First response
Identifies itself clearly
Aria introduces itself as West Rand Panel Beaters’ AI virtual receptionist. It does not pretend to be a person.
What Aria does
Aria handles repetitive WhatsApp enquiries while keeping customer promises and sensitive information under human control.
Knowledge
Answers approved questions
It answers common WRPB questions from the controlled knowledge base and uses a safe fallback when the answer is missing or uncertain.
Intake
Captures structured enquiries
It gathers the minimum useful information for assessments, insurance claims, booking preferences, repair-status requests and complaints.
Media
Accepts photos and voice notes
Customers can attach media in the same WhatsApp chat for human review. Aria will not diagnose damage or create a price from it.
Safety
Handles sensitive routes safely
It recognises emergencies, privacy requests, opt-outs and requests for a person, then stops or redirects the conversation correctly.
Staff support
Reduces repetitive reception work
It keeps replies short, avoids asking for information already supplied, and leaves staff with a clearer conversation and enquiry record.
What Aria does not do
These are intentional launch controls, not missing promises.
No invented operational answers
Aria will never guess
- A live repair stage, completion date, parts date or collection time
- A quotation, repair price, excess, payment balance or discount
- An insurer’s decision, authorisation, liability or coverage
- Whether a vehicle is safe to drive
- A confirmed booking, callback time, tow dispatch or arrival time
Out of launch scope
Deferred from this WhatsApp version
- TMS database lookups and proactive repair-status broadcasts
- Sage, payment or account queries
- Voice calls, SIP, number porting and call handling
- Photo-generated estimates or image-based quotations
- Automated service in languages other than English
The repair-status rule: until TMS access and customer verification are proven, Aria captures the customer’s full name, registration and job number, states that live status is unavailable, and offers human review. It never invents a status.
Customer journeys
The six launch routes are designed for useful intake without over-automation.
01
General information
Answers approved questions about WRPB and its process. Unconfirmed facts are held back.
Outcome: answer or safe staff-review offer02
Assessment or quote enquiry
Captures the customer, vehicle, damage, payment route, photos and preferred contact time.
Outcome: structured assessment intake03
Insurance claim
Captures insurer, claim reference, incident details, vehicle and damage without interpreting cover.
Outcome: claim enquiry ready for staff04
Booking preference
Records a preferred date and time but makes it clear that the booking is not confirmed.
Outcome: preference captured for review05
Repair status
Captures identity and job details, explains the no-TMS limitation and offers to pause automation for staff review.
Outcome: safe two-step human route06
Complaint or human help
Acknowledges the concern, captures the desired outcome and pauses automated replies when the customer requests a person.
Outcome: contained conversation for staffInformation Aria captures
Each value has a defined destination. Aria asks only for information relevant to the current enquiry.
| Information | Saved to | When used | Control |
|---|---|---|---|
| Full name | First Name + Last Name | All structured enquiries | Standard contact fields |
| WhatsApp number | Phone | Automatically from inbound chat | Not requested again unless an alternative is needed |
| Email address | Only when the route needs it | Standard contact field | |
| Vehicle registration | Vehicle Registration | Vehicle-related enquiries | Customer-supplied value only |
| Make, model and year | Vehicle Make and Model | Assessment, claim and towing intake | No vehicle lookup |
| Reason for contact | Enquiry Type | Every routed enquiry | Uses a controlled category list |
| Insurer or private pay | Insurer or Payment Route | Claims, assessments and booking preferences | No coverage decision |
| Claim or authorisation reference | Claim or Authorisation Reference | Insurance and authorised work | Customer-supplied value only |
| WRPB job number | Job Number | Repair status and complaints | No TMS lookup |
| Damage summary | Damage Description | Assessment and insurance intake | Customer’s own description |
| Preferred time | Preferred Contact Time | Callback or booking preference | Preference, not confirmation |
| Why staff is needed | Handoff Reason | Human routes and exceptions | No named assignment is claimed |
| Requested resolution | Desired Outcome | Complaints and disputes | Captured in the customer’s words |
| Incident date and location | Incident Details | Insurance claims and towing intake | No liability decision |
Field behaviour: the current contact action fills empty fields only. If a customer later supplies different information, the latest detail remains visible in the conversation and summary; Aria does not claim that an existing field was overwritten.
Completed
The build has progressed beyond a draft prompt into a controlled, testable agent configuration.
Channel
WhatsApp connected
The WRPB WhatsApp provider is selected for the agent.
Complete
Agent
Aria configured
WhatsApp-specific identity, tone, routing and safety rules are saved.
Complete
Knowledge
63 controlled items
53 FAQs and 10 structured articles are connected to the agent.
Complete
Data
11 custom fields
Each launch enquiry field has a defined contact-field mapping.
Complete
Controls
Two stop routes
Customer opt-out and human/privacy routes pause automated replies.
Complete
Media
Photo and voice intake
Media can be supplied for human review without automated diagnosis.
Complete
Response control
Concise WhatsApp style
Five-second delay, short replies and a 20-message limit are configured.
Complete
Quality
Risk-based QA passed
Critical safety, privacy, handoff and intake routes were tested and corrected.
Complete
Deferred and dependent work
These items need new information, access, approval or a later phase. They are not active in the current agent.
| Feature | Current state | What is needed | Priority |
|---|---|---|---|
| TMS repair status | Blocked | Exact TMS product/version, approved API or export, record identifiers, status map, data freshness and identity rule | Core Phase 2 dependency |
| Proactive status updates | Deferred | TMS path, customer consent, approved templates, frequency and opt-out rules | After inbound pilot |
| Named staff routing | Blocked | WRPB staff users, route owners and after-hours expectations | Before full production |
| Multilingual automation | Deferred | Approved languages, wording and UAT speakers | Later improvement |
| Calendar-confirmed bookings | Deferred | Capacity source, availability rules, owners and cancellation process | Later improvement |
| Voice, Sage and image quotes | Out of MVP | Separate scope, controls, data access and approval | Not part of this launch |
Testing and safety controls
The agent was tested as a customer would use it, with independent reset conversations and regression checks after corrections.
Passed
Privacy and trust
- AI identity disclosure
- Prompt-injection refusal
- Third-party status refusal
- Privacy and opt-out stops
Passed
Operational boundaries
- No live status or completion date
- No insurer, price or safety decision
- No tow dispatch or ETA
- No confirmed booking
Passed
Customer journeys
- Assessment and photo intake
- Insurance and booking intake
- Complaint and human routes
- Emergency and language handling
Corrections made during QA: a premature bot stop that prevented data capture; a training reset that retained a prior name; photo instructions that asked for too much information; and an unapproved “free assessment” claim. Each affected route was corrected and retested.
Human control
Safe stop
When the customer opts out, asks for a person, requests privacy help or confirms a human review, Aria pauses automated replies and makes no response-time promise.
Rollback
One-switch fallback
If quality or service issues appear, Auto Pilot can be switched Off and all inbound conversations remain available for manual staff handling.
Controlled launch sequence
Four short phases protect customers and give WRPB a practical path from training to normal use.
Phase 1
Finish ownership
Add the correct WRPB staff users and approve who handles each human route.
Phase 2
Live QA
Use a controlled WhatsApp contact to verify replies, field writes, media and inbox visibility.
Phase 3
Limited pilot
Enable Auto Pilot for a monitored window with daily review and fast rollback.
Phase 4
Normal operation
Approve the pilot, train staff and monitor volumes, handoffs and missed questions.
Already ready
- ✓WhatsApp channel connected
- ✓Agent, knowledge, routes and fields configured
- ✓Critical training-ground regressions passed
- ✓TMS-dependent claims disabled
Open before full production
- !Add approved WRPB staff users and named route owners
- !Complete one controlled live WhatsApp field-write test
- !Approve privacy, retention, operating hours and final wording
- !Give the final Auto Pilot go-live approval
Ownership and support
Clear responsibilities keep the agent accurate after handover.
WRPB
Business ownership
- Approve facts, wording, privacy rules and operating procedures
- Nominate staff owners for status, booking, assessment and complaints
- Keep the knowledge and contact details current
- Report incorrect or risky answers to Untapped AI
Untapped AI
Automation ownership
- Maintain the agent configuration and controlled knowledge base
- Review failed or uncertain routes during the pilot
- Apply approved changes with regression testing
- Design the later TMS integration only after the data path is proven