I recently onboarded two new callers to our sales floor. The costly part of their training was not the app or the script. It was the leads they would burn while they learned.
So I gave them a cheaper place to make their first mistakes: a voice AI that plays the prospect.
Why is a new caller’s first week so expensive?
A lead is a limited chance. On our floor at Canadian Funding Partners, which I co-founded and built AI-first, every caller works from a list in a power dialer. When a call ends, the caller taps what happened (voicemail, interested, not interested) and the dialer moves to the next number after a short pause. It is built for pace.
Pace is good once a caller knows what to say. Before that, it means the dialer feeds real prospects to someone still finding their words.
Some of the leads we load are older ones. Other callers tried them before and could not connect. Those are already hard. A new caller who fumbles one of them may not get another chance.
The prospects add to the pressure. We call small businesses about a government funding program. Many of them are wary of cold calls that mention government money. The first thing in their head is whether it is a scam. A caller who stumbles on the opening loses the call in seconds.
None of this means a new caller should wait. I told both of them to make mistakes, have lots of phone numbers you can call back, and get comfortable. But the first few mistakes do not need to happen on real leads.
How does AI role-play work for sales call coaching?
On the onboarding call, I gave the new callers three ways to learn their market before they lean on the dialer.
- Talk to people in the niche. If you know a few accountants, have longer conversations with them. Ask how their practice works and what wears them down.
- Book time with a few more and ask the same questions. The more you talk to people in that niche, the better you can speak to them on the phone.
- Use AI. In my words on the call: “You can voice call AI, ask the AI to roleplay an accountant. I can even build an AI that you can roleplay with.”
The third one is the cheapest by far. The first two take time to arrange. The AI prospect is ready whenever the caller is, and it will raise the same objection again and again without getting tired.
It is a voice call, not a chat. That matters. Cold calling is a spoken skill. Typing a good answer to “Is this a scam?” is not the same as saying it out loud with a stranger waiting.
We already run an AI voice agent on our floor alongside our human callers. Turning that same kind of tool around so it plays the prospect is a small step.
What should the AI prospect brief include?
The quality of the role-play depends on the brief. A vague brief gets you a polite prospect who agrees with everything. That teaches nothing.
What I wanted the callers to understand before they dial is the industry: the typical problems, the typical priorities, and what the owner believes. The thing I don’t know about a new niche is what they believe. One owner may think AI will never touch their business. Another may think less admin work would give them more money or more free time. Those two need very different calls.
So the brief should carry that. Here is the shape of one, for illustration only:
- Who you are: owner of a small accounting firm with a few staff.
- Your week: document chasing before tax season, routine client emails, phone inquiries that pull you away from billable work.
- What you believe: this call is probably a scam; government programs mean paperwork; AI is for big firms.
- Objections to raise: “Where are you based?”, “Who have you worked with?”, “What does this cost me?”, “I don’t have time.”
- What would make you book: the caller shows they understand your tax-season crunch and keeps the ask small.
- Rule: stay skeptical until the caller earns it. Hang up if they read a script at you.
The caller then practises the opening, the legitimacy questions and the ask until the answers come out in their own words.
How do you know the practice is working?
This is where the claim has to earn its place.
An AI system is only as good as the feedback it gets. The same goes for a new caller. Role-play gives fast feedback on the parts of a call you can rehearse. It cannot tell you whether the real market responds.
That is why I did not want the callers to spend weeks in practice. My advice on the call was to get your feet wet first. Make sure the system works, start calling, and when you hit specific challenges, we go deeper on those.
That sets up a loop:
→ practise the opening and common objections with the AI prospect → make real calls and listen back to the recordings → note the objections that went badly → add those to the AI prospect’s brief → practise again
The number to watch is meetings booked. I told the callers that is the number they control.
What would prove this wrong: if callers who practise with the AI prospect book meetings at no better rate than callers who skip it, over their first few weeks, then the practice is not paying for itself. We have not measured that yet. Treat the benefit as a claim until we do.
Where does AI coaching stop and real calls start?
AI role-play is a rehearsal room. It is not the market.
The AI prospect only knows what the brief tells it. If the brief is wrong about what owners believe, the caller rehearses the wrong call. Real conversations, with people who run these businesses, are what keep the brief honest. That is why the first two learning methods stay on the list.
It also only matters if calling is where your business is stuck. Our working assumption is that in owner-led firms, the step that limits output is most often revenue follow-up: how fast you reach people, how well you follow up, how clean the records are. If that is true for you, coaching callers is AI on the bottleneck, and that is the kind of AI that raises what the whole business produces.
If your bottleneck is somewhere else, say delivery or fulfilment, then better callers will not lift output. They will pile more work in front of the real constraint. At best you trim the cost of training. Be clear about which case you are in before you build anything.
What to watch: whether new callers reach steady booking faster than earlier hires did, and whether fewer of the hard, recycled leads get burned in their first week.