AI sales coaching: which scores to trust, and what still needs you

You manage a handful of reps, and on a good week you listen to maybe two of their calls each. AI coaching tools are built to cover the rest. Strip the marketing away and they do four jobs:

1. They analyse calls: record, transcribe and score each one against a checklist you set, like "asked about budget" or "agreed a next step".
2. They give feedback on how the rep talks, such as how long the rep talked compared with the buyer, how many questions they asked and how fast they spoke.
3. They run roleplay, where the rep practises a cold call or a price objection with an AI playing the buyer and gets a score and notes at the end.
4. They suggest next steps after a real call, by drafting the follow-up or naming what the rep should do before the next meeting.

Here's how a few tools describe themselves on their own sites:

• Gong Enable has an AI Call Reviewer that scores calls automatically, an AI Trainer that turns past calls into roleplay scenarios, and a coaching inbox for managers.
• Hyperbound runs roleplays with AI buyers, records and scores real calls against a rubric, and shows each rep's gaps rather than a team average.
• Second Nature runs AI roleplays with avatars for discovery calls, objection handling and cold calls.
• Yoodli gives private roleplay practice with speaking analytics such as pacing and conciseness, and tracks progress over time.

Check prices and plans on each tool's own site. The AI sales tools post covers tools for other jobs.

AI coaching output falls into three groups. Treat each one differently.

Counts are the part to trust most, since a transcript either has a question in it or it doesn't. Judgments are where the tools slip, and a tool can mark a rep down for not asking about budget on a call where the buyer said the budget in the first minute.

The tools improve every quarter, but there are still four things they leave to you.

• Picking the one thing. A scorecard can list eight weak spots. A rep can work on one. Choosing which one, for this rep this month, is your job.
• Knowing the context. The AI doesn't know the rep's mum is in hospital, that the territory changed, or that the buyer on Tuesday's call was always going to say no.
• Making it stick. Feedback from a tool is easy to ignore. A manager who asks "how did the new opener go this week?" is harder to ignore.
• Deal judgment. Whether to push on price, walk away or bring in help is still a call a person makes.

AI coaching fails when it's a login reps open once a month, and it works when it has a fixed place in the week, on the same days and in the same shape. Here's a routine that fits a team of 5 to 15 reps.

1. Monday, 10 minutes: every rep does one roleplay on this month's skill, like handling "we're already using someone".
2. Midweek, 15 minutes each: you listen to one real call per rep, picked from the tool's scores. Pick the call that scored oddly, high or low.
3. Friday, 20 minutes: a short 1:1 with each rep. Look at the roleplay, the real call and one number. Agree one thing to try next week. The 1 on 1 meeting template has an agenda to start from.
4. Month end: check whether the number moved. If it did, pick the next skill. If it didn't, keep this one for another month.

One skill a month per rep. You don't need twelve training topics. You need a short list of the things that lose your team deals, worked through in order.

Some reps will hear "AI records and scores every call" as "my manager is building a file on me". That's a fair worry, and ignoring it kills adoption.

• Say why before you switch it on. The tool is there to help them practise and to save you guessing, and it isn't a ranking.
• Let reps see their own scores first, before you do. Some tools let them practise privately.
• Coach from the tool and never discipline from it. If a score shows up in a performance review with no conversation before it, trust is gone.
• Use it on your own calls too, if you still sell. Reps notice.
• Tell buyers you're recording and follow your country's rules on it.

The sales accountability guide covers how to hold reps to their numbers without it feeling like surveillance.

1. Pick one skill the whole team is weak at. Look at last month's lost deals if you're not sure.
2. Trial one tool on that skill only, with two or three reps who volunteer.
3. Count one number before and after, like first calls that end with a booked meeting.
4. Keep your 1:1s exactly as they are. Add the tool's output to them, and don't replace them with it.
5. After two weeks, ask the volunteers whether they'd keep it. If they wouldn't, don't roll it out to the rest.

• Gong, Gong Enable: AI Call Reviewer scores calls automatically, AI Trainer turns past calls into roleplay scenarios, coaching inbox for managers, read 28 September 2026
• Hyperbound, homepage: roleplay with AI buyers, recording and scoring real calls against a rubric, gaps per rep, read 28 September 2026
• Second Nature, homepage: AI roleplays with avatars for discovery calls, objection handling and cold calls, read 28 September 2026
• Yoodli, homepage: private roleplay coaching, speaking analytics such as pacing and conciseness, progress over time, read 28 September 2026