Client case
Client anonymised at their requestThe routine calls get made.The difficult ones still reach a person.
A large local-services platform put AI phone service into its sales division. The system calls stores on a schedule to ask how their opening is progressing and answer what it can, and it answers the support line when stores call in: sorting each request, resolving some on the spot, and handing the rest to a person. Every conversation is recorded, and every record can be traced afterwards.
- Store progress stays current
- Routine questions get answered or routed
- Every call stays traceable
Written from the project's own account of the work. The client's name, region, organisation and store identities are withheld.
The situation
Starting from the work itself
A store needs steady contact between preparing to open and trading normally. The sales team wants to know which step a store is on and what is in the way. The store wants someone to answer when something goes wrong. Those two needs run down the same phone line, which makes it both the support queue and the place where follow-up actually happens.
Progress does not update itself
A store's opening is not settled in one conversation. Someone has to check in on a schedule and notice the ones where something has changed.
Routine questions eat the day
Repeat calls and standard queries take real time. The people who should be having the difficult conversations spend their day having the easy ones.
Not every problem is the same problem
Requests range from trivial to tangled. They have to be sorted before they can be routed, so each one ends up on the path that suits it.
A call log is not a record
Knowing a call happened is not the same as knowing what was said. Without the second part there is nothing to improve against.
These are the management concerns the work is organised around, drawn from the real setting. They are not a measured finding about how large the problem was before launch.
The work
Two directions, one set of records
The service runs along two paths that share the same record. Calls go out on a schedule; calls come in when a store needs something. AI carries the routine part of both, and a person takes over wherever judgement is needed.
Outbound follow-up
Call on schedule
Ask how the store's opening is progressing.
Answer what it can
Deal with the question, or note it down.
Keep the record
Retain the notes and the recording.
Carry it forward
Give the next conversation something to start from.
Inbound service
Answer the call
Find out what the store needs.
Sort the request
Separate what can be handled from what cannot.
Route it
Resolve it directly, or hand it to a person.
Stay traceable
Leave something a manager can return to.
Every conversation starts the same way: understand the situation, deal with the problem, and leave something behind for the next review.
AI phone service
Handles scheduled contact, first-line questions, inbound sorting, and a share of the problems themselves.
Sales and support staff
Take anything needing a longer conversation, a business judgement, or coordination with someone else.
The manager
Reads the records and listens back to calls to work out what needs to change and who to follow up with.
Walkthrough
A store describes its progress on a scheduled call and raises a question; the AI answers within what it can handle and keeps the record. When the store calls the support line later, the AI sorts the request, handling it if it can and passing it to a person if it cannot. Afterwards the manager can go back through the record and the recording to check how it was handled. This is a walkthrough of the confirmed capability, not a transcript of a client call.
What it amounts to
Routine contact, phone questions and human handoff joined into one thread, so a store's progress keeps being carried forward instead of explained again from the start.
What it changes
Covering the routine, so the difficult calls get the attention
What is confirmed here is the capability described above. Conversion, lower cost, higher satisfaction and return on investment have no measured figures behind them, so this case is written as mechanism and effect rather than as a result.
Follow-up keeps its continuity
Someone always knows where each store is, which is what makes the next conversation possible.
Human effort goes further
Routine contact and part of the question load are carried elsewhere, leaving people for the complicated end.
A store has somewhere to go
Outbound contact and inbound answering run side by side, and the route to a person stays open.
Review has something to stand on
Records and recordings locate the actual problem, instead of relying on recollection.
How you would tell whether it is worth keeping
- Follow-up coverage
- Of the stores due a call, how many had a conversation that went somewhere.
- Problem handling
- Whether the problems it should resolve stay resolved, and whether the ones passed to people are actually picked up.
- Team time
- How the total hours spent on contact, handover and review move at a similar volume of work.
Judge it on what came back: who was reached, what was resolved, and who took it from there.
Product reference Β· redacted
Scheduling follow-up for hours it can actually happen
From needs-a-call to called-on-plan
What this means for the work
In work like this, a screen of this kind is how the calling rhythm gets organised. The list of stores has to move with the work, contact times have to suit when a store is actually reachable, and whether a call was made on time should be recorded separately from whether it was worth anything.
Every call leaves something behind
From a contact record to evidence of service
What this means for the work
When a store asks the same thing twice, or a previous answer needs checking, the manager can go back to that specific call and see how the problem arose and how it was handled. That is a different starting point from reconstructing it from memory, and it is what any later coordination gets built on.
Turning conversations into something you can review
From listening to calls to knowing what to change
What this means for the work
The reference product can sort an agreed set of conversation outcomes into categories. Applied to store work, the ones worth watching would be opening stage and outstanding items, with spot checks on whether the sorting is right. Those categories are a suggestion for this kind of work, not a capability confirmed as live in this project.
Test it on what came back rather than on what it can do: was the answer right, was the problem dealt with, and did somebody pick it up from there?
Widening it
Decide the edges first, then widen
From a business boundary to something that runs
Pick one job
Choose one kind of store and one clear follow-up goal.
Agree the rules
Settle what can be answered, how problems are graded, and who owns the handoff.
Check the calls
Sample the calls and verify the answers, the sorting and the outcome.
Widen on evidence
Extend it on experience, hours saved and cost per conversation.
This is our advice for similar work. It is not a timeline from the project described above.
Buy the outcome, not the ability to make calls
- Which stores get called, and when
- What can be answered, and the rules for handing over
- Call records, recordings and the review routine
- How the result will be judged
Assess it on service delivered rather than on dialling. Beyond answering and calling, check whether problems actually get resolved, whether the handoff holds, and how much ongoing attention it takes to keep running. If you want to talk about sales value, that has to be judged alongside how each store is actually placed and what it goes on to do.
Basis: written from the project's own account of the work. Client name, region, organisation and store identities are withheld. Interface images are redacted reference screens, not this client's system.
Is your team making these calls?
If follow-up keeps slipping, or the support line is where your best people spend their day, the self-check is the quickest way to see which part is worth doing first.