A few years ago I spent an embarrassing amount of money on a carbon road bike. Featherlight frame, aero wheels, the sort of thing that arrives with a price tag you quietly hide from your wife. I was convinced it would make me faster.

It did, technically. It made me faster at being exactly as unfit as I'd always been.

The bike was an amplifier. It took the engine I already had and multiplied it. Trouble was, my engine was a bloke who hadn't trained properly since university. A lighter frame doesn't build a bigger heart. It just gets your existing lungs to the bottom of the hill a little quicker, so you can be out of breath somewhere new.

Watch what's happening with AI in sales right now and it's the same mistake, running at industry scale.

Every company I speak to is buying the carbon bike. AI SDRs, copilots, generative-everything, a shiny new tab in the CRM that promises to write the email, draft the call plan, summarise the deal. The pitch is always some version of "bolt this on and the number goes up." Then six months later the number hasn't moved, and everyone's quietly confused about why the expensive bike didn't win the race.

Here's the bit nobody wants to say out loud: AI on its own is an amplifier, not a coach. It multiplies whatever capability is already in the room. Give it to a rep who's already good and it makes them faster. Give it to a rep who's stuck and it helps them be stuck at a much higher volume.

I don't just believe this, there's now proper research behind it. ValueSelling Associates, working with Aberdeen Strategy & Research, surveyed 610 sales and enablement leaders and split them into two camps: companies using AI alongside sales coaching, and companies using AI on its own.

Companies that paired AI with sales coaching saw 3.3x the year-on-year growth in quota attainment of companies using AI without it. The same group cut sales-cycle time by 56% and lifted sales productivity by 95% year on year.

Read that again. Same technology. Same models. The only variable is whether there's a coaching layer wrapped around the AI. 3.3x is not a rounding error. It's the difference between AI being a line item you defend at renewal and AI being the reason you hit the year.

(Full disclosure: ValueSelling is one of the methodologies we build coaching around, so I'm hardly a neutral observer here. But it's their data, their independent research partner, and 610 leaders is a sample I'm happy to stand behind.)

Does AI improve sales performance on its own?

Barely, and not in the way the brochure promises. AI on its own improves activity, not capability. It writes more emails, drafts more call plans, summarises more calls. What it doesn't do, unprompted, is make the rep better at the one conversation that actually decides the deal.

Think about what "faster" buys you when the underlying skill hasn't moved. A rep who never confirms the cost of doing nothing will now fail to confirm it in half the time, across twice as many accounts. You haven't fixed the leak. You've just pumped more water through the same broken pipe.

Why does AI alone fail to move quota?

Because it does nothing about Performance Drift: the gap between what a rep learned in training and what they actually do on a live call. That gap is a behaviour problem, and behaviour doesn't change just because a tool got quicker.

We learned this the hard way. When we started, we ran standalone AI coaching and watched fewer than 5% of reps genuinely improve. The technology was fine. The reps weren't lazy. The missing piece was the loop: something that watches what happens on the real deal, tells the rep the one thing they missed, and holds them to it next time. Without that loop, AI just industrialises the habits reps already have. You end up with better-educated underperformers, the AI edition.

Sound familiar? It should. It's the same reason a decade of LMS completions and two-day workshops never moved the needle. Knowledge was never the bottleneck. Execution was.

What does "AI with coaching" actually mean?

It means three things the generic tools skip: methodology, memory, and a loop.

Methodology. The AI has to know how your team actually sells. Not "sales" in the abstract, but Gap Selling, or MEDDPICC, or ValueSelling, or whatever your reps are meant to be running, enforced inside the product rather than printed on a poster nobody reads.

Memory. It has to know this rep and this deal. What got promised on the last call, which stakeholder went quiet, where the discovery was thin. Coaching that starts from zero every session isn't coaching, it's a very expensive search bar.

A loop. It has to close the gap between where the work happens and where the feedback happens. We call that closing the Context Gap. The coaching has to land on the deal in front of the rep this afternoon, not in a classroom next quarter.

That's the whole idea behind how we build: Show Me, Let Me, Coach Me. On day one the AI does the execution work at full fidelity, so the rep gets an immediate result on a real deal. Then it hands more of the wheel over as the rep gets stronger, until it's coaching rather than doing. Immediate performance today, capability that compounds tomorrow. The amplifier and the training programme, in the same tool.

Should you stop buying AI sales tools?

No. Buy the fast bike. Just stop pretending the bike is the fitness.

The mistake was never investing in AI. The mistake is buying AI as if the software is the transformation, when the software is only ever the amplifier. The 3.3x doesn't come from the model. It comes from wrapping the model in a coaching loop that knows your methodology and your deals. Strip that away and you've bought a very clever way to do the wrong things quicker.

What actually makes AI move the sales number?

Depth, not horsepower. An AI that knows your methodology, remembers your rep, sits on the actual deal, and closes the loop between the call and the feedback will move quota in a way a faster email generator never will, because it's working on capability, not just output.

We built Replicate Labs around that single idea. Not a copilot that makes a mediocre call quicker, but an AI management team, a coach, a performance manager, a RevOps manager, that turns every real deal into a coaching moment and every coaching moment into a better next call. The technology is the easy part. The loop is the product.

AI will happily make your reps faster. Whether it makes them better is entirely down to what you wrap around it.

If your team has already bought the AI and the number still hasn't budged, the software was never the missing piece. The coaching loop was. You can put one around a real deal today: start free with Keenan at https://replicatelabs.ai, bring a deal that's actually stalling, and watch what methodology-native coaching flags on your real pipeline rather than a demo script. It's free to get started, whether you're a rep trying to close the one in front of you, a manager trying to lift the whole team, or a business tired of paying for AI that made everyone faster and nobody better.