A few weeks ago I watched a sales manager walk through his pipeline on a screen share. Every deal had a little coloured badge next to it. Green, amber, red. The red ones had a tidy explanation underneath: engagement declining, no activity in fourteen days, single-threaded, timeline unrealistic. It was genuinely impressive. The model had read the room better than half the reps owning the deals.
Then he scrolled, sighed, and said the quiet part out loud. "So I know which seven deals are dying. Now what do I actually do about them by Friday?"
That sentence is the whole problem with where revenue tooling has landed. We have built an extraordinary detection machine and bolted it onto a team that still has no more time, no more skill, and no more help than it had before the machine arrived. The flag lights up. Nobody is any closer to saving the deal.
The whole category has quietly admitted detection is not enough
Here is the thing I find genuinely funny about deal-risk scoring in 2026. Read the marketing for almost any of these tools and you will find, somewhere past the hero section, a paragraph confessing that detection on its own does nothing. "Knowing a deal is at risk doesn't save it." "Detection is necessary but insufficient." "The gap isn't the intelligence, it's the execution." They are right. They are describing the exact moment their own product hands the problem back to you.
And the data backs the confession. Only around 7% of sales organisations hit the forecast accuracy that risk scoring is supposed to unlock. Not because the scores are wrong. Because a score sitting in a dashboard is not an action. It is a more articulate way of worrying.
So we have arrived somewhere slightly absurd. We spent a decade getting computers good enough to tell a rep their deal is in trouble, and the rep's response is the same as it was in 2015: stare at it, feel a bit sick, and try to remember what worked the last time something like this happened.
A risk score is not a fix. It is a better-informed version of the panic you already had. The intelligence layer got ten times smarter. The execution layer, where deals are actually saved, got nothing.
Why reps ignore the red badges (and they do)
If you have rolled out any of these tools, you already know the dirty secret: the scores get ignored. Not out of stubbornness. Out of self-defence.
Picture the rep again. They open Monday's pipeline and eleven deals are amber, four are red. Every single flag is technically accurate. And every single flag arrives with no answer to the only question the rep has, which is "and so what do I do, specifically, today, with this deal, given this buyer and this politics and this silence." The tool has handed them fifteen problems and zero next moves. That is not help. That is a louder to-do list.
So the rep does what any sane person buried in alerts does. They learn to scroll past the colour. Alert fatigue sets in within a fortnight, the scores stop changing behaviour, and win rates do not budge. The org concludes the reps "aren't using the insights." The reps concluded, correctly, that the insights were not actually insights. They were observations. An observation is what a buyer is doing. An insight is what you should do about it. Most of the market is shipping the first and charging for the second.
Detection, recommendation, execution: only one of them closes deals
It helps to be precise about the three layers, because most tools collapse them and the collapse is where the value leaks out.
Detection is "this deal is at risk." A model reads activity, sentiment, timeline, and threading and lights up the deals that look wrong. This is genuinely good now, and it is also the cheapest layer to build. Everyone has it.
Recommendation is "here is roughly what tends to work in deals that look like this." A step up. The smarter tools have reached here, suggesting a play based on historical patterns. Useful, but it is still a leaflet slid under the door. It assumes the rep can take a generic recommendation and turn it into the right words to the right stakeholder in a deal with a hundred specific details the leaflet never saw.
Execution is the rep actually doing the work to de-risk the deal, well, on this specific opportunity. Re-engaging the buyer who went dark with a message that fits what was actually said on the last call. Building a second thread into an account that is dangerously single-threaded. Reframing a slipping timeline around the buyer's real driver instead of your quarter-end. This is the only layer where a deal moves. And it is the only layer almost nobody is helping with.
The reason the gap exists is that detection scales with compute and execution scales with coaching, and coaching has always been the bottleneck. There was never enough manager time to sit with every rep on every red deal and work it through. So the industry did the convenient thing. It automated the part that was easy to automate, called it transformation, and left the hard, human, deal-saving part exactly where it has always been: starved.
What closing the gap actually looks like
The fix is not a better score. It is moving help to where the deal is decided.
Imagine the red badge does not just sit there. Imagine the rep can open that exact deal and, instead of a generic playbook, work it with a coach that already knows the context. It knows who the buyer is, where it went quiet, who has to sign and who has gone missing, because it has been in the deal with them. The rep does not get told "re-engage the champion." They get the question that surfaces why the champion went cold, then they draft the actual re-engagement in the rep's own voice, against the real history, in the minutes they have before the next call. The flag becomes a starting point for work, not a verdict to dread.
That is the difference between a tool that watches your pipeline and a tool that works it with you. One produces a more accurate forecast of which deals you will lose. The other changes which deals you lose. Sit with that distinction, because the entire market is selling you the first while implying the second.
This is the layer we built Replicate Labs to own. Not the dashboard that tells a manager seven deals are dying. The coach that sits with the rep and helps them save three of them. Detection is a commodity now. Execution on the actual deal, in the rep's actual workflow, is where the result lives, and it is the part the rest of the category keeps confessing it cannot do.
If your reps are drowning in accurate flags and nothing is moving, the missing piece is not a smarter score. Bring a real deal, one that is genuinely slipping, and work it. Sixty minutes with Keenan is free, no card, no demo gauntlet. See what happens when the red badge turns into a next move you can actually make. Start at replicatelabs.ai.
You do not have a forecasting problem. You have an execution problem wearing a forecasting problem's clothes.
James Pursey is the CEO of Replicate Labs, an AI sales coaching platform built on named sales methodologies. Previously enablement at SimilarWeb.