onboarding.txt
Rama main · réplica de lectura del repositorio original.
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.