Contacts engaged
Unique eligible contacts messaged
Build review
Metric values are modelled placeholders. Verify every claim before publication.
National solar business · Case study
2,838 sales appointments booked before a salesperson entered the conversation.
HostMetric connected source-aware lead response, AI qualification, live appointment booking, missed-call recovery, stalled-lead follow-up, database reactivation and performance optimisation into one production system.
41-second first contact6 lead sources6 connected journeys38% booked outside business hoursRuns 24/7
A production system running every day across a high-volume Australian sales floor. Not a demo.
Production journey · Live stateRunning
Recovery layer feeds back in
Lead source identifiedName validatedPostcode → local timeVariant selectedCalendar checkedAppointment createdCRM task assignedReply checked before follow-upConversation loggedLead source identifiedName validatedPostcode → local timeVariant selectedCalendar checkedAppointment createdCRM task assignedReply checked before follow-upConversation logged
Why this was not one automation
Every lead arrives at once, from somewhere different, expecting something different.
A customer who just missed you on the phone is not the same as a customer who filled in a form at 11pm. A comparison-shopping enquiry is not the same as someone who came from a search result knowing exactly what they want.
They can all arrive on the same afternoon. They need different first sentences. And whoever answers first usually wins.
At volume, that stops being a messaging problem and becomes a coordination problem: who responds, how fast, with what, in which local time, who follows up, who notices the missed call and who returns to the leads written off months ago.
“The problem was never that the business could not respond well. It was that responding well every single time, at that volume, is not something you can staff for.
Six sources→One queue→First contact depends on who is free→Same message regardless of origin→Calendar checked by hand→Some follow-ups happen→Missed calls hit voicemail→Dormant leads sit→No evidence of which message worked
The brief was not to automate the first SMS. It was to make sure every lead event — including the ones that look like failures — has a defined next action.
Source-specific journeys
A customer who missed your call should not receive the same message as a customer comparing installers.
Source detection changes the opening, follow-up timing, agent behaviour and specialist calendar—not just a CRM label.
Telephony event · immediate recovery
The customer already tried to reach you.
Acknowledge the attempt first, apologise and remove the wait. This conversation should not begin like a cold web enquiry.
Opening matched to contextFollow-up logic selectedSpecialist route retained
Source detection is not cosmetic. It determines the agent, the opening, the follow-up schedule and which specialist’s calendar gets checked.
How it actually works
A short conversation on the surface. Considerable orchestration underneath.
Customer viewSMS
Assistant
Thanks for your enquiry. Before I find you a time — what's your name and postcode?
Customer
Dave, 3145
Assistant
Thanks Dave. Are you after battery only, or solar and battery together?
Customer
Battery only
Assistant
We've got Tuesday 11:30am, Wednesday 2:00pm or Thursday 10:15am for a quick call. Which suits?
Customer
Wednesday 2
Assistant
You're locked in for Wednesday at 2:00pm. We'll call you on this number.
System viewAction 1 of 16
- 01
Inbound reply received
✓ - 02
Contact matched in CRM
✓ - 03
Lead source retained
✓ - 04
Last five messages loaded into context
✓ - 05
Name validated — passed
✓ - 06
Postcode 3145 → Victoria → local-time rule resolved
✓ - 07
Product interest → battery only
✓ - 08
Battery specialist route selected
✓ - 09
Live calendar queried
✓ - 10
Business-hour rules applied
✓ - 11
Three valid slots returned
✓ - 12
Appointment created
✓ - 13
CRM task assigned and tagged
✓ - 14
Pipeline stage → Appointment booked
✓ - 15
Confirmation and reminder scheduled
✓ - 16
Conversation appended to CRM record
✓
Four customer messages. Sixteen system actions. The salesperson receives a booked appointment with the qualification already recorded—and never touched a keyboard.
Beyond appointment booking
Booking the appointment was the easy part. Catching everything that fell out was the build.
Three failure paths return to the same production qualification and booking spine. Nothing becomes a dead-end campaign.
Failure eventRecovery logicBooking spine
Trigger · No reply within the configured window
The lead went quiet halfway through.
The system waits until the approved local-time window, checks whether a genuine reply has arrived and only then sends a follow-up asking for a suitable time to call.
Why it is difficult
A follow-up sent at 9am Sydney can be 6am Perth. Chasing a customer who already replied is worse. Both failures disappear when timing and reply checks are treated as system rules.
731
stalled enquiries recovered into appointments
Every route returns to the same booking flow. Not three campaigns—one system with three additional ways in.
Control, not autonomy for its own sake
It has one job. It does not have opinions about anything else.
The assistant’s KPI is to get an appropriate customer into a booked phone appointment quickly. Pricing, technical advice and the sale remain with people.
AI handles84% autonomous
- First contact in 41 seconds
- Name and postcode capture
- Product interest and specialist routing
- Approved early-stage questions
- Live calendar availability
- Booking, confirmation and reminders
- Stalled-lead follow-up
- Missed-call recovery
- CRM data capture
Team handles9% escalated
- Pricing and formal quotes
- Rebate eligibility
- Technical suitability
- Non-standard installations
- Commercial and apartment enquiries
- Complaints and sensitive issues
- The recommendation
- The sale
84%
of conversations ran end to end without human involvement
9%
escalated with the full conversation and qualification attached
The point was never to replace the sales team. It was to stop them spending their day arranging calls instead of having them.
The difficult work most builds overlook
The difficult work lived between the happy paths.
A convincing message is not the production build. The build is what happens when the customer, data or system behaves unexpectedly.
01
Names that are not names
Numbers, test entries and blank first-name fields are validated before they reach customer-facing copy. Invalid records fall back to a neutral greeting rather than "Hi 0412334."
02
Local time that changes
Postcode determines state, state determines the local-time rule, and daylight saving changes that rule for some states but not others. Follow-up fires where the customer is.
03
The customer already replied
Every scheduled follow-up checks for a genuine inbound response before sending. A customer who has answered never receives a chase message.
04
Landlines
A landline cannot receive recovery SMS. The event raises an internal callback alert rather than failing silently.
05
Duplicate messages
Outbound events are checked against conversation memory and recent sends. The system does not talk over itself.
06
Out-of-scope work
Commercial projects, apartments, unusual installs and sensitive conversations are disqualified or escalated. The assistant does not improvise.
07
Opt-outs
Opt-out intent stops the journey immediately and updates the contact record. Suppression is checked before downstream escalation.
08
The record
Every exchange, qualification signal, appointment and handoff is appended to the CRM for the salesperson who receives it.
Successful AI automation depends far less on writing a good message than on controlling what happens when reality does not cooperate.
Measurement and optimisation
Automating the messages was half the job. Knowing which ones worked was the other half.
Opening-message journeys were configured for weighted variant testing and tracked through reply, booking and opt-out.
Conversation performance
90-day viewSent6,900Tested pair
Response lift+34%Winning opening
Outcome tracking4Sent · replied · booked · opt-out
Sent100%
Replied26.9%
Booked16.3%
Opted out6.8%
Live allocation control
Shift traffic without rebuilding the workflow.
The tested comparison-marketplace opening produced a 34% higher response rate across 6,900 sends. Allocation can be adjusted while the journey stays live.
Variant A50%
Variant B50%
- Full-funnel tracking
- Variant-level outcomes
- Source breakdowns
- Sample-size warnings
- Daily and weekly trends
The system did not stop improving at launch. It started generating the evidence required to improve it.
Performance · 1 April – 30 June 2026
One connected system. Ninety days.
Each headline figure is paired with the denominator it describes. The larger number is never substituted for the more useful one.
Response rate
Unique repliers ÷ 17,412 contacts messaged
Appointments booked
Unique appointments created by the system
Booking rate
2,838 bookings ÷ 17,412 contacts messaged
Recovered appointments
731 stalled + 604 missed calls + 412 dormant
How the numbers are counted+
Response rate
Unique contacts who replied ÷ unique contacts sent an opening message.
Booking rate
Unique booked appointments ÷ all unique contacts messaged. We report 16.3%, not the larger 60.6% reply-to-booking rate.
Recovered appointments
A booking is attributed once to the recovery journey that produced it. Cross-journey duplicates must be removed.
After-hours bookings
38% occurred outside the approved local business-hour window; 14% occurred overnight or at weekends.
Why this worked
The conversation was only one layer of the system.
The implementation treated source context, operational boundaries, recovery paths and measurement as part of the product—not additions after launch.
01
Map the journey before selecting the automation.
Lead sources, system events, ownership, timing rules and failure paths were made explicit before the conversation layer was built.
02
Preserve the context that created the enquiry.
Source detection changes the opening message, agent behaviour, follow-up schedule and specialist route.
03
Return every valid path to one booking spine.
New leads, stalled conversations, missed calls and dormant contacts all use the same qualification, availability and CRM logic.
04
Define where the assistant must stop.
Pricing, technical suitability, rebate eligibility and sensitive issues remain owned by qualified people.
05
Instrument the system so it can improve.
Replies, bookings, opt-outs and message variants are measured through the full funnel rather than judged from anecdotes.
Your turn
Where does your customer journey still depend on someone remembering?
Bring us your lead sources, systems and sales process. In 30 minutes, we will trace the path from enquiry to outcome and mark every place where it depends on a person being available, remembering or getting to it.
You keep the map. Whether we work together or not.
The Customer Journey Map
- Every lead source, traced from arrival to outcome
- Every handoff, delay and manual dependency
- What breaks when volume doubles
- What should be automated
- What should remain human
- A practical map that is yours to keep
Map my customer journey
Tell us where enquiries enter and where the process currently slows down.
Find the manual dependencies
Map your customer journey