AI Sales Coaching Only Works If Someone Actually Acts On It

A sales manager in Jeddah once showed me his dashboard. Fourteen charts. Colour-coded. Beautiful, honestly.

Then I asked him what he did with it last Monday. He paused. "I looked at it," he said. That was the whole answer.

And that's the problem with most so-called AI in field sales. It tells you things. Loads of things. But telling isn't coaching. A rep dropping 22% on his weekly outlet coverage doesn't need a chart confirming he dropped 22%. He needs someone (or something) to tell him what to do about it before Thursday.

I got this wrong at first with zivni. We built a really pretty analytics layer early on. Sales manager insights everywhere. Trends, heatmaps, all of it. And our best customers barely opened it. Took me a while to understand why. The data was accurate. It just wasn't useful in the moment when a manager had eleven reps and forty minutes.

The gap between knowing and doing is where deals die

Here's the thing about field rep performance. It doesn't improve because a report exists. It improves when a specific person changes one specific behaviour on a specific route.

So we flipped the whole thing. Instead of "here's what happened," we started building toward "here's what to say to Ahmed tomorrow."

Big difference. And harder to build than it sounds.

Let me give you a real example. A distributor we work with in Karachi — mid-size, roughly 180 reps across FMCG snacks and beverages. Their managers were spending most of their coaching time on the loudest problems. The rep who missed target by a mile. The one who quit calling. Reactive stuff.

Meanwhile the quiet middle — the reps doing okay but leaving money on the table — got ignored. Nobody had time to dig into 180 people.

That middle is where the growth actually sits. A rep hitting 80% could probably hit 95% with two small tweaks. But no manager on earth is manually spotting those two tweaks across 180 people every week.

That's the job AI is genuinely good at. Not replacing the manager. Doing the noticing.

What a good AI recommendation actually looks like

Most tools give you this: "Rep coverage down 15%." Okay. And?

What we push for in zivni is closer to this: "Bilal visited 6 of his 9 high-value outlets in Deira this week. The three he skipped bought 4,200 AED last month combined. Two are open before 10am — suggest he front-loads those before traffic."

See the difference? One's an observation. The other's a plan a manager can hand over in thirty seconds.

The AI sales coaching that works isn't the flashy kind. It's boring, specific, and tied to money. It answers three questions a manager would otherwise have to work out by hand:

• Who needs attention this week (not everyone, just the few where a nudge moves numbers)
• What exactly to talk to them about
• What good looks like so the conversation isn't vague

We pull this from stuff reps are already generating — GPS-tracked visits, order patterns, shelf photos, time-per-outlet. Nobody fills in extra forms. That matters more than people think. The second you ask reps to do admin to feed the AI, they stop, and your data rots.

One merchandising manager in the UK told me something that stuck. She said the AI didn't make her smarter. It made her faster. She already knew coaching mattered. She just never had the hours to prep for eleven separate conversations. Now she walks into each one with the exact thing to focus on. Her one-on-ones went from vague pep talks to actual coaching in about three weeks.

That's the whole point. Speed to a useful conversation.

Where this quietly goes wrong

But let me be honest about the failure modes, because there are some.

First — if managers don't trust the recommendation, they ignore it. Trust comes from the AI being right about the small stuff first. If it flags a rep as underperforming when everyone knows that rep just had two weeks of medical leave, the whole system loses credibility. We learned to build in context. Bad recommendations are worse than no recommendations.

Second — over-coaching. Some managers took the AI as a to-do list and started drowning reps in feedback. Every week, five things to fix. Reps hated it. Felt like surveillance, not support. Look, coaching works when it's focused. One or two things at a time. We actually capped how many recommendations surface per rep per week, on purpose. Restraint is a feature.

Third — and this is the big one — none of it works without the human. I don't believe in fully automated coaching. A message from an app doesn't build a relationship. A manager who says "hey, I noticed you cracked those three Deira outlets, nice work" does. The AI teed it up. The human landed it. That split is non-negotiable for me.

We've had customers ask if they can just auto-send recommendations straight to reps and skip the manager. I always push back. You'd be automating away the exact thing that makes field sales human. Reps don't stay for dashboards. They stay for managers who see them.

The numbers back the human bit up too. Across the accounts where managers actually used the recommendations in weekly one-on-ones — versus the ones who just let the app notify reps directly — we saw roughly 3x better follow-through on the flagged actions. Same AI. Different delivery. The manager is the multiplier.

So if you're evaluating any of this — us, or FieldAssist, or whoever — don't get sold on the insight generation. Everyone can generate insights now. That's commodity. Ask them one question instead. What does the manager actually do differently on Monday morning because of your tool?

If the answer is "they look at a nicer dashboard," keep walking.

The real test isn't how smart the AI sounds. It's whether the rep in Muscat or Manchester or Multan does one thing better next week because of it. Everything else is decoration.

And honestly, most of the industry is still selling decoration.