Exactly. I would make the AI ask the internal team first, not the client. That is the key to making this genuinely fast. The workflow should feel like an AI onboarding conversation, not a traditional onboarding form. The ideal flow Step 1 — Deal becomes "Won" Salesperson clicks: Convert to Client The CRM creates the client and activates the AI Onboarding Assistant. The AI already knows most things from the CRM: Company Contact Lead source/campaign Service/product Sales notes Meeting notes Proposal Quoted price, if available Previous communications Then it asks the internal user only for information it cannot reliably determine. Step 2 — AI asks the salesperson a few questions Instead of a 30-field form: AI: I have enough information to start onboarding ABC Ltd. I need 5 details: 1. Project value? £________ 2. Project duration? ○ One-off ○ 1 month ○ 3 months ○ 6 months ○ 12 months ○ Custom 3. Payment terms? ○ Full upfront ○ Monthly ○ 50/50 ○ Custom 4. Planned start date? 📅 18 September 5. Who will manage this client? [Select team member] Then: Anything else that is different from your standard service? [Optional] And that's it. Start Onboarding → The AI takes over. Step 3 — AI builds the onboarding plan It should dynamically decide what is required. For example: Client purchased: AI Automation Implementation £6,000 3 months Monthly support included The AI determines: Required Commercial Service agreement Invoice Payment terms Legal NDA Terms DPA, if applicable Operational Client requirements Existing systems Access requirements Implementation plan Kick-off meeting Communication Welcome Signature request Payment communication Requirements request Kick-off confirmation It creates everything. Step 4 — Client gets a very small onboarding experience This is where I'd make your system different. Do NOT send them a huge form. Send: Welcome to Ascentra We're ready to begin your project. It will take approximately 3 minutes to complete the first step. Let's get started → Then use one question at a time. Screen 1 What would you like us to achieve? [Type your answer] Next → Screen 2 Which systems do you currently use? ☑ Xero ☑ HubSpot ☐ Salesforce ☐ Microsoft 365 ☐ Other Next → Screen 3 Who should we contact for this project? Name: ______ Email: ______ Phone: ______ Screen 4 Do you have any existing documents or processes you'd like us to use? 📎 Upload files Skip for now → Done. The AI should adapt the questions This is the really important part. Suppose they answer: "We want to automate our lead qualification." The AI can understand that. It doesn't need to ask: "What is your lead qualification objective?" It already knows. Instead it might ask: Where do your leads currently come from? ○ Website ○ Facebook ○ LinkedIn ○ Email ○ Other Then based on their answer: Where are new leads currently recorded? If they choose HubSpot: We'll use HubSpot as the starting point. Do you want us to automate: ☑ Lead capture ☑ Lead qualification ☑ Follow-up ☐ Reporting This makes the onboarding feel intelligent rather than bureaucratic. Use "progressive disclosure" This is the UX principle I'd use throughout your CRM. Don't ask for everything at once. Ask: What is necessary now? Then ask the next question only when the answer creates a reason for it. For example: Do you process personal data? No → Don't ask about DPA. If: Yes → Then ask the relevant data-processing questions. Similarly: Do you need API integration? No → Don't ask for API credentials. Yes → Ask which system. Then: Which system? HubSpot → Show HubSpot-specific requirements. This can reduce a 40-question onboarding form to perhaps 5–10 questions for many clients. AI should also reuse information This is critical. If the client already told your salesperson: "We use Xero." Don't ask the client: "Which accounting system do you use?" Instead show: We understand you currently use Xero. Is that correct? Yes / Change That's much faster. Same with: Company name Address Website Contact Service Project objective Budget Timeline Number of users The CRM should never ask a question twice. I'd also add "Save & Continue Later" Clients are busy. If they stop halfway: You're 60% complete. We've saved your progress. [Continue onboarding] No need to restart. And let the client use their phone Since your signing system is already yours, make the whole experience mobile-first. They should be able to: Open link → answer questions → upload document → sign agreement → finish without creating an account or sitting at a computer. The final experience should look something like this DEAL WON ↓ AI reads CRM ↓ AI asks INTERNAL USER 5-7 missing commercial details ↓ Human submits ↓ AI creates personalised onboarding plan ↓ AI generates documents ↓ AI prepares communications ↓ ┌──────────────────────────────┐ │ HUMAN APPROVAL GATE │ │ │ │ Approve / Edit / Regenerate │ └──────────────┬───────────────┘ ↓ Client receives link ↓ ┌────────────────────┐ │ 3–5 min onboarding │ │ │ │ Question → Answer │ │ Question → Answer │ │ Upload if needed │ └─────────┬──────────┘ ↓ Documents signed ↓ Invoice/payment ↓ AI checks everything ↓ ONBOARDING COMPLETE ↓ ACTIVE CLIENT And I would add one killer feature: AI should tell the internal team what it still needs — rather than making them understand the process. For example: ABC Ltd is 92% ready for activation. ✅ Agreement signed ✅ NDA signed ✅ Invoice paid ✅ Client requirements received ✅ Implementation owner assigned ❌ API access missing AI recommendation: Request HubSpot API access from client. [Generate request] → [Review] → [Send] That is the difference between a CRM with "AI features" and a CRM where AI actually operates the onboarding process. My recommendation would be to design this around three separate AI conversations: 1. Internal AI → Sales/Operations "Give me the missing commercial details." 2. AI → Client "Give me only the information I need, one step at a time." 3. AI → Human approver "I've prepared everything. Here's what I want to send and why." That structure will keep the system fast for the client, extremely light for your staff, and still controlled by humans where it matters. Lets finalise the onboarding system so i can share the full development plan to my dev to start building. We will use that only for us. Not for public use Absolutely. Since this is an internal CRM system only, I would build it for operational efficiency rather than trying to make it a polished public SaaS onboarding platform. Below is the version I would give directly to your developer. The important point is that AI prepares and coordinates the entire process, but a human approval gate remains before anything external is sent or any legally/commercially significant action is executed. AI-Powered Internal Client Onboarding System Development Specification — V1 1. Core objective When a lead becomes a client, the CRM should automatically start an AI-powered onboarding workflow. The system should: Understand what the client purchased. Ask the internal team only for missing commercial/project information. Determine what onboarding steps are required. Generate the required documents. Generate client communications. Collect only necessary information from the client. Handle digital signatures through our existing signing system. Connect with the existing invoicing system. Track compliance and completion. Automatically prepare follow-ups. Maintain a complete audit trail. Require human approval before external communication or significant execution. Core philosophy AI does the work. Human approves the work. CRM records the work. 2. Conversion trigger When: Lead → Opportunity → Won/Client the CRM should automatically create: Client Record Client ├── Company information ├── Contacts ├── Services purchased ├── Commercial information ├── Documents ├── Agreements ├── Invoices ├── Communication history ├── Onboarding ├── Projects ├── Support └── Audit history The existing CRM data should be reused. Never ask the user/client for information already available in the CRM unless verification is required. 3. AI Onboarding Initiation Immediately after conversion, the AI analyses the available information: Lead record Opportunity Campaign Service/product Proposal Sales notes Meeting notes Previous communications Price Existing client information Then AI identifies missing information. Internal AI prompt interface Instead of a large form: I'm ready to onboard ABC Ltd. I have most of the information. I need 5 details: Project value: £6,000 Duration: 3 months Payment: Monthly Start date: 18 September Account manager: [Select] Anything unusual about this client or agreement? [Optional text] Start AI Onboarding The number of questions should be dynamic. If CRM already contains price and duration, don't ask again. 4. Service Onboarding Templates This should be the foundation of the system. Create an internal: Service Onboarding Template Engine Every service/product can have its own configuration. Example: AI Automation TalentBridge Cybersecurity Management Consultancy Digital Product Training Enterprise Service Custom Service Each template defines: Documents required Example: Welcome Letter NDA Service Agreement Terms & Conditions Privacy Notice DPA Invoice Implementation SOP Project Brief Information required Company information Project objective Technical requirements Contacts Access requirements Data requirements etc. Signatures required NDA → Client Agreement → Client DPA → Client Commercial requirements Price Duration Payment terms Start date Renewal Workflow Agreement → Signature → Invoice → Payment → Requirements → Kickoff → Activation This means developers don't need to create a new workflow every time you introduce a service. 5. AI Onboarding Planner After receiving the missing internal information, AI creates an: Onboarding Plan Example: Client: ABC Ltd Service: AI Automation Value: £6,000 Duration: 3 months AI-generated plan Commercial Generate invoice Prepare payment terms Legal NDA Service Agreement DPA Client information Requirements Current systems Project contacts Access requirements Operational Implementation SOP Project brief Kickoff meeting Communication Welcome Signature request Payment communication Requirements request Kickoff confirmation AI should also explain: DPA required because this service involves processing client personal data. 6. Document Generation Engine The CRM needs a central: Document Template Library Do not allow AI to freely invent important legal documents. Instead: Approved Template + Client Data + Service Data + Jurisdiction + Commercial Terms ↓ AI-generated document Examples: NDA Service agreement Terms & Conditions DPA Privacy documents Welcome letter Project brief SOP Invoice Payment reminder Kickoff document 7. Document version control Every generated document must have: Document ID Client Document type Version Generated by Generated date Approved by Approval date Status Statuses: Draft AI Generated Human Review Changes Requested Approved Sent Signed Rejected Superseded Expired Never silently overwrite an approved document. If changed: Version 1 → Version 2 → Version 3 8. Human Approval Centre This is one of the most important components. Create: AI Approval Inbox Example: ABC Ltd AI has prepared 11 actions. Documents: 6 Welcome Letter — Ready NDA — Ready Service Agreement — Ready DPA — Ready Invoice — Ready Communications: 5 Welcome email Signature request Requirements request Payment email Kickoff email Each item: Preview | Edit | Regenerate | Approve | Reject 9. Approval rules Different actions should have different approval levels. Mandatory human approval Contracts NDA DPA Terms Legal documents Invoices Signature requests Important client communications Price changes Scope changes Refunds Cancellation Service activation Can potentially be automated Internal task creation Internal notifications Internal status changes Progress calculations Reminder scheduling AI summaries However, make the automation configurable. 10. Client onboarding interface The client should not receive a huge questionnaire. The system generates a secure onboarding link. The client sees: Welcome to [Company] We're ready to begin. This should take approximately 3 minutes. Start → Then: Question 1 What is the main outcome you want from this project? Answer box. Question 2 Which systems do you currently use? Selectable options. Question 3 Who should we contact regarding implementation? Contact details. Question 4 Do you have existing documents/processes you'd like us to review? Upload. Then finish. 11. Progressive questioning This is essential. The AI should decide the next question based on previous answers. Example: Do you need system integration? │ ├── NO → Continue │ └── YES ↓ Which system? ↓ HubSpot ↓ What access? Don't ask unnecessary questions. Rule: Never ask the client for information that is not required for the purchased service or compliance process. 12. Information already known If CRM already knows: Company = ABC Ltd show: Company name: ABC Ltd Is this correct? ✓ Yes ✎ Change Rather than asking: "What is your company name?" This makes onboarding much faster. 13. Save and continue Client can leave and return later. The system stores progress. Example: Onboarding: 60% complete [Continue] No need to restart. 14. Digital signature integration Integrate the existing internal signing system. Required workflow: Document approved ↓ Generate signature request ↓ Secure signing link ↓ Client opens on device ↓ Reviews document ↓ Signs ↓ Signature verification ↓ Signed document stored in CRM ↓ Audit record ↓ Workflow continues The signing system should support: Mobile Desktop Tablet Typed signature Drawn signature Signature fields Date Name Checkbox/consent Multiple signatories where required The CRM should know: Unsigned / Sent / Viewed / Partially Signed / Signed / Declined / Expired 15. Invoice integration Use the existing CRM invoicing module. Example: Agreement approved ↓ Invoice generated ↓ Human approval ↓ Invoice sent ↓ Payment status monitored Payment states: Not Created Draft Approved Sent Viewed Partially Paid Paid Overdue Cancelled Payment events should trigger onboarding workflows. 16. Communication engine Create a central: AI Communication Centre AI generates: Welcome emails Signature requests Missing-information requests Payment reminders Follow-ups Kickoff messages Progress updates Delay notifications Completion messages Renewal reminders Every communication should show: AI-generated Reason: Client has not provided required project information for 3 days. Recommended action: Send reminder. Then: Edit → Approve → Send 17. Follow-up automation Example: Client hasn't completed onboarding Day 0 Initial request. Day 2 AI prepares friendly reminder. Day 5 Second reminder. Day 7 Internal escalation. Day 10 AI flags: 🔴 Onboarding blocked This should be configurable per service. 18. AI Onboarding Health Each client receives an onboarding score. Example: 84% — 🟢 Healthy Legal 100% Commercial 100% Information 70% Operations 80% Client response 90% AI should identify the blocker. Current blocker: API access not provided. Recommended action: Request API access. [Generate message] 19. Onboarding dashboard Management should have: All Onboarding Total: 32 🟢 Healthy 22 🟡 At Risk 7 🔴 Blocked 3 Filters: Service Account manager Campaign Client Date Status Payment Compliance Overdue 20. Client onboarding timeline Every event should appear chronologically. Example: 08 Sep 14:20 Deal converted to client 08 Sep 14:21 AI onboarding initiated 08 Sep 14:23 Internal details completed 08 Sep 14:24 6 documents generated 08 Sep 14:35 Documents approved by John 08 Sep 14:36 Signature request sent 08 Sep 16:02 NDA signed 09 Sep 09:10 Agreement signed 09 Sep 09:15 Invoice sent 10 Sep 13:20 Payment received 10 Sep 13:21 Client questionnaire completed 11 Sep 10:00 Kickoff scheduled 21. Onboarding completion criteria Don't allow someone to simply click: "Complete onboarding." Instead the system calculates completion. Example: Required documents ✓ Required signatures ✓ Payment ✓ Client information ✓ Compliance ✓ Internal tasks ✓ Project setup ✓ Then: ✅ ONBOARDING COMPLETE Automatically create: ACTIVE CLIENT and move the client into: Delivery / Project / Support 22. Exception management This is where AI becomes really useful. The AI shouldn't only follow the happy path. It should detect: Client hasn't responded Payment overdue Document rejected Contract changed Client requested different scope Missing information Signature expired Internal task overdue Project start date approaching Compliance requirement missing Then: 🔴 Exception detected Client requested a change to the agreed project scope. This may affect pricing and contract terms. Recommended action: Human review required. This prevents AI from blindly continuing. 23. Audit trail Every important action must be logged. Who What When Why Previous value New value AI/Human Document version Approval status For example: 08/09/26 18:42 Service Agreement v2 approved Approved by: Operations Manager Generated by: AI Previous version: v1 This is particularly important for legal/commercial documents. 24. AI permissions I recommend building an AI Permission Matrix into the CRM. Action AI Human approval Read CRM data ✓ — Summarise meeting ✓ — Create internal tasks ✓ — Generate welcome ✓ ✓ Generate NDA ✓ ✓ Generate agreement ✓ ✓ Generate invoice ✓ ✓ Send contract — ✓ Request signature — ✓ Send payment reminder Configurable Configurable Change price — ✓ Change contract — ✓ Activate client — ✓/rules Create follow-up ✓ Configurable The permission system should be configurable rather than hardcoded. 25. Data model Your developer should create dedicated entities/tables for: Client Client Contact Opportunity Service Service Onboarding Template Onboarding Instance Onboarding Step Onboarding Requirement Document Document Template Document Version Approval Signature Request Invoice Payment Client Response Communication Communication Template AI Action AI Recommendation Exception Audit Event The most important relationship is: Client ↓ Onboarding Instance ↓ Onboarding Steps ↓ Documents / Signatures / Payments / Communications 26. AI action architecture I would make AI actions structured rather than allowing the AI to directly manipulate everything. For example: AI Recommendation Action: GENERATE_DOCUMENT Document: SERVICE_AGREEMENT Client: ABC Ltd Template: AI_AUTOMATION_V3 Variables: price = £6,000 duration = 3 months start_date = 18/09/2026 payment_terms = monthly Requires approval: YES Then the CRM executes the action only after approval. This is much safer than giving the AI unrestricted database access. 27. The overall user experience Your employee should ultimately see something as simple as: ABC Ltd AI Onboarding — 87% 🟢 Commercial 🟢 Compliance 🟢 Agreement 🟢 Payment 🟡 Client information 🟢 Internal setup AI says: I've prepared 8 actions. 3 require your approval. The client has completed 80% of onboarding. One item is outstanding: API access. [Review AI Actions] 28. The ideal complete workflow LEAD ↓ CAMPAIGN ↓ MEETING ↓ OPPORTUNITY ↓ WON ↓ CLIENT CREATED ↓ AI READS ALL EXISTING DATA ↓ AI IDENTIFIES MISSING INTERNAL DETAILS ↓ INTERNAL USER ANSWERS 3–7 QUESTIONS ↓ AI CREATES ONBOARDING PLAN ↓ AI GENERATES DOCUMENTS ↓ AI GENERATES CLIENT COMMUNICATIONS ↓ AI CREATES INVOICE ↓ ━━━━━━━━━━━━━━━━━━━━━━ HUMAN APPROVAL GATE ━━━━━━━━━━━━━━━━━━━━━━ ↓ APPROVED ITEMS SENT ↓ CLIENT ONBOARDING PORTAL ↓ PROGRESSIVE QUESTIONS ↓ DOCUMENT UPLOAD ↓ DIGITAL SIGNATURE ↓ PAYMENT ↓ AI MONITORS PROGRESS ↓ AUTOMATIC FOLLOW-UP PREPARATION ↓ EXCEPTIONS ESCALATED ↓ ALL REQUIREMENTS COMPLETE ↓ HUMAN FINAL APPROVAL ↓ ACTIVE CLIENT ↓ DELIVERY ↓ SUPPORT ↓ RENEWAL / UPSELL 29. What I would call V1 Don't try to build every AI capability at once. Phase 1 — Core engine Build: Client conversion trigger Service templates AI missing-information questions Onboarding instance Document templates AI document generation Approval centre Client onboarding link Progressive questions Digital signatures Invoice integration Communication generation Audit trail Onboarding dashboard Phase 2 — Intelligence Add: AI onboarding health score AI exception detection Automated follow-up recommendations AI document checking AI meeting-note extraction AI onboarding summaries Intelligent question generation Better service-specific workflows Phase 3 — Autonomous operations Eventually: AI manages the onboarding unless it encounters something requiring human judgement. That's the end goal. The human shouldn't be doing the onboarding. The human should be supervising the AI onboarding operation. One architectural decision I strongly recommend Build the system around this object: ONBOARDING INSTANCE Not around documents. A client may have 8 documents, 4 signatures, 2 invoices, 12 questions and 20 communications. Those are all components of one onboarding instance. That gives you a single source of truth: "Where exactly is this client in the journey, and what needs to happen next?" And the AI becomes the intelligence sitting on top of that state. If your developer builds the Onboarding Instance + Service Template + AI Action + Human Approval architecture correctly, you can later add virtually any new service to the CRM without rebuilding the onboarding system from scratch. Give me a prompt for developing the onboarding system itself Yes. I’d give your developer one master development prompt that tells them to build the actual internal system, not just design the concept. You can paste the following directly into Claude Code, Cursor, OpenCode, or give it to your development team. MASTER DEVELOPMENT PROMPT Build an AI-Powered Internal Client Onboarding System You are a senior full-stack architect and developer. You are building an internal AI-powered client onboarding module inside our existing CRM. This is NOT a public SaaS product. It is an internal operational system used by our team to convert won opportunities into fully onboarded clients with as little manual work as possible. The system must be designed around this principle: AI prepares, organises, checks and recommends. Humans approve. The CRM executes and records. The goal is to reduce manual onboarding work dramatically while maintaining human control, legal/commercial oversight, security and a complete audit trail. 1. PRIMARY OBJECTIVE When a lead/opportunity is converted into a client, the CRM must automatically initiate an AI-powered onboarding workflow. The system should: Read existing CRM information. Determine what information is already known. Identify only the information still missing. Ask the internal team for missing commercial/project details. Build a service-specific onboarding plan. Generate required documents. Generate required client communications. Generate the invoice through the existing CRM invoicing system. Prepare digital signature requests through our existing signing system. Present all externally significant actions to a human for approval. Send approved communications/documents. Give the client a simple, mobile-friendly onboarding experience. Ask the client only necessary questions. Use progressive questioning instead of a large questionnaire. Monitor onboarding progress. Detect delays, missing information and exceptions. Prepare follow-ups. Maintain a complete audit history. Automatically mark onboarding complete when all mandatory requirements are satisfied. Move the client into the normal active-client/delivery/support workflow. 2. IMPORTANT DESIGN PRINCIPLE DO NOT build this as a large onboarding form. DO NOT ask the client 20–50 questions at once. DO NOT ask the client for information already available in the CRM. DO NOT allow AI to independently send contracts, legal documents, invoices or important commercial communications. Instead use: CRM data ↓ AI analysis ↓ Missing information ↓ Internal AI questions ↓ Onboarding plan ↓ AI-generated actions ↓ Human approval ↓ Client interaction ↓ Digital signature/payment ↓ AI monitoring ↓ Completion 3. ONBOARDING INSTANCE Create a central entity called: OnboardingInstance Every client onboarding must have one. Example: OnboardingInstance id client_id opportunity_id service_id template_id status health_status progress_percentage onboarding_owner started_at target_completion_date completed_at created_at updated_at Statuses: NOT_STARTED INITIALISING WAITING_INTERNAL_INFORMATION AI_PLANNING WAITING_HUMAN_APPROVAL WAITING_CLIENT IN_PROGRESS AT_RISK BLOCKED READY_FOR_ACTIVATION COMPLETED CANCELLED Health: HEALTHY AT_RISK BLOCKED The OnboardingInstance is the central object connecting: Documents Questions Client responses Signatures Invoices Payments Communications Tasks Approvals Exceptions Audit events 4. CONVERSION TRIGGER When an opportunity changes to: WON automatically: Create/update Client. Attach opportunity to Client. Identify purchased service/product. Select appropriate Service Onboarding Template. Create OnboardingInstance. Start AI analysis. Do not duplicate existing client/contact information. 5. AI INITIAL ANALYSIS The AI should receive structured context from the CRM: Client information Contact information Opportunity information Campaign Lead source Product/service Proposal Sales notes Meeting notes Quoted price Duration Existing communications Existing documents Existing client history The AI should determine: Known information What is already known and reliable. Missing information What is genuinely required before onboarding can begin. Verification information What is known but should be confirmed. Optional information What would be useful but is not required. The AI must NOT ask unnecessary questions. 6. INTERNAL AI QUESTIONING Before contacting the client, AI should ask the internal user for missing commercial/project information. Example: ABC Ltd is ready for onboarding. I already have: ✓ Client information ✓ Service ✓ Proposal ✓ Project objective ✓ Contact information I need: 1. Project price 2. Duration 3. Payment terms 4. Start date 5. Account manager Anything unusual about the agreement? [Optional] Use appropriate UI controls: Currency field Date picker Dropdown Radio buttons Multi-select Short text Long text only where necessary Do not make this a generic form. The AI should dynamically generate the questions. 7. SERVICE ONBOARDING TEMPLATE ENGINE Create: ServiceOnboardingTemplate A template defines how a particular service/product is onboarded. Example: Service ↓ Onboarding Template ├── Required documents ├── Required information ├── Required signatures ├── Payment requirements ├── Compliance requirements ├── Client questions ├── Internal tasks ├── Communication sequence └── Completion conditions Templates must be configurable by authorised administrators. Do NOT hardcode onboarding logic for individual services. Example templates: AI Automation Cybersecurity TalentBridge Management Consultancy Training Digital Product Enterprise Service Custom Service The architecture must allow new services to be added without rewriting the onboarding engine. 8. AI ONBOARDING PLANNER After internal information is supplied, AI creates an onboarding plan. Example: Client: ABC Ltd Service: AI Automation Price: £6,000 Duration: 3 months Start: 18 September AI determines: Commercial Invoice Payment terms Legal NDA Service Agreement Terms DPA if applicable Operational Requirements System information Access requirements Implementation SOP Kickoff Communication Welcome Signature request Requirements request Payment communication Kickoff communication Each action must become an explicit workflow object. 9. AI ACTION ENGINE Create: AIAction Example: AIAction id onboarding_id action_type description payload risk_level requires_approval status created_at approved_at approved_by executed_at Action types: GENERATE_DOCUMENT GENERATE_COMMUNICATION CREATE_INVOICE REQUEST_SIGNATURE REQUEST_INFORMATION CREATE_TASK SCHEDULE_FOLLOWUP UPDATE_STATUS FLAG_EXCEPTION GENERATE_SOP GENERATE_PROJECT_BRIEF AI must NOT directly execute sensitive actions. Instead: AI proposes action ↓ Human approval ↓ CRM executes 10. HUMAN APPROVAL CENTRE Create a dedicated internal screen: AI Approval Centre Show all actions requiring approval. Example: ABC Ltd AI prepared 9 actions. Documents: ✓ Welcome Letter ✓ NDA ✓ Service Agreement ✓ DPA ✓ Invoice Communications: ✓ Welcome email ✓ Signature request ✓ Requirements request ✓ Kickoff email Each item must provide: Preview Edit Regenerate Approve Reject The human should also be able to: Approve All when appropriate. Every approval must be recorded. 11. APPROVAL LEVELS Implement configurable approval requirements. Always require human approval: Contracts NDA DPA Terms Legal documents Invoices Signature requests Scope changes Price changes Refunds Important commercial communications Client activation Potentially automated: Internal tasks Internal notifications Progress calculation Internal summaries Non-sensitive status updates Reminder preparation Never bypass mandatory approval rules. 12. DOCUMENT GENERATION SYSTEM Create: DocumentTemplate and: Document and: DocumentVersion Document generation must work like: Approved Template + CRM Data + Service Data + Commercial Data + AI-generated client-specific information ↓ Generated Document Document types: WELCOME_LETTER NDA SERVICE_AGREEMENT TERMS PRIVACY_NOTICE DPA INVOICE PROJECT_BRIEF IMPLEMENTATION_SOP REQUIREMENTS_DOCUMENT KICKOFF_DOCUMENT PAYMENT_REMINDER OTHER For legal documents, AI must populate approved templates rather than freely invent legal language. 13. DOCUMENT VERSION CONTROL Every document needs: document_id client_id onboarding_id document_type template_id version status generated_by generated_at approved_by approved_at sent_at signed_at superseded_by Statuses: DRAFT AI_GENERATED UNDER_REVIEW CHANGES_REQUESTED APPROVED SENT VIEWED SIGNED REJECTED SUPERSEDED EXPIRED Never overwrite an approved version. 14. CLIENT ONBOARDING PORTAL Create a lightweight secure client onboarding interface. The client should NOT need to navigate the CRM. They receive a secure onboarding link. Example: Welcome to [Company] We're ready to begin your project. This should take approximately 3 minutes. [Start] The client should be able to: Answer questions Confirm existing information Upload files Review documents Sign documents See progress Save and continue later Complete onboarding on mobile 15. PROGRESSIVE QUESTIONING This is mandatory. The AI must decide the next question based on: Service Previous answers Existing CRM data Compliance requirements Project requirements Example: Do you require integration? NO ↓ Skip integration questions. YES ↓ Which system? HubSpot ↓ Ask HubSpot-specific information. Never show a huge questionnaire. 16. INFORMATION REUSE If the CRM already knows: Company = ABC Ltd Website = abc.com CRM = HubSpot Contact = John Smith the client should see: We have your company listed as: ABC Ltd ✓ Correct ✎ Change The system should follow: Never ask twice unless verification is required. 17. CLIENT RESPONSE ENTITY Create: ClientResponse Store: question_id client_id onboarding_id response response_type submitted_at source verified AI should be able to use previous responses when deciding what to ask next. 18. DIGITAL SIGNATURE INTEGRATION We already have our own DocuSign-style digital signing capability. Integrate it into onboarding. Required states: NOT_REQUIRED DRAFT READY SENT VIEWED PARTIALLY_SIGNED SIGNED DECLINED EXPIRED Workflow: Document approved ↓ Signature request created ↓ Secure signing link ↓ Client opens on device ↓ Client reviews ↓ Client signs ↓ Signature completed ↓ Signed document stored ↓ Audit event created ↓ Next onboarding step triggered Support: Mobile Tablet Desktop Multiple signatories Signature fields Date fields Name fields Consent/checkbox fields Typed signature Drawn signature Use the existing signing infrastructure rather than creating another signing system. 19. INVOICE INTEGRATION Use the existing CRM invoicing system. Possible workflow: Agreement approved ↓ AI prepares invoice ↓ Human approves ↓ Invoice issued ↓ Payment monitored Payment states: DRAFT APPROVED SENT VIEWED PARTIALLY_PAID PAID OVERDUE CANCELLED Payment events must be available to the onboarding workflow engine. 20. COMMUNICATION ENGINE Create: Communication and: CommunicationTemplate AI can generate: Welcome Signature request Requirements request Missing information reminder Payment reminder Kickoff confirmation Progress update Completion message Delay notification Follow-up Every external communication should initially enter: WAITING_FOR_APPROVAL unless the action is explicitly configured as safe for automatic sending. 21. FOLLOW-UP ENGINE The workflow engine should monitor deadlines. Example: Day 0 Initial request Day 2 AI prepares reminder Day 5 AI prepares second reminder Day 7 Internal escalation Day 10 Onboarding marked BLOCKED These intervals must be configurable per service/template. 22. AI ONBOARDING MONITOR AI should continuously evaluate the onboarding state. It should identify: Missing client information Missing signatures Missing documents Payment delays Internal task delays Approaching deadlines Client inactivity Contract changes Scope changes Other exceptions Example: AI ALERT ABC Ltd Onboarding Health: AT RISK Reason: Client has not supplied required API access for 4 days. Recommended action: Request API access. [Generate message] [Review] 23. ONBOARDING HEALTH SCORE Calculate a health status from: Commercial Legal Compliance Client information Documents Payment Internal tasks Client responsiveness Deadline risk Display: HEALTHY AT_RISK BLOCKED Also display the exact reason. Do not make the score a meaningless percentage. The user must be able to understand: Why is this client at risk? 24. EXCEPTION ENGINE Create: OnboardingException Examples: MISSING_INFORMATION PAYMENT_OVERDUE SIGNATURE_DECLINED SIGNATURE_EXPIRED DOCUMENT_REJECTED SCOPE_CHANGE PRICE_CHANGE DEADLINE_RISK CLIENT_UNRESPONSIVE INTERNAL_TASK_OVERDUE COMPLIANCE_BLOCK Each exception should have: severity description detected_at recommended_action status assigned_to resolved_at resolved_by AI may recommend a solution but must not make sensitive decisions automatically. 25. COMPLETION ENGINE Do not allow users to manually declare onboarding complete without validation. The system must check required conditions. Example: Required documents ✓ Required signatures ✓ Required information ✓ Payment ✓ Compliance ✓ Internal tasks ✓ Project setup ✓ Then: READY_FOR_ACTIVATION A human can give final approval. Then: ACTIVE CLIENT 26. ONBOARDING TIMELINE Every action must appear in a chronological timeline. Example: 08 Sep 14:20 Opportunity converted to client 08 Sep 14:21 AI onboarding initiated 08 Sep 14:24 Internal information completed 08 Sep 14:25 AI generated 6 documents 08 Sep 14:40 Documents approved 08 Sep 14:41 Signature request sent 08 Sep 16:02 NDA signed 09 Sep 09:15 Agreement signed 09 Sep 09:20 Invoice issued 10 Sep 13:20 Payment received 10 Sep 14:00 Client requirements completed 11 Sep 10:00 Onboarding completed Timeline events must be immutable audit records. 27. AUDIT SYSTEM Create: AuditEvent Every significant action must record: event_id client_id onboarding_id user_id actor_type action timestamp old_value new_value document_version metadata Actor types: HUMAN AI SYSTEM CLIENT Examples: AI generated agreement Human approved agreement System sent agreement Client viewed agreement Client signed agreement AI detected missing information Human rejected document 28. AI CONTEXT The AI should have controlled access to relevant CRM information. Create a structured context object rather than sending the entire database blindly. Example: ClientContext Client Contacts Opportunity Products Services Proposal Meeting notes Sales notes Documents Previous communication Invoices Payments Onboarding state Previous responses Use permissions so AI only receives information necessary for the task. 29. AI SAFETY / PERMISSIONS The AI must operate under explicit permissions. Example: READ_CRM_DATA GENERATE_DOCUMENT GENERATE_COMMUNICATION CREATE_INTERNAL_TASK CREATE_INVOICE_DRAFT CREATE_SIGNATURE_REQUEST_DRAFT RECOMMEND_ACTION UPDATE_PROGRESS DETECT_EXCEPTION Separate these from: SEND_EXTERNAL_COMMUNICATION SEND_CONTRACT ISSUE_INVOICE REQUEST_SIGNATURE CHANGE_PRICE CHANGE_CONTRACT ACTIVATE_CLIENT Sensitive actions require human approval. 30. ADMIN SETTINGS Create an internal administration area where authorised users can configure: Services Service name Description Price model Onboarding template Documents Templates Versions Required approval Required signature Questions Question Type Required/optional Conditional logic Communication Templates Follow-up timing Approval requirements Compliance Required documents Required confirmations Service-specific requirements Workflow Steps Conditions Dependencies Escalations 31. SECURITY This is an internal system but must still follow strong security practices. Implement: Role-based access control Secure client onboarding links Expiring tokens Authentication where appropriate Encryption in transit Encryption at rest Document access controls Audit logging Secure file uploads Access expiration No sensitive data in URLs CSRF protection Rate limiting Input validation Permission checks server-side Never rely solely on frontend permissions. 32. GDPR / DATA GOVERNANCE The system should support appropriate GDPR/data governance controls. Implement architecture for: Data minimisation Purpose limitation Access control Audit history Retention policies Document retention Data export Data deletion where legally appropriate Consent/acknowledgement records where applicable Do not automatically assume that every client needs the same compliance documents. The service onboarding template should determine applicable requirements, with human oversight. 33. INTERNAL DASHBOARD Create: AI Onboarding Dashboard Display: Total onboarding Healthy At Risk Blocked Awaiting Client Awaiting Approval Awaiting Payment Ready for Activation Completed Filters: Client Service Campaign Account Manager Onboarding Owner Status Health Date Payment 34. CLIENT 360 INTEGRATION Once onboarding is complete, the OnboardingInstance should remain permanently connected to the Client record. Client 360 should show: Client ↓ Lead history ↓ Opportunity ↓ Onboarding ↓ Documents ↓ Contracts ↓ Invoices ↓ Payments ↓ Projects ↓ Support ↓ Communications ↓ Renewals Do not create duplicate client records when they purchase additional services. 35. AI SUMMARY At any time, the CRM should be able to generate: Client Onboarding Summary Example: ABC Ltd purchased AI Automation Implementation for £6,000 over 3 months. Agreement and NDA signed. Invoice paid. Client currently uses HubSpot and Xero. Project objective: Automate lead qualification and CRM entry. Outstanding: HubSpot API access. Onboarding health: AT RISK Recommended next action: Request API access. This summary should be generated from live CRM data. 36. TECHNICAL REQUIREMENTS Before coding: Inspect the existing CRM architecture. Identify current database structure. Identify existing authentication. Identify existing client/contact entities. Identify existing opportunity/deal entities. Identify existing service/product entities. Identify existing invoicing system. Identify existing document/file storage. Identify existing communication system. Identify existing digital signing system. Identify existing notification system. Identify existing AI/API infrastructure. DO NOT duplicate functionality that already exists. Extend existing systems wherever practical. 37. API-FIRST DESIGN Create clean internal APIs/services for: POST /onboarding/start GET /onboarding/{id} POST /onboarding/{id}/internal-details POST /onboarding/{id}/generate-plan GET /onboarding/{id}/actions POST /onboarding/actions/{id}/approve POST /onboarding/actions/{id}/reject POST /onboarding/actions/{id}/regenerate POST /onboarding/{id}/client-link GET /onboarding/{id}/progress POST /onboarding/{id}/complete Use appropriate REST/API conventions based on the existing CRM architecture. Do not blindly implement these exact endpoints if the existing application follows a different architecture. Adapt them consistently. 38. ASYNCHRONOUS JOBS AI generation, document generation, email preparation, file processing and monitoring should not block the main UI. Use the existing job/queue infrastructure if available. Potential jobs: AI_ANALYSE_ONBOARDING GENERATE_DOCUMENT GENERATE_COMMUNICATION CHECK_ONBOARDING CHECK_DEADLINES GENERATE_FOLLOWUP PROCESS_SIGNATURE_EVENT PROCESS_PAYMENT_EVENT CALCULATE_HEALTH DETECT_EXCEPTION Jobs must be idempotent where possible. 39. ERROR HANDLING If an AI action fails: Do not silently fail. Show: AI action failed. Action: Generate Service Agreement Reason: Required project duration is missing. Required action: Provide project duration. [Complete Information] If an external service fails: Signature service unavailable. Action has NOT been sent. Retry available. Never mark an action complete unless execution was actually confirmed. 40. TESTING Create automated tests for: Workflow Won opportunity creates onboarding. Correct template selected. Missing information detected. Plan generated. AI AI doesn't ask known questions. AI asks required questions. Conditional questions work. AI actions are structured. Approval Unapproved sensitive actions cannot execute. Approved actions execute correctly. Rejected actions don't execute. Documents Correct template selected. Correct client variables inserted. Versioning works. Signature Signature events update onboarding. Signed document stored correctly. Invoice Invoice status synchronises. Client portal Client can resume onboarding. Conditional questions work. Responses save correctly. Security Client cannot access another client's onboarding. Internal users only see authorised clients. Expired links cannot be used. 41. DEVELOPMENT PRIORITY Build in this order: PHASE 1 Core: OnboardingInstance Service templates Conversion trigger Internal AI questioning AI onboarding plan Action engine Approval centre PHASE 2 Documents: Template library AI document generation Version control Document approval PHASE 3 Client experience: Secure onboarding link Progressive questions File upload Save/resume Mobile interface PHASE 4 Commercial/legal: Digital signature integration Invoice integration Payment events PHASE 5 Communication: AI communication generation Approval Follow-up engine Reminder system PHASE 6 Intelligence: Health score Exception detection AI recommendations AI summaries Deadline monitoring PHASE 7 Management: Dashboard Analytics Audit system Service template administration AI permission management 42. DO NOT OVERBUILD V1 V1 should focus on making this workflow work perfectly: WON DEAL ↓ AI asks internal user for missing details ↓ AI creates onboarding plan ↓ AI generates documents ↓ Human approves ↓ Client receives simple onboarding link ↓ Client answers progressive questions ↓ Client signs required documents ↓ Invoice/payment ↓ AI monitors ↓ Human resolves exceptions ↓ CLIENT ACTIVATED Do not build unnecessary public-facing functionality. Do not build a separate CRM. Do not build another invoicing system. Do not build another e-signature platform. Integrate with existing CRM components. 43. FINAL UX PRINCIPLE The internal user should feel like they are working with an AI operations assistant. Not: "Complete this 30-field onboarding form." Instead: AI: ABC Ltd is ready for onboarding. I already have most information. I need five details before I can prepare everything. Then: AI: I've prepared 9 actions. Three require your approval. Then: AI: The client has completed onboarding except for API access. I've prepared a request for your approval. Then: AI: Everything required is complete. Client is ready for activation. The human's job becomes: Answer → Review → Approve → Handle Exceptions The AI's job becomes: Analyse → Prepare → Generate → Monitor → Recommend → Coordinate The CRM's job becomes: Execute → Track → Store → Audit 44. SUCCESS CRITERIA The system will be considered successful when a normal client can go from: Won Deal → Active Client with minimal human intervention. Target: Internal setup: under 2 minutes Client onboarding: 3–10 minutes for a normal client No duplicate questions No unnecessary forms All required documents generated automatically All sensitive external actions require approval Digital signing works from mobile Invoice integrates automatically Follow-ups are automatically prepared Every action has an audit record Exceptions are clearly surfaced New services can be onboarded by creating a new template rather than rebuilding the system MOST IMPORTANT RULE Do not build a collection of disconnected onboarding features. Build a single AI-driven Onboarding Engine where: Client + Service + Template + Workflow + AI Actions + Human Approval + Client Responses + Documents + Signatures + Invoice + Communication + Audit Trail all belong to one OnboardingInstance. The system should always be able to answer: Who is this client? What did they buy? What has already been completed? What is still required? Who needs to act? What should happen next? Can AI do it, or does it require human approval? What evidence do we have that it happened? Build the system around those questions.