clear preferences = 12–15. |
| Constraints / risk factors | 10 | Visa, mobility, diet, flights, elderly/children, seasonality risks identified. | Unknown = 0; partly known = 5. |
| Intent / decision chain | 5 | Decision maker, proposal deadline and willingness to proceed/deposit known. | Casual enquiry = 1; clear intention = 5. |
Decision rule: ≥75 can move to Source Quote; 60–74 continues shaping; <60 returns to qualification. Even if total score is ≥75, if budget fit is below 0.8, no formal RFQ unless human approves a soft exploratory RFQ.
6. Reference-Itinerary-First + No-Freeform-Itinerary Rule
In Shape Direction, AI should not invent itineraries from general domain knowledge. It either adapts a proven Dynasty reference itinerary, qualifies the client further, or asks DMC when intent is high.
| Step | Rule | AI action | Human role |
|---|---|---|---|
| 1 | Search reference itinerary database first. | Match by destination, season, duration, pax profile, travel style, budget level, hotel tier and interests. | Review only if match quality is uncertain or client is high value. |
| 2 | If match quality is strong, adapt the reference itinerary. | Use the closest itinerary as base; adjust pacing, hotels, activities and logistics to the client. Mark which parts are confirmed vs assumptions. | Approve only if client-facing proposal, pricing or unusual promise is involved. |
| 3 | If no good match exists and intent is high, ask DMC. | Send a neutral RFQ / proposal request to qualified DMCs with no client price promise and no booking/blocking commitment. | No client quote until supplier evidence, QC, margin and approval are complete. |
| 4 | If no good match exists and intent is low, keep qualifying. | Ask the minimum missing questions, offer budget bands/examples, and gather intel until intent becomes high enough. | Human not needed unless VIP/sensitive or the case becomes risky. |
| 5 | If the client remains poor-fit, decline nicely. | Examples: budget clearly below bespoke threshold, refuses essential info, wants lowest-price shopping only, or request is not executable. | Use a polite close/nurture template; do not consume DMC capacity. |
7. High-Intent DMC Fast Track
If no good reference itinerary exists, AI should check whether the client shows high intent. If intent is high, AI can skip LLM itinerary design and go straight to a DMC proposal/RFQ path.
| Signal | Points | How to interpret |
|---|---|---|
| Repeat customer / referral / past converted client | 25 | Strong signal that the enquiry is serious and worth supplier effort. |
| Detailed brief | 25 | Destination, dates, pax, duration, budget/hotel level and preferences are mostly provided. |
| Specific greeting | 15 | Email is addressed to Dynasty or a named staff member, not generic “Hi” / “Dear Sir”. |
| Clear travel window or proposal deadline | 15 | Client has a real timeline and wants to move forward. |
| Luxury / spending ability signal | 10 | 5-star hotels, private driver, suites, business class, special access, luxury ryokan, etc. |
| Active engagement | 10 | Client replies quickly, has multiple exchanges, asks practical next-step questions. |
Decision rule: High Intent Score ≥70/100 → AI may prepare a DMC proposal/RFQ request. If the RFQ is neutral, makes no price promise to the client, and does not ask DMC to block resources, AI may handle it directly and log evidence. Anything involving money, pricing, discount, payment, refund, supplier lock-in, complaint, cancellation or material last-minute change is risky and requires human approval.
| Example | High-intent decision | AI action |
|---|---|---|
| Repeat client asks for “12 Dec Hokkaido, 4 pax, 5-star ryokan, private driver”. | High intent. | Search reference itinerary first; if no good match, send RFQ to Japan DMCs with no booking commitment. |
| New client writes “Dear Dynasty team, 6 pax Spain/Portugal Dec 2026, prefer 5-star and private guide”. | Likely high intent. | Use reference itinerary if available; otherwise DMC fast track is acceptable. |
| Email only says “Hi, please send Europe package”. | Low intent. | Do not ask DMC yet; ask clarifying questions first. |
8. Two-DMC RFQ and First-Qualified Reply Rule
Default sourcing should ask two qualified DMCs for the same neutral RFQ. But the system should not wait for both if one DMC replies first and the reply is already complete and strong enough.
| Rule | AI action | Important control |
|---|---|---|
| Send to two DMCs where possible. | Send the same structured RFQ brief to two approved DMCs, with the required response format and deadline. | If only one strong DMC exists or time is too short, record why one-DMC sourcing was used. |
| Track reply completeness as a checklist, not just free text. | Parse each DMC reply into mandatory fields: itinerary, specific hotels/rooms, price/currency/basis, inclusions/exclusions, guide/vehicle, terms/cancellation, validity. | Missing hard-required fields block client quote even if the itinerary sounds attractive. |
| First qualified DMC wins first proposal path. | If DMC A replies first and passes mandatory completeness + itinerary quality threshold, use A as the working proposal for QC, pricing and client quote approval. | Do not delay unnecessarily waiting for DMC B unless A is weak, expensive, incomplete, or the case is strategically important. |
| Second DMC remains comparison / backup. | If DMC B later replies before client quote is sent, compare value and switch only if materially better. Otherwise store B for benchmark and backup. | Keep evidence of why A was selected: response speed, completeness, quality, price, fit. |
| Targeted DMC follow-up workflow. | If a reply is incomplete, AI sends a precise follow-up listing missing fields instead of a generic “please advise”. | Examples: “Please provide hotel names and room categories”, “please separate inclusions/exclusions”, “please confirm validity and cancellation terms”. |
| Apple-to-apple only when useful. | If first proposal is strong but price needs benchmarking, sanitize the structure and ask the other DMC to quote same assumptions. | Remove DMC name, logo, watermark, metadata and identifying formatting before sharing. |
9. Bespoke Scenario Examples
| Case | Stage | AI judgement | AI action / recommendation | Risk |
|---|---|---|---|---|
| New client says only “Japan / Europe trip” | 1 Qualify Lead | Valid lead, insufficient details. | Create workspace; ask key questions in one message; then search reference itinerary database before proposing direction. | Non-risky |
| Budget below bespoke threshold | 2 Shape Direction | Expectation gap. | Explain budget logic; suggest shorter trip/off-season/lower hotel/alternative destination. | Non-risky |
| Client refuses budget disclosure | 2 Shape Direction | Cannot design accurately. | Offer 2–3 budget tiers and ask client to choose direction. | Non-risky |
| Destination changes after RFQ | 2 Shape Direction / 3 Source Quote | Core brief changed. | Pause old RFQ, re-score concept, notify supplier if needed. | Non-risky/High |
| Flight trade-off | 2 Shape Direction / 4 Price & Close | Comfort/budget/time trade-off; prices must be sourced. | Use Sabre/NDC/API; show source, route impact; approval before client quote. | High |
| Detailed proposal before deposit | 3 Source Quote | High intent but no commitment. | Soft RFQ only; no blocking/booking commitment. | Non-risky |
| DMC incomplete / poor reply | 3 Source Quote | Not quote-ready. | Targeted follow-up, QC comments, compare with second DMC or backup DMC if needed. | Non-risky |
| Client confirms but payment not verified | 4 Price & Close | Cannot make irreversible commitment. | Payment reminder; no ticket/hotel/DMC lock until evidence. | High |
| Complaint/refund/compensation | 5 Operate Trip | Brand and financial risk. | Collect facts, propose remedy, human approval for refund/compensation. | High |
10. Three End-to-End Bespoke Examples
These examples show how the system should run from first enquiry to post-trip learning, based on common Dynasty-style historical cases and patterns.
| Example | Case profile | How the system runs | AI behaviour at each stage |
|---|---|---|---|
| A. Family Japan winter private tour | Family asks for Japan in December, wants snow, food, onsen and private driver; budget unclear at first. | 1 Qualify → 2 Shape → 3 Source → 4 Price → 5 Operate. | 1: creates workspace, asks pax/children ages/dates/rooms/budget/hotel level. 2: searches reference itineraries first; if a Hokkaido ski+onsen or Tokyo+Hakone+Karuizawa itinerary matches, adapts it and explains winter cost/hotel constraints. 3: sends RFQ to at least two Japan DMCs where possible; asks for driver, guide, hotel category, snow activities, meals, cancellation terms. 4: QC itinerary, checks margin, prepares quality and cost-controlled options; quote requires approval. 5: collects passports/flights/diet, confirms DMC contacts, monitors arrival, then updates guide/hotel/experience quality after trip. |
| B. Spain & Portugal family private tour | Family wants Spain/Portugal, dates far ahead, several pax/room details may change, client compares routing and flight comfort. | 1 Qualify → 2 Shape with routing trade-offs → 3 DMC RFQ → 4 revised quote loop → 5 operate. | 1: matches or creates client/trip workspace and captures pax, target month, route interest. 2: compares Madrid/Seville/Lisbon/Porto routing, pace and seasonality; flags if date/pax/rooms are still unstable. 3: RFQ goes to at least two Iberia DMCs; if one proposal has stronger routing, AI can sanitize it and ask the other DMC to quote the same hotels/restaurants/car type for apple-to-apple comparison. 4: AI compares proposal design and value-for-money; any date, room or pax change triggers change-request logic; human approves client-facing price. 5: locks services only after payment evidence, generates pre-departure checklist and stores final routing for future Iberia cases. |
| C. Low-budget Europe enquiry that may not fit Dynasty | Client wants Switzerland/Europe private trip but budget is below bespoke threshold and refuses to commit initially. | 1 Qualify → 2 Shape / expectation management → either 3 soft RFQ with approval or nurture/close. | 1: lead is valid but flagged budget-risk. 2: Shape Direction score may be high on destination/date but low on budget fit; AI does not RFQ automatically. It explains realistic budget and offers alternatives: fewer cities, off-season, 4-star hotels, Austria/Northern Italy, shorter trip. 3: only if human approves or client shows strong intent, AI sends soft RFQ with no booking/blocking commitment. 4: if supplier quote confirms mismatch, AI recommends either adjusted scope or polite close/nurture, not low-margin discounting. 5: if converted, normal operate flow; if closed, store preference for future EDM/follow-up. |
11. Human + AI Collaboration Mechanism for Bespoke
For the early AI phase, keep the operating model deliberately simple: AI handles non-risky work; humans approve risky work. Remove medium risk as a separate category.
| Category | Execution mechanism | Human involvement | Examples |
|---|---|---|---|
| Non-risky | AI executes directly and logs the action, evidence and confidence. | Human reviews through activity log / digest only when needed. | Create workspace, classify enquiry, extract facts, ask clarification questions, search reference itinerary DB, send neutral DMC RFQ with no commitment, DMC follow-up for missing fields, internal summaries. |
| Risky | Mandatory approval. AI prepares an approval card; no execution until human approves/edits/rejects. | Human must decide inside ERP approval center. | Anything involving money transfer, client quote, discount, payment confirmation, refund, compensation, invoice, supplier lock-in/prepayment, DMC quote much higher than estimate, complaint, sudden cancellation, last-minute DMC itinerary change, legal/brand-sensitive promise. |
Important: Simpler is better for early AI adoption. The system should not ask humans to classify low/medium/high every time; it should ask one question: “Could this create financial, supplier, client, legal or brand commitment?” If yes → approval. If no → AI proceeds with audit trail.
12. AI Action Queue for Non-Risky Work
Without a medium-risk category, the queue becomes simpler: it is not an approval queue. It is a transparent activity lane for actions AI is allowed to execute, with pause/edit controls for managers.
| Action type | Default behaviour | Human options | Escalate to risky when… |
|---|---|---|---|
| Ordinary client clarification | AI sends when confidence is high; otherwise drafts for review. | Edit template, pause case, assign owner. | VIP/sensitive wording, complaint tone, or client asks about price/discount/refund. |
| Neutral DMC RFQ / follow-up | AI can send to approved DMCs with no booking/blocking commitment. | Add/remove DMC, adjust deadline, pause supplier. | New untested DMC, high-value case, request to hold space, price promise, or urgent last-minute change. |
| Missing-info reminders | AI follows up politely and records response. | Change frequency, stop reminders, assign human. | Departure within 72h, missing passport/visa/flight, elderly/children risk, or client frustration. |
| Supplier data/database updates | AI extracts and updates structured fields with source link. | Review changes in digest. | Commercial commitment, blacklist/termination, public-facing content use. |
Recommended notification rhythm: default to case activity log + 2–3 daily digests. Instant notifications only for newly created approval cards, failed actions, urgent departure risk or complaints.
13. A Full Day: Human + AI Interaction Example
Example of how a planner or manager can supervise many Bespoke cases without checking every action manually.
| Time | What happens | AI action | Human experience |
|---|---|---|---|
| 09:15 | New Japan family enquiry arrives by email. | Creates Trip Workspace, classifies as Bespoke, extracts destination/date/pax signals, lead score 62/100, drafts clarification email. | No interruption. The action appears in AI Action Queue: “send clarification if confidence remains high”. |
| 09:40 | Planner opens dashboard briefly. | Shows 4 pending non-risky actions and 1 high-risk quote approval. | Planner edits one Japan clarification question, reviews one DMC RFQ, leaves the rest for AI to send because they are non-risky. |
| 10:30 | Morning digest. | Sends summary: 3 actions auto-executed, 2 upcoming, 1 blocked by missing budget. | Planner reads digest, no need to open every case. |
| 11:20 | DMC replies to Spain/Portugal but missing room assumptions and cancellation terms. | Supplier completeness score 68/100. Sends targeted follow-up or drafts only if confidence is low. | No instant alert because normal DMC follow-up. It appears in queue. |
| 12:05 | Client asks if quoted price can be reduced. | Classifies as high-risk because it affects price/margin. Prepares approval card with options: hold value, reduce hotel level, shorten route, manager discount. | Human receives approval notification. No auto-send. |
| 14:10 | Japan client confirms budget and dates. | Shape Direction score rises to 82/100. AI recommends moving to Source Quote and drafts RFQ to 2 Japan DMCs. | Non-risky RFQ can be sent by AI; planner can edit the supplier shortlist if needed. |
| 15:30 | Afternoon digest. | Summarizes auto-sent RFQs, DMC deadlines, blocked high-risk quote, near-departure missing passport. | Planner focuses only on exceptions. |
| 16:45 | Client travelling tomorrow has missing flight details. | Near-departure exception; AI sends instant alert and drafts reminder. | Human sees urgent notification, approves send immediately. |
| 18:00 | End-of-day digest. | Lists completed actions, pending approvals, tomorrow’s DMC deadlines, and cases needing human decision. | Manager reviews only the exception list, not every email/task. |
Product takeaway: humans should spend time on exceptions and judgement, not monitoring. The ERP should let AI run routine non-risky actions with a visible countdown and strong audit trail.
1. Other Business Lines — Timeline and Conditions
| Line | Timeline | Key gates | Human approval points |
|---|---|---|---|
| MICE / Incentive | Corporate Qualification → Scope Event/Trip → Supplier/Tender RFQ → Proposal/Contract/Payment → Operate Group/Event → Debrief | Decision chain known; pax/date/objective/budget clear; tender deadline; rooming/list needs; onsite responsibility defined. | Proposal price, contract terms, deposit/payment schedule, supplier commitment, tender exception, onsite risk plan. |
| Group Tour | Product Match → Seat/Price Check → Booking → Payment/Documents → Tour Operation → Feedback | Product exists, departure open, seat available, price valid, passenger data collected, payment verified. | Discount, waitlist, cancellation/refund, non-standard rooming, product exception. |
| Ticketing | Capture Ticket Request → Search Fare & Availability → Reserve / Hold PNR → Invoice & Payment → Issue Ticket → Servicing / Changes | Fare source is live; client-approved PNR hold/TTL is recorded before invoicing and issue. | PNR hold, ticket issue, refund/reissue, schedule-change options, financial exception. |
| Ancillary | Attach to Trip → Confirm Requirement → Source Provider → Price/Payment → Confirm → Support | Active trip or strategic client; provider available; terms clear; payment/approval before confirmation. | VIP service confirmation, special-access promise, refund/compensation, supplier prepayment. |
| DMC / Supplier | Identify Supplier → Registration/Profile → Sample Itinerary/Pricing → QC/Score → Database Update → Use in RFQ | Contact, destination, capabilities, pricing, terms, sample itinerary and quality score captured. | Commercial commitment, sensitive negotiation, termination/blacklist, public-facing content use. |
2. Other Business Scenario Examples
| Case | Line / stage | AI action | Risk |
|---|---|---|---|
| Corporate incentive enquiry | MICE / Qualification | Ask pax, destination/date, objective, budget, decision owner, tender deadline, flight/land split. | Non-risky |
| Client asks for fixed group tour | Group Tour | Check active product; if none, propose private/bespoke or nurture for future group products. | Non-risky |
| Air ticket only | Ticketing | Quote from ticketable source; collect payment before issue; ticket issue needs approval. | High |
| Standalone visa request | Ancillary | If tied to trip, add task; otherwise decline/nurture/refer politely. | Low/Non-risky |
| DMC sends registration form or itinerary | DMC / Supplier | Parse registration/pricing/capabilities; QC itinerary; update DMC/experience database. | Low/Non-risky |
1. AI Confidence and Risk Control
flowchart TD
A[AI proposes judgement/action] --> B{Confidence >= 0.80?}
B -- No --> H[Human review or collect more evidence]
B -- Yes --> C{Could this create money / commitment / complaint / change risk?}
C -- Yes --> H[Mandatory approval card]
C -- No --> L[AI executes immediately
log audit trail] L --> Z[Update workspace, task, evidence, next step] H --> Z
log audit trail] L --> Z[Update workspace, task, evidence, next step] H --> Z
| Category | Execution | Examples |
|---|---|---|
| Non-risky | AI directly executes and logs evidence. | Workspace, classification, field updates, internal summary, neutral DMC RFQ/follow-up with no commitment. |
| Risky | Human approval required. | Client quote, invoice, payment/refund, discount, supplier lock-in, complaint, cancellation, low margin / high DMC quote exception, material last-minute change. |
2. ERP Structure Recommendation
Lead / Trip Workspace
Central object. All emails, WhatsApp, calls, files, tasks, quotes, proposals, invoices, payments and approvals attach here.
Sticky Business Tabs
ERP should mirror this document: Overview, Bespoke, Other Business, ERP/AI controls.
Business-Line Classifier
Routes intake to Bespoke, MICE, Group, Ticketing, Cruise, FIT, Ancillary, DMC/Supplier or Other.
Stage Engine
Each business line has fixed timeline and quantitative gates. Complex events become subtasks/revisions.
AI Action Queue
Every proposed action stores confidence, whether it is risky, evidence, deadline, owner, action text and result/approval status.
Approval Center
Risky actions show financial impact, supplier/client commitment, source links and alternatives.
Evidence Ledger
No price, payment, supplier lock-in or refund without source evidence.
Exception Dashboard
Shows stuck cases, overdue tasks, pending approvals, payment risk, low margin, DMC delay, VIP/complaint and departure risks.
Recommendation: Do not build a prettier task list. Build the case-centric operating system: classification → stage gate → AI action → risk/approval → evidence → learning.
1. Product Document Recommendations
Turn the flowchart into a PRD draft
This document should not be treated as a static process chart. It should become the first product requirements document for Dynasty ERP / AI Workflow.
Add field requirements by stage
Each stage should define required fields, AI-generated fields, human-confirmed fields and output artifacts.
Use required fields before scores
In MVP, hard required fields should block stage movement. Scores should diagnose what is missing, rank priority, and recommend the next action. Human override and reason must be logged.
Keep examples close to real cases
Every workflow rule should be tested against real case types: family Japan, Spain/Portugal, low-budget Europe, MICE, ticketing-only, DMC proposal.
2. ERP System Recommendations
| Recommendation | Why it matters | What to build |
|---|---|---|
| Case-centric, not task-list-centric | Tasks are only useful when attached to a client/trip/supplier context. | Lead / Trip Workspace as the core object. |
| Exception dashboard | Management should not read every chat/task. They need stuck cases, overdue approvals, payment risk, DMC delays and complaints. | Dashboard with exception cards and SLA filters. |
| AI Action Queue | AI should propose structured actions, not just chat responses. | Action object: case, stage, confidence, risk, evidence, timer, owner, result. |
| Independent Approval Center | High-risk financial/client/supplier actions need consistent approval UX. | Approval cards with evidence, financial impact, recommendation and alternatives. |
| Evidence Ledger | Prices, payments and supplier commitments must be source-linked. | Evidence records attached to quotes, payments, approvals and supplier confirmations. |
3. Early AI Adoption Recommendations from Historical Cases
| Recommendation | Why | MVP implementation |
|---|---|---|
| Approved DMC whitelist by destination | AI should not decide supplier quality from scratch. | Per destination: primary DMC, backup DMC, specialty, response SLA, quality score, allowed RFQ template. |
| Fit/intent before supplier work | Low-budget or vague enquiries have historically consumed too much planner/DMC time. | Show intent score and missing-info checklist; only high-intent/no-match cases trigger DMC RFQ. |
| One approval definition | Early users adopt faster when rules are obvious. | “Money/commitment/complaint/change/cancellation = approval”; everything else AI logs and proceeds. |
| Evidence-first pricing | Prevents fabricated/rounded numbers and protects margin. | No quote card without supplier/API/source link, margin math and approval trail. |
| Learning loop after every DMC reply | The database improves only if AI captures supplier behaviour. | Auto-update response time, completeness score, price variance, routing quality and reusable experiences. |
4. Suggested MVP Priority
| Priority | Module | Scope |
|---|---|---|
| 1 | Trip Workspace | Client, enquiry, conversations, files, stage, tasks, approvals, evidence. |
| 2 | Bespoke 5-stage timeline | Qualify Lead → Shape Direction → Source Quote → Price & Close → Operate Trip. |
| 3 | AI scoring + missing-fields diagnosis | Required-field gates plus diagnostic scores: lead priority, RFQ readiness, DMC reply completeness, quote readiness, execution readiness. |
| 4 | Task panel + AI Action Queue | Next actions, risky/non-risky flag, owner, due time, evidence, result. |
| 5 | Approval Center | Client quote, invoice, payment, supplier lock, refund, low-margin exception. |
| 6 | Evidence Ledger | Source links for price, payment, supplier quote, confirmation and approval. |
Dashboard Demo UI — Timeline + Task Panel
This is a lightweight product mockup showing how the ERP could feel: cases on the left, stage timeline and evidence in the center, AI task/action panel on the right.
Trip Workspace: Spain & Portugal Family Private Tour
1. Overview — 完整业务流程
Dynasty 的流程应从统一入口开始，再进行业务分类。不同业务线有各自 timeline，但共享财务、审批、服务和知识沉淀控制。
flowchart TD
A[统一入口
Email / WhatsApp / Website / Call / Walk-in / Partner / Staff] --> B[创建或匹配 Client + Lead/Trip Workspace] B --> C[抓取最低信息
联系人、来源、原始需求、目的地/主题、日期、人数、预算信号] C --> D{业务分类} D --> BT[Bespoke / Private Tour] D --> M[MICE / Incentive] D --> G[Group Tour 询问] D --> T[Ticketing] D --> A2[附加服务] D --> S[DMC / Supplier 合作] D --> O[其他 / 行政 / 无关] BT --> BT1[接单判断] --> BT2[需求定向] --> BT3[供应商询价] --> BT4[报价成交] --> BT5[出行服务] --> L[回访学习 + 数据库更新] M --> M1[企业资格判断] --> M2[活动/旅行范围确认] --> M3[供应商/Tender 询价] --> M4[Proposal + 合同 + 付款] --> M5[团队/活动执行] --> L G --> G1{是否有固定团产品?} G1 -- 无 / 当前默认 --> G2[如适合，转 Bespoke / Private] G1 -- 有 / 未来产品 --> G3[查出发、座位、价格、tour code] --> G4[报名 + 付款 + 文件] --> G5[团队执行] --> L T --> T1[识别产品] --> T2[实时库存 + 价格来源] --> T3[先收款] --> T4[出票 / 确认] --> T5[变更服务] --> L A2 --> A21{是否附属于 active trip?} A21 -- 是 --> A22[加入 Trip Workspace 的附加 task] A21 -- 否 --> A23[通常婉拒 / nurture / 转介] S --> S1[注册供应商/DMC] --> S2[收 profile + pricing + sample itinerary] --> S3[QC 和评分] --> S4[更新供应商 / 体验数据库] O --> O1[关闭 / 内部转派 / 归档] BT4 --> F[财务控制
毛利、发票、付款、退款] M4 --> F G4 --> F T3 --> F F --> P[审批中心] P --> BT5 P --> M5 P --> G5 P --> T4
Email / WhatsApp / Website / Call / Walk-in / Partner / Staff] --> B[创建或匹配 Client + Lead/Trip Workspace] B --> C[抓取最低信息
联系人、来源、原始需求、目的地/主题、日期、人数、预算信号] C --> D{业务分类} D --> BT[Bespoke / Private Tour] D --> M[MICE / Incentive] D --> G[Group Tour 询问] D --> T[Ticketing] D --> A2[附加服务] D --> S[DMC / Supplier 合作] D --> O[其他 / 行政 / 无关] BT --> BT1[接单判断] --> BT2[需求定向] --> BT3[供应商询价] --> BT4[报价成交] --> BT5[出行服务] --> L[回访学习 + 数据库更新] M --> M1[企业资格判断] --> M2[活动/旅行范围确认] --> M3[供应商/Tender 询价] --> M4[Proposal + 合同 + 付款] --> M5[团队/活动执行] --> L G --> G1{是否有固定团产品?} G1 -- 无 / 当前默认 --> G2[如适合，转 Bespoke / Private] G1 -- 有 / 未来产品 --> G3[查出发、座位、价格、tour code] --> G4[报名 + 付款 + 文件] --> G5[团队执行] --> L T --> T1[识别产品] --> T2[实时库存 + 价格来源] --> T3[先收款] --> T4[出票 / 确认] --> T5[变更服务] --> L A2 --> A21{是否附属于 active trip?} A21 -- 是 --> A22[加入 Trip Workspace 的附加 task] A21 -- 否 --> A23[通常婉拒 / nurture / 转介] S --> S1[注册供应商/DMC] --> S2[收 profile + pricing + sample itinerary] --> S3[QC 和评分] --> S4[更新供应商 / 体验数据库] O --> O1[关闭 / 内部转派 / 归档] BT4 --> F[财务控制
毛利、发票、付款、退款] M4 --> F G4 --> F T3 --> F F --> P[审批中心] P --> BT5 P --> M5 P --> G5 P --> T4
2. 业务线总览
| 业务线 | 状态 | 包含什么 | 进入条件 | 退出/转向条件 |
|---|---|---|---|---|
| Bespoke / Private Tour | 重点核心 | 个人、家庭、affluent 小团体定制私家旅行。 | 客户要定制；有目的地/主题/日期/人数/预算信号；大概率符合 Dynasty 定位。 | 纯低价比较、不愿提供基础信息、需求不可执行、或预算不匹配。 |
| MICE / Incentive | 核心 | 企业奖励旅游、会议、活动、retreat、团队出行。 | 企业主体、pax、目标、决策链、日期和预算信号。 | 决策人不清、预算/时间不现实、纯低价团购。 |
| Group Tour | 传统模式暂停/重构 | 固定出发团或未来主题团。 | 客户询问固定团或已有产品。 | 无 active product 时，适合则转 bespoke/private，否则 nurture。 |
| Ticketing | 选择性处理 | 机票出票与退改签服务。 | 有 live source；客户接受服务费/付款条款；确认前先收款。 | 低毛利不支持、拒绝服务费、或有信用风险。 |
| Ancillary | 附属 | 保险、签证建议、VIP 机场服务、fast track、餐厅、特殊 access。 | 通常附属于 active trip 或战略客户。 | 单独 visa-only / 低价值行政，除非战略例外。 |
| DMC / Supplier | 基础设施 | 供应商 onboarding、注册、价格、样板行程、体验数据库。 | 潜在伙伴有有用目的地能力/价格/内容。 | 资料不完整、质量差、重复、或商业不匹配。 |
1. Bespoke / Private Tour — 核心 Timeline
这是重点业务流程，也应该是 ERP 中最强的视觉锚点。主阶段保持简单，复杂情况放在 subtask、revision 和 approval 内处理。
1接单判断
有没有真实客户？是否适合 Dynasty？资料是否够开始推进？
2需求定向
优先从参考行程库匹配最接近客户需求的行程；没有好匹配时，高意向才问 DMC，低意向继续补问或礼貌关闭。AI 不自行凭知识编行程。
3供应商询价
客户必填 brief 完整且意向清楚后，尽量向两家 DMC / 供应商拿可执行方案。
4报价成交
做 QC、毛利检查、客户报价、修改、确认、收款。
5出行服务
确认资源、出发前准备、旅途中支持、回访和知识沉淀。
2. Bespoke / Private Tour 流程图
这张图展示 Bespoke case 如何通过 5 个主阶段，并处理资料不足、预算不匹配、需求修改、供应商问题、付款和审批 gate。
flowchart TD
A[新的 bespoke 询问] --> B[1 接单判断
建 workspace、分类、评分] B --> C{接单评分 >= 60
且核心信息 >= 3项?} C -- 否 --> C1[一次性补问缺失信息
或 nurture / 归档] --> B C -- 是 --> D[2 需求定向
优先搜索参考行程库] D --> D0{是否有高匹配参考行程?} D0 -- 是 --> D1[使用最匹配参考行程 按客户需求调整] D0 -- 否 --> D2{客户意向是否高?} D2 -- 是 --> D3[直接进入 DMC proposal/RFQ 路径] D2 -- 否 --> D4[AI 提出行程方向 必须人工 review] D1 --> E{方案清晰度 >= 75
且预算匹配 >= 0.8?} D3 --> E D4 --> E E -- 否: 资料不足 --> E1[补日期、人数、房型、风格、限制条件] --> D E -- 否: 预算差距 --> E2[期望管理
缩短天数、淡季、降酒店、换目的地] --> D E -- 是 --> F{客户 commitment?} F -- 已付定金 / 高意向 --> G[3 供应商询价
正式 RFQ 或 soft RFQ] F -- 意向弱 --> F1[Nurture / 给方向建议
不做深度供应商工作] --> D G --> H{供应商回复完整度 >= 80?} H -- 否 --> H1[追 DMC / QC 反馈 / 找 backup] --> G H -- 是 --> I[4 报价成交
QC、毛利、客户 proposal] I --> J{报价/付款/承诺风险?} J -- 高风险 --> K[必须人工批准] J -- 无风险 --> K1[延时执行队列] J -- 低风险 --> L[AI 执行并记录] K --> M{客户确认且付款核实?} K1 --> M L --> M M -- 否 --> M1[付款提醒 / 修改范围
不得锁不可逆资源] --> I M -- 是 --> N[5 出行服务
确认供应商、出发前、旅途中支持] N --> O{旅途中问题?} O -- 是 --> O1[AI 联系 DMC、记录证据、高风险费用/投诉升级] --> N O -- 否 --> P[回访学习
供应商评分、体验库、SOP 更新]
建 workspace、分类、评分] B --> C{接单评分 >= 60
且核心信息 >= 3项?} C -- 否 --> C1[一次性补问缺失信息
或 nurture / 归档] --> B C -- 是 --> D[2 需求定向
优先搜索参考行程库] D --> D0{是否有高匹配参考行程?} D0 -- 是 --> D1[使用最匹配参考行程 按客户需求调整] D0 -- 否 --> D2{客户意向是否高?} D2 -- 是 --> D3[直接进入 DMC proposal/RFQ 路径] D2 -- 否 --> D4[AI 提出行程方向 必须人工 review] D1 --> E{方案清晰度 >= 75
且预算匹配 >= 0.8?} D3 --> E D4 --> E E -- 否: 资料不足 --> E1[补日期、人数、房型、风格、限制条件] --> D E -- 否: 预算差距 --> E2[期望管理
缩短天数、淡季、降酒店、换目的地] --> D E -- 是 --> F{客户 commitment?} F -- 已付定金 / 高意向 --> G[3 供应商询价
正式 RFQ 或 soft RFQ] F -- 意向弱 --> F1[Nurture / 给方向建议
不做深度供应商工作] --> D G --> H{供应商回复完整度 >= 80?} H -- 否 --> H1[追 DMC / QC 反馈 / 找 backup] --> G H -- 是 --> I[4 报价成交
QC、毛利、客户 proposal] I --> J{报价/付款/承诺风险?} J -- 高风险 --> K[必须人工批准] J -- 无风险 --> K1[延时执行队列] J -- 低风险 --> L[AI 执行并记录] K --> M{客户确认且付款核实?} K1 --> M L --> M M -- 否 --> M1[付款提醒 / 修改范围
不得锁不可逆资源] --> I M -- 是 --> N[5 出行服务
确认供应商、出发前、旅途中支持] N --> O{旅途中问题?} O -- 是 --> O1[AI 联系 DMC、记录证据、高风险费用/投诉升级] --> N O -- 否 --> P[回访学习
供应商评分、体验库、SOP 更新]
3. Bespoke 阶段定义与必填信息 Gate
主 gate 应该是必填信息优先，分数第二。缺少关键必填字段时，即使总分看起来高，也不能进入下一阶段。分数仍有价值，但主要用于诊断、优先级和 AI 置信度。
| 阶段 | 目标 | 进入/运行必填字段 | 进入下一阶段条件 | 未满足时 | 示例 |
|---|---|---|---|---|---|
| 1 接单判断 | 判断是否真实旅行询问，分类业务线并建立 workspace。 | 必填：联系人/渠道、原始需求、来源、至少一个旅行信号（目的地/主题/日期/人数/产品）。AI 自动字段：疑似业务线、紧急度、owner、是否重复/thread match。 | 进入需求定向：真实联系人 + 旅行需求 + 无 hard reject + 足够提出有用补问。分数帮助排序，但不能替代必填字段。 | 补问最少必要信息。若是 spam、不可执行、或合理跟进后无回应，则 archive/nurture。 | “想去日本”足够建 workspace，但不够 RFQ；AI 补问日期/月、人数、天数、酒店等级/预算、兴趣。 |
| 2 需求定向 | 把模糊需求变成供应商可报价 brief。必须先搜索参考行程库。 | RFQ 前必填：目的地/区域/主题；日期/月 + 天数；人数 + 成人儿童/房型假设；预算或酒店等级/消费能力；旅行风格/兴趣；特殊限制已知或标记未知；客户意向信号。加分项：航班、餐饮风格、导游语言、床型、must-not-do。 | RFQ brief 完整，且客户意向足够高。无参考行程匹配时：高意向 → 问 DMC；低意向 → 继续补问；不适合 → 婉拒/nurture。 | 定向补问缺失字段；客户不说预算时给预算区间；不要为模糊或不匹配客户消耗 DMC。 | 瑞士预算太低 → 不正式 RFQ；建议缩短行程、淡季、降低酒店、换目的地。 |
| 3 供应商询价 | 从 DMC/供应商拿到真实、可执行、可报价方案。 | RFQ 包必须包含：客户 brief、日期/月/天数、人数/房型、预算/酒店等级、路线/主题、must-haves、限制、回复 deadline、要求的 quote 格式。DMC 回复必须包含：逐日行程、具体酒店或清楚酒店等级+房型假设、价格/币种/人数 basis、包含/不含、车辆/导游、条款、有效期、取消条款。 | 至少一个 DMC 回复通过必填完整度和行程质量门槛。通常发两家 DMC；谁最先回复且达标，就先用谁，不必无谓等待。 | 针对缺失字段追 DMC；第一家不完整时，一边追一边等第二家；deadline 前无人达标则升级或找 backup DMC。 | DMC 只给总价但无酒店名/房型假设 → 不能报价；AI 要求酒店清单、房型、税费 basis、取消条款。 |
| 4 报价成交 | 把供应商方案转成客户报价，并保证毛利、证据和条款清楚。 | 必填：选定 DMC 方案、source-linked 成本、毛利计算、汇率/税费假设、客户版包含/不含、报价有效期、付款计划、取消/变更条款、审批卡。 | 客户报价必须审批。进入出行服务：客户书面确认 + 付款条件已核实 + 供应商承诺审批完成。 | 做 QC、毛利和风险 gate。任何价格、付款、折扣、供应商锁位或例外必须人工审批。 | DMC 报价明显高于初估 → 审批卡写明原因、毛利影响和替代方案，再对客户报价。 |
| 5 出行服务 | 交付旅程，并把经验回流数据库。 | 必填：客户确认、付款状态、供应商确认、最终 itinerary、按需收集旅客姓名/护照/航班、紧急/当地联系人、内部 owner 和升级路径。 | 旅程完成，未结费用/投诉/退款/invoice/反馈解决，供应商评分更新后关闭。 | 出发前 checklist；监控 deadline；处理紧急情况；归档反馈并更新 DMC/experience 数据库。 | 航班延误 → 联系 DMC 调整接送，记录费用/责任；如涉及钱或投诉则升级。 |
4. Bespoke 各阶段评分量化标准
是的，建议用评分作为阶段 gate。评分决定 case 是否准备好进入下一阶段；但是否可以自动执行，仍要再经过 AI 置信度和风险判断。
4.1 接单评分 — 从接单判断进入需求定向，门槛 ≥60/100
| 评分项 | 分值 | 满分标准 | 部分/低分例子 |
|---|---|---|---|
| 真实联系人和归属 | 15 | 姓名/联系方式/渠道清楚；内部 owner 或销售负责人明确。 | 转发的模糊消息 = 5–8。 |
| 业务线匹配 | 20 | 明确是 bespoke/private、MICE 或高价值旅行服务。 | 旅行需求不清 = 8–12；非旅行业务 = 0。 |
| 基础旅行信息 | 25 | 至少有 3 项：目的地/主题、日期/月、人数、天数、酒店/偏好。 | 只有目的地 = 8–10。 |
| 商业匹配 / 预算信号 | 20 | 预算、酒店等级、客户历史或需求显示符合 Dynasty 定位。 | 预算明显偏低 = 0–8。 |
| 意向 / 紧迫度 | 10 | 客户有真实出行窗口、要 proposal/follow-up，或是转介绍/老客户。 | 随便看看 = 2–4。 |
| 无硬性拒绝因素 | 10 | 不是 spam，不是明显不可执行，不是完全超出服务范围。 | 纯低价且无弹性 = 0–3。 |
4.2 RFQ Ready 评分 — 从需求定向进入供应商询价，门槛 ≥75/100
| 评分项 | 分值 | 满分标准 | 部分/低分例子 |
|---|---|---|---|
| 目的地 / 路线范围 | 20 | 主目的地或路线已收敛到供应商可报价。 | 只有“欧洲”=5；两个国家二选一=12。 |
| 日期 / 天数 | 15 | 明确日期，或可报价月份 + 天数。 | “学校假期”=5–8。 |
| 人数 / 房型 | 15 | pax、成人儿童、房型假设足够清楚；误差 ≤20%。 | 只有“一家人”=5。 |
| 预算和酒店等级 | 20 | 预算匹配 ≥0.8，或酒店/消费画像支持预期成本。 | 预算不匹配=0–8；不说预算但有 luxury hotel 信号=12–15。 |
| 旅行风格 / 必要需求 | 15 | 兴趣、节奏、导游/司机、餐食/活动需求清楚。 | 泛泛说“轻松”=5–7。 |
| 特殊限制 | 10 | 行动、饮食、签证、儿童老人、航班限制已识别。 | 未知=0–3。 |
| 客户 commitment | 5 | 已付定金或强意向/deadline；否则只能 soft RFQ。 | 随便问问=1。 |
4.3 供应商回复完整度 — 从供应商询价进入报价成交，门槛 ≥80/100
| 评分项 | 分值 | 满分标准 | 部分/低分例子 |
|---|---|---|---|
| 行程内容 | 20 | 逐日路线和体验可用、合理、符合客户。 | 通用复制行程=8–12。 |
| 酒店 / 房型假设 | 15 | 酒店名或等级、房型假设、替代方案清楚。 | 只写“4星酒店”=6–8。 |
| 车辆 / 导游细节 | 15 | 车型、导游语言、司机/导游时间和路线物流清楚。 | 只说有交通无车型/导游细节=6–8。 |
| 价格和币种 | 20 | 成本、币种、人数基础、有效期、税费/服务费假设清楚。 | 只有总价无 basis=8–10。 |
| 包含 / 不包含 | 10 | 餐食、门票、小费、机票、行李、税费清楚分开。 | 包含项模糊=3–5。 |
| 条款和取消 | 10 | 付款 deadline、取消/修改条款、报价有效期清楚。 | 无取消条款=0–3。 |
| 质量匹配 | 10 | 符合 Dynasty 品质，无明显路线/品牌/可行性问题。 | luxury fit 差或路线不合理=0–5。 |
4.4 报价 Ready 评分 — 准备客户报价，门槛 ≥85/100
| 评分项 | 分值 | 满分标准 | 部分/低分例子 |
|---|---|---|---|
| 来源证据 | 20 | 供应商 quote/API fare/source link 已存 evidence ledger。 | 价格无来源=0–8。 |
| 成本和毛利计算 | 25 | 内部成本、markup/margin、汇率/税费假设和例外清楚。 | 有成本但未核毛利=8–12。 |
| 客户报价假设 | 15 | 人数、房型、日期、包含/不含、有效期清楚。 | 假设隐藏或模糊=4–8。 |
| QC / 可行性 | 15 | 路线、季节、酒店、导游/车辆、活动可行性已检查。 | 供应商行程未 QC=0–6。 |
| 付款和条款 | 15 | 定金/尾款、due date、取消/修改条款准备好。 | 付款计划缺失=3–6。 |
| 审批准备度 | 10 | 审批卡有价格、毛利、证据、风险和推荐。 | 审批缺证据/理由=3–5。 |
4.5 执行 Ready 评分 — 从报价成交进入出行服务，门槛 ≥90/100
| 评分项 | 分值 | 满分标准 | 部分/低分例子 |
|---|---|---|---|
| 客户确认 | 15 | 客户书面确认 itinerary、价格、条款、旅客假设。 | 口头/不清楚确认=5–8。 |
| 付款证据 | 25 | 定金/全款已到账并匹配 case。 | 客户说会付但未核实=0–5。 |
| 供应商确认 | 20 | DMC/酒店/航司/邮轮确认或 booking reference 已保存。 | 供应商只说 noted，无确认=5–8。 |
| 旅客资料 | 15 | 姓名/护照/航班/饮食/行动/紧急联系人按需完整。 | 缺护照或航班=5–10。 |
| 最终文件 | 15 | 最终 itinerary、voucher、联系人、briefing 准备好。 | 只有 draft itinerary=5–8。 |
| 运营责任 | 10 | 内部 owner、DMC 当地联系人、升级渠道、deadline 清楚。 | 无升级联系人=3–5。 |
建议：评分要作为 gate 和诊断工具，而不是盲目自动化。分数要显示“缺什么”；AI 动作应该是“补问最高影响的信息”或“升级审批”，不是简单 fail。
5. “需求定向”100 分评分表
需求定向不能凭感觉“差不多”。ERP 应显示缺口，AI 优先补问最影响推进的信息。
| 评分项 | 分值 | 满分标准 | 部分得分例子 |
|---|---|---|---|
| 目的地 / 主题清晰度 | 20 | 国家/城市/路线主题清楚，例如北海道滑雪+温泉。 | “欧洲”=5；“瑞士或意大利”=12。 |
| 日期 / 天数清晰度 | 15 | 明确日期或 ≤14 天窗口，并知道天数。 | “年底”=5；“12月中，约10天”=12。 |
| 人数 / 房型 / 旅客结构 | 15 | pax、成人儿童、房数、床型、年龄/行动需求已知。 | “一家人”=5；“2大2小，2间房”=13。 |
| 预算 / 消费能力匹配 | 20 | 预算、酒店等级或历史消费证明符合 Dynasty bespoke 定位。 | 不说预算但指定 Aman/Four Seasons=15；明显低=0–8。 |
| 旅行风格 / 兴趣 | 15 | 节奏、酒店偏好、餐食、文化/自然/购物/亲子等清楚。 | “轻松一点”=6；偏好明确=12–15。 |
| 限制 / 风险因素 | 10 | 签证、行动、饮食、航班、老人儿童、季节风险已识别。 | 未知=0；部分已知=5。 |
| 推进意愿 / 决策链 | 5 | 决策人、proposal deadline、是否愿意推进/deposit 清楚。 | 随便问问=1；明确推进=5。 |
规则：≥75 可以进入供应商询价；60–74 继续定向；<60 回到接单补问。即使总分 ≥75，如果预算匹配度低于 0.8，也不能正式 RFQ，除非人工批准 soft RFQ。
6. 参考行程库优先 + AI 不自由编行程规则
在需求定向阶段，AI 不应该凭通用知识自行编行程。它只能：改造已有 Dynasty 参考行程、继续补问客户，或在高意向时问 DMC。
| 步骤 | 规则 | AI 动作 | 人类角色 |
|---|---|---|---|
| 1 | 先搜索参考行程库。 | 按目的地、季节、天数、旅客画像、旅行风格、预算等级、酒店等级和兴趣匹配。 | 只有匹配质量不确定或客户高价值时需要 review。 |
| 2 | 如果匹配质量高，用参考行程作为基础。 | 使用最接近的行程，再按客户节奏、酒店、活动和交通需求调整，并标明哪些是已确认、哪些是假设。 | 只有涉及客户 proposal、价格或不寻常承诺时审批。 |
| 3 | 如果没有好匹配但客户意向高，问 DMC。 | 向合格 DMC 发 neutral RFQ / proposal request；不承诺客户价格，不要求 DMC booking/blocking。 | 客户报价必须等供应商证据、QC、毛利和审批完成。 |
| 4 | 如果没有好匹配且客户意向低，继续补问。 | 用最少必要问题补齐资料，可给预算区间/例子，引导客户明确需求，直到意向足够高。 | 除 VIP/敏感或进入风险事项外，不需要人类介入。 |
| 5 | 如果客户仍不适合，礼貌关闭。 | 例如预算明显低于 bespoke 门槛、拒绝提供必要信息、只比最低价、或需求不可执行。 | 用礼貌 close/nurture 模板；不要消耗 DMC 资源。 |
7. 高意向 DMC 快速通道
如果参考行程库没有好匹配，AI 应判断客户是否高意向。若高意向，可以跳过 AI 自由设计行程，直接进入 DMC proposal/RFQ 路径。
| 信号 | 分值 | 判断方式 |
|---|---|---|
| 老客户 / 转介绍 / 过去成交客户 | 25 | 强烈说明询问值得供应商投入。 |
| Brief 很详细 | 25 | 目的地、日期、人数、天数、预算/酒店等级、偏好大多已提供。 |
| 具体称呼 | 15 | 邮件称呼 Dynasty 或具体员工名字，而不是泛泛 “Hi” / “Dear Sir”。 |
| 明确出行窗口或 proposal deadline | 15 | 客户有真实时间线并希望推进。 |
| Luxury / 消费能力信号 | 10 | 5星酒店、私家车、套房、商务舱、特殊 access、luxury ryokan 等。 |
| 积极互动 | 10 | 客户回复快、多轮沟通、问具体下一步问题。 |
决策规则：High Intent Score ≥70/100 → AI 可以准备 DMC proposal/RFQ。若 RFQ 是 neutral、不承诺客户价格、不要求 DMC 锁资源，AI 可直接处理并记录证据。任何涉及钱、报价、折扣、付款、退款、供应商锁位、投诉、取消或重大临时更改的事项，都属于有风险，必须人工审批。
| 例子 | 高意向判断 | AI 动作 |
|---|---|---|
| 老客户询问 “12 月北海道，4 pax，5-star ryokan，private driver”。 | 高意向。 | 先搜参考行程；无好匹配时，向日本 DMC 发 no booking commitment RFQ。 |
| 新客户写 “Dear Dynasty team, 6 pax Spain/Portugal Dec 2026, prefer 5-star and private guide”。 | 大概率高意向。 | 有参考行程则使用；没有好匹配则可走 DMC fast track。 |
| 邮件只写 “Hi, please send Europe package”。 | 低意向。 | 不问 DMC，先补问关键问题。 |
8. 两家 DMC RFQ 与首个达标回复规则
默认向两家合格 DMC 发送同一份 neutral RFQ。但系统不应为了等齐两家而拖慢：若第一家先回复且完整度和行程质量已经达标，就先用它进入 QC、报价和审批。
| 规则 | AI 动作 | 重要控制 |
|---|---|---|
| 条件允许时发两家 DMC。 | 向两家 approved DMC 发送同一份结构化 RFQ brief，并明确回复格式和 deadline。 | 若目的地只有一家强 DMC 或时间太紧，记录为什么只询一家。 |
| 用 checklist 追踪回复完整度，而不是只看 free text。 | 把每家 DMC 回复解析成必填字段：行程、具体酒店/房型、价格/币种/basis、包含/不含、导游/车辆、条款/取消、有效期。 | 缺硬性字段时，即使行程听起来不错，也不能报价给客户。 |
| 首个达标 DMC 进入第一 proposal path。 | 若 DMC A 最先回复，并通过必填完整度 + 行程质量门槛，就用 A 作为 working proposal 做 QC、pricing 和客户报价审批。 | 除非 A 方案弱、太贵、不完整或 case 战略重要，否则不要无谓等待 DMC B。 |
| 第二家 DMC 作为比较 / backup。 | 若 DMC B 在客户报价前回复，则比较 value；只有明显更好才切换，否则保存为 benchmark 和 backup。 | 保留选择 A 的证据：回复速度、完整度、质量、价格、匹配度。 |
| 定向 DMC follow-up 工作流。 | 若回复不完整，AI 发精确补问清单，而不是泛泛说 “please advise”。 | 例：“请提供酒店名和房型”“请分开包含/不含”“请确认有效期和取消条款”。 |
| 必要时才 apple-to-apple。 | 若第一家方案强但需要价格 benchmark，AI sanitise 结构后让另一家按同假设报价。 | 发送前删除 DMC 名字、logo、水印、metadata 和可识别格式。 |
9. Bespoke 场景示例
| 情况 | 阶段 | AI 判断 | AI 动作/建议 | 风险 |
|---|---|---|---|---|
| 新客户只说“日本/欧洲旅行” | 1 接单判断 | 有效 lead，资料不足。 | 建 workspace；一次性补问关键问题；之后先搜索参考行程库再提出方向。 | 无风险 |
| 预算低于 bespoke 门槛 | 2 需求定向 | 期望差距。 | 解释预算逻辑；建议缩短、淡季、降酒店、换目的地。 | 无风险 |
| 客户拒绝透露预算 | 2 需求定向 | 无法准确设计。 | 给 2–3 个预算档位让客户选方向。 | 无风险 |
| RFQ 后改目的地 | 2 需求定向 / 3 供应商询价 | 核心 brief 改变。 | 暂停旧 RFQ，重新评分；如已有承诺则审批。 | 无风险/高 |
| 机票价格取舍 | 2 需求定向 / 4 报价成交 | 舒适度/预算/时间取舍，价格必须有来源。 | 用 Sabre/NDC/API；显示来源和路线影响；客户报价前审批。 | 高 |
| 未付定金要详细 proposal | 3 供应商询价 | 高意向但无 commitment。 | 只做 soft RFQ；不做 blocking/booking commitment。 | 无风险 |
| DMC 回复不完整/质量差 | 3 供应商询价 | 不能报价。 | 定向追问、QC 反馈，并与第二家 DMC 或 backup DMC 比较。 | 无风险 |
| 客户确认但未核实付款 | 4 报价成交 | 不能做不可逆承诺。 | 付款提醒；无证据不得出票/锁酒店/DMC。 | 高 |
| 投诉/退款/赔偿 | 5 出行服务 | 品牌和财务风险。 | 收集事实，提出补救，退款/赔偿必须审批。 | 高 |
10. 三个 Bespoke 从头到尾典型示例
这些示例说明系统如何从第一封询问运行到旅程结束和知识沉淀，参考 Dynasty 常见历史案例类型和运营模式。
| 示例 | 案例画像 | 系统如何流转 | AI 在各阶段怎么做 |
|---|---|---|---|
| A. 日本冬季家庭私家团 | 家庭想 12 月去日本，要雪景、美食、温泉、私家车；一开始预算不清。 | 1 接单 → 2 定向 → 3 询价 → 4 报价 → 5 出行。 | 1：建 workspace，补问人数、儿童年龄、日期、房型、预算、酒店等级。 2：先搜索参考行程库；若有北海道滑雪温泉或东京+箱根+轻井泽高匹配行程，则基于它调整，并说明冬季成本和酒店限制。 3：尽量向至少两家日本 DMC 发 RFQ；要求司机、导游、酒店等级、雪地活动、餐食、取消条款。 4：QC 行程、检查毛利，准备保品质版和控成本版；报价必须审批。 5：收护照/航班/饮食，确认 DMC 联系人，监控抵达，旅后更新 guide/hotel/experience 质量。 |
| B. 西班牙葡萄牙家庭私家团 | 家庭想 Spain/Portugal，出行时间较远，人数/房型可能变化，客户也会比较路线和机票舒适度。 | 1 接单 → 2 路线定向 → 3 DMC RFQ → 4 修改报价循环 → 5 出行。 | 1：匹配或创建客户/trip workspace，记录 pax、月份、路线兴趣。 2：比较 Madrid/Seville/Lisbon/Porto 路线、节奏和季节；标记日期/人数/房型是否稳定。 3：尽量向至少两家 Iberia DMC 发 RFQ；如果一家行程设计更好，AI 可 sanitise 后让另一家按同酒店/餐厅/车型报价，做 apple-to-apple 比较。 4：AI 比较设计质量和 value-for-money；任何日期、房型、人数变化触发 change request；客户价格需人工审批。 5：付款证据确认后才锁服务；生成出发前 checklist，并把最终 Iberia 路线沉淀给未来案例。 |
| C. 预算偏低的欧洲询问 | 客户想瑞士/欧洲 private trip，但预算低于 bespoke 门槛，一开始也不愿明确 commitment。 | 1 接单 → 2 需求定向/期望管理 → 可能 3 soft RFQ 或 nurture/关闭。 | 1：有效 lead，但标记 budget-risk。 2：目的地/日期可能分数高，但预算匹配低；AI 不自动 RFQ。先解释真实预算，并提供替代：减少城市、淡季、4星酒店、奥地利/北意、缩短天数。 3：只有在人类批准或客户明显高意向时，AI 才发 no blocking 的 soft RFQ。 4：若供应商报价确认不匹配，AI 建议调整范围或礼貌关闭/nurture，而不是低毛利硬接。 5：若成交则正常进入出行服务；若关闭，则保存偏好用于未来 EDM/follow-up。 |
11. Bespoke 人类 + AI 协作机制
AI 接入早期建议刻意简化：无风险事项 AI 处理；有风险事项人工审批。 不再单独设置中风险。
| 类别 | 执行机制 | 人类参与 | 例子 |
|---|---|---|---|
| 无风险 | AI 直接执行，并记录动作、证据和置信度。 | 人类通过 activity log / digest 事后查看。 | 建 workspace、分类、提取信息、补问客户、搜索参考行程库、发送无承诺 neutral DMC RFQ、追 DMC 缺失字段、内部总结。 |
| 有风险 | 必须审批。AI 生成审批卡，批准/修改/拒绝前不执行。 | 人类必须在 ERP 审批中心决策。 | 任何涉及钱款流转、客户报价、折扣、付款确认、退款、赔偿、invoice、供应商锁位/预付款、DMC 报价明显高于初估、投诉、客户突然取消、DMC 临时重大改行程、法律/品牌敏感承诺。 |
重点：AI 早期不要让员工每次判断低/中/高。系统只问一个问题：“这件事会不会形成财务、供应商、客户、法律或品牌承诺？” 是 → 审批；否 → AI 带 audit trail 自动推进。
12. 无风险事项的 AI Action Queue
取消中风险后，queue 更简单：它不是审批队列，而是 AI 可执行动作的透明活动车道，管理者可以暂停/修改。
| 动作类型 | 默认行为 | 人类可操作 | 何时升级为有风险 |
|---|---|---|---|
| 普通客户补问 | 置信度高则 AI 发送；低置信度则只起草。 | 改模板、暂停 case、指定 owner。 | VIP/敏感措辞、投诉语气、客户问价格/折扣/退款。 |
| Neutral DMC RFQ / follow-up | AI 可发给已批准 DMC，但不做 booking/blocking 承诺。 | 增删 DMC、调整 deadline、暂停供应商。 | 新 DMC、高价值 case、要求 hold space、价格承诺、紧急临时更改。 |
| 缺资料提醒 | AI 礼貌跟进并记录回复。 | 改频率、停止提醒、转人工。 | 72 小时内出发、缺护照/签证/航班、老人儿童风险、客户不满。 |
| 供应商数据/数据库更新 | AI 抽取结构化字段并绑定 source link。 | 在 digest 中 review 改动。 | 商业承诺、blacklist/termination、公开使用供应商内容。 |
推荐通知节奏：默认进入 case activity log + 每天 2–3 次 digest。只有新审批卡、动作失败、紧急出发风险或投诉才即时通知。
13. 一整天：人类 + AI 交互示例
示例说明 planner / manager 如何监督多个 Bespoke case，而不需要逐条人工检查。
| 时间 | 发生什么 | AI 动作 | 人类体验 |
|---|---|---|---|
| 09:15 | 日本家庭询问邮件进来。 | 创建 Trip Workspace，分类为 Bespoke，提取目的地/日期/人数信号，接单评分 62/100，起草补问邮件。 | 不打断人类。动作进入 Autopilot Queue：“30 分钟后发送补问”。 |
| 09:40 | Planner 短暂打开 dashboard。 | 显示 4 个待执行无风险动作和 1 个高风险报价审批。 | Planner 修改一个日本补问，检查一个 DMC RFQ，其余因无风险让 AI 发送。 |
| 10:30 | 上午摘要。 | 汇总：3 个动作已自动执行，2 个即将执行，1 个因缺预算 blocked。 | Planner 看摘要，不需要逐个打开 case。 |
| 11:20 | 西葡 DMC 回复，但缺房型假设和取消条款。 | 供应商完整度 68/100。AI 直接发送定向 follow-up；低置信度时只起草。 | 普通 DMC follow-up，不即时打扰，只出现在 queue。 |
| 12:05 | 客户问能不能降价。 | 分类为高风险，因为影响价格/毛利。AI 生成审批卡：坚持价值、降酒店、缩短路线、经理折扣。 | 人类收到审批通知。不会自动发。 |
| 14:10 | 日本客户确认预算和日期。 | 需求定向分数升到 82/100。AI 建议进入供应商询价，并起草给 2 家日本 DMC 的 RFQ。 | 无风险 RFQ 可由 AI 发送；Planner 可按需要调整 DMC shortlist。 |
| 15:30 | 下午摘要。 | 汇总已发 RFQ、DMC deadline、blocked 高风险报价、临近出发但缺护照 case。 | Planner 只处理异常。 |
| 16:45 | 明天出发的客户缺航班信息。 | 临近出发例外，AI 即时通知并起草提醒。 | 人类看到紧急通知，立即批准发送。 |
| 18:00 | 下班前摘要。 | 列出已完成动作、待审批、明天 DMC deadline、需要人类决策的 case。 | Manager 只看 exception list，不看每封邮件/task。 |
产品结论：人类应该把时间花在异常和判断上，而不是监控。ERP 应让 AI 用可见倒计时和完整 audit trail 运行常规无风险动作。
1. 其他业务 Timeline 与条件
| 业务线 | Timeline | 关键 Gate | 人工审批点 |
|---|---|---|---|
| MICE / Incentive | 企业资格判断 → 活动/旅行范围确认 → 供应商/Tender 询价 → Proposal/合同/付款 → 团队/活动执行 → Debrief | 决策链清楚；pax/date/objective/budget 清楚；tender deadline；房表/名单需求；现场责任定义。 | proposal 价格、合同条款、付款计划、供应商承诺、tender 例外、现场风险计划。 |
| Group Tour | 产品匹配 → 座位/价格检查 → 报名 → 付款/文件 → 团队执行 → 反馈 | 有产品、出发开放、有座位、价格有效、旅客资料、付款核实。 | 折扣、waitlist、取消/退款、特殊房型、产品例外。 |
| Ticketing | 捕获机票需求 → 查票价与可订性 → PNR 占位 → 发票与收款 → 出票 → 退改签服务 | 票价来源 live；客户确认后记录 PNR 占位/TTL；出票前完成发票与收款。 | PNR 占位、出票、退款/改签、航变选项、财务例外。 |
| Ancillary | 附加到 Trip → 确认需求 → 找供应商 → 报价/付款 → 确认 → 服务支持 | active trip 或战略客户；供应商可用；条款清楚；确认前付款/审批。 | VIP 服务确认、特殊 access 承诺、退款/赔偿、供应商预付款。 |
| DMC / Supplier | 识别供应商 → 注册/Profile → 样板行程/价格 → QC/评分 → 数据库更新 → 用于 RFQ | 联系人、目的地、能力、价格、条款、样板行程、质量评分已捕获。 | 商业合作承诺、敏感谈判、终止/黑名单、公开使用内容。 |
2. 其他业务场景示例
| 情况 | 业务线/阶段 | AI 动作 | 风险 |
|---|---|---|---|
| 企业 incentive 询问 | MICE / 资格判断 | 问 pax、目的地/日期、objective、预算、决策人、tender deadline、机票/陆地分工。 | 无风险 |
| 客户问固定团 | Group Tour | 查 active product；无合适团则建议 private/bespoke 或 nurture。 | 无风险 |
| 只要机票 | Ticketing | 用可出票来源报价；出票前收款；出票本身高风险审批。 | 高 |
| 单独签证请求 | Ancillary | 若附属于 trip，则加 task；否则婉拒/nurture/转介。 | 低/无风险 |
| DMC 发注册表或行程 | DMC / Supplier | 解析注册/价格/能力；QC 行程；更新 DMC/experience 数据库。 | 低/无风险 |
1. AI 置信度和风险控制
flowchart TD
A[AI 提出判断/动作] --> B{置信度 >= 0.80?}
B -- 否 --> H[人工 review 或补证据]
B -- 是 --> C{是否产生钱/承诺/投诉/变更风险?}
C -- 是 --> H[必须生成审批卡]
C -- 否 --> L[AI 直接执行
记录 audit] L --> Z[更新 workspace、task、evidence、下一步] H --> Z
记录 audit] L --> Z[更新 workspace、task、evidence、下一步] H --> Z
| 类别 | 执行 | 例子 |
|---|---|---|
| 无风险 | AI 直接执行并记录证据。 | Workspace、分类、字段更新、内部总结、无承诺 neutral DMC RFQ/follow-up。 |
| 有风险 | 必须人工审批。 | 客户报价、invoice、付款/退款、折扣、供应商锁位、投诉、取消、低毛利/DMC 报价过高例外、重大临时变化。 |
2. ERP 架构建议
Lead / Trip Workspace
核心对象。所有邮件、WhatsApp、电话、文件、task、quote、proposal、invoice、payment、approval 都归属这里。
Sticky Business Tabs
ERP 可参考本文结构：Overview、Bespoke、Other Business、ERP/AI controls。
Business-Line Classifier
统一入口分流到 Bespoke、MICE、Group、Ticketing, Cruise, FIT、Ancillary、DMC/Supplier 或 Other。
Stage Engine
每条业务线有固定 timeline 和量化 gate。复杂情况做 subtask/revision。
AI Action Queue
每个 AI 动作都有置信度、是否有风险、证据、deadline、负责人、动作内容、结果/审批状态。
Approval Center
有风险动作展示财务影响、客户/供应商承诺、source link、替代方案。
Evidence Ledger
没有来源证据，不允许价格、付款、供应商锁位、退款。
Exception Dashboard
看 stuck cases、overdue tasks、pending approvals、payment risk、low margin、DMC delay、VIP/complaint、departure risk。
建议：不要只做更好看的 task list。先做 case-centric operating system：分类 → 阶段 gate → AI action → 风险/审批 → evidence → learning。
1. 产品文档建议
把流程图升级成 PRD 初稿
这份文档不应只是静态流程图，而应作为 Dynasty ERP / AI Workflow 的产品需求文档初版。
按阶段补 ERP 字段
每个阶段要定义必填字段、AI 自动字段、人工确认字段和输出物。
必填字段先于评分
MVP 阶段，硬性必填字段应阻断阶段流转；评分用于诊断缺什么、排序优先级、建议下一步。人类 override 必须记录原因。
用真实案例验证规则
每条规则都要用真实案例类型测试：日本家庭、西葡、低预算欧洲、MICE、机票-only、DMC proposal。
2. ERP 系统建议
| 建议 | 为什么重要 | 要做什么 |
|---|---|---|
| 以 Case 为中心，不以 task list 为中心 | Task 只有挂在客户/trip/supplier context 上才有价值。 | Lead / Trip Workspace 作为核心对象。 |
| 异常驾驶舱 | 管理层不需要看每个聊天/task，而要看卡住的 case、逾期审批、付款风险、DMC 延迟和投诉。 | 带 exception cards 和 SLA filters 的 Dashboard。 |
| AI Action Queue | AI 不应只是聊天回复，而要生成结构化动作。 | Action object：case、stage、confidence、risk、evidence、timer、owner、result。 |
| 独立审批中心 | 高风险财务/客户/供应商动作需要统一审批体验。 | 审批卡展示证据、财务影响、推荐和替代方案。 |
| Evidence Ledger | 价格、付款和供应商承诺必须有 source link。 | 把 evidence record 绑定到 quote、payment、approval、supplier confirmation。 |
3. 基于历史案例的 AI 早期接入建议
| 建议 | 为什么 | MVP 做法 |
|---|---|---|
| 按目的地建立 DMC 白名单 | AI 早期不应自己判断供应商质量。 | 每个目的地维护 primary DMC、backup DMC、专长、回复 SLA、质量分、可用 RFQ 模板。 |
| 先判断 fit/intent，再消耗供应商资源 | 历史上低预算或模糊询问最容易浪费 planner 和 DMC 时间。 | 显示意向分和缺失项；只有高意向且无参考行程匹配时才问 DMC。 |
| 审批规则只保留一个定义 | 员工越容易理解，AI 越容易落地。 | “钱/承诺/投诉/变更/取消 = 审批”；其他 AI 记录后推进。 |
| 价格先证据后报价 | 防止数字虚构/四舍五入，也保护毛利。 | 没有供应商/API/source link、毛利计算和审批轨迹，就不能生成 quote card。 |
| 每次 DMC 回复都沉淀学习 | 数据库只有持续吸收供应商表现才会越来越强。 | 自动更新回复速度、完整度、报价偏差、路线质量和可复用体验。 |
4. MVP 优先级建议
| 优先级 | 模块 | 范围 |
|---|---|---|
| 1 | Trip Workspace | 客户、询问、对话、文件、阶段、task、approval、evidence。 |
| 2 | Bespoke 5-stage timeline | 接单判断 → 需求定向 → 供应商询价 → 报价成交 → 出行服务。 |
| 3 | AI scoring + 缺失项诊断 | 必填字段 gate + 诊断评分：lead 优先级、RFQ ready、DMC 回复完整度、报价 ready、执行 ready。 |
| 4 | Task panel + AI Action Queue | 下一步动作、有风险/无风险标记、负责人、due time、证据、结果。 |
| 5 | Approval Center | 客户报价、invoice、付款、供应商锁位、退款、低毛利例外。 |
| 6 | Evidence Ledger | 价格、付款、供应商报价、确认、审批的 source link。 |
Dashboard Demo UI — Timeline + Task 面板
这是一个轻量产品 mockup：左侧 case 列表，中间是阶段 timeline 和证据/评分，右侧是 AI task/action 面板。