AI Spots the Deal Risk. Acting on It Still Takes Days.

Everstage published a survey last week of 195 VP-and-above revenue leaders at North American B2B SaaS companies, and the finding that stuck with me wasn't about AI at all, really. 45% of them said it takes four or more days to act after a deal-risk signal appears. And 57% said they first learn that a deal is at risk during a pipeline review, compared with 17% who first hear about it from an automated alert.

Every leader in that survey reported having AI in production, so this isn't a group that's behind on tools. And yet 51% said AI's biggest shortcoming is that it surfaces an insight without taking action on it. One in four said they learn a deal was at risk only after it has already been lost.

A quick, fair caveat: Everstage sells sales performance software, and the survey's framing leans toward "AI should act, not just inform," which happens to be the direction vendors in that space are selling. I'd take the diagnosis seriously and hold the prescription a little more loosely. Because where I've landed, after a lot of time in revenue enablement, is that the four days aren't an AI problem. They're what happens when a signal shows up and nobody's job description says what to do with it.

The Tool Isn't the Slow Part

Think about what actually happens when a risk signal fires in most revenue orgs. The champion goes quiet, or a competitor shows up on a call, or the economic buyer hasn't been in a meeting in three weeks. Some tool notices and puts a flag on the opportunity, or a line in a digest email, or a red dot on a dashboard. And then the flag sits there, because the rep assumes their manager saw it, the manager assumes the rep is on it, and RevOps assumes it'll come up in pipeline review on Thursday. So it comes up on Thursday.

That's why 57% of leaders hear about risk in pipeline review, and I don't think it's because the signal arrived late. Pipeline review is just the only place in the week where someone is clearly expected to look at a deal and decide something. The meeting has an owner, a time and a format, and the alert doesn't. So the meeting wins, every time, and the alert becomes a very expensive way of generating agenda items.

I think this is the part that gets lost in a lot of AI buying conversations. A signal is only worth something if it shortens the time between "something changed" and "someone did something about it." If the time to action is the same with the tool as without it, you bought a nicer way to find out on Thursday.

A Signal Needs an Owner, a Play and a Clock

The fix I keep coming back to is pretty unglamorous. For each signal that actually matters to your forecast, you write down three things, and you write them down before you buy or turn on anything else:

• An owner, meaning one named role (not "the deal team") who is expected to act when this signal fires. For a single-threaded deal that might be the AE; for a stalled renewal it might be the CSM with the account manager copied.
• A play, which is what acting actually looks like, in a sentence or two. "Re-engage the champion" is a wish. "Ask the champion for a 20-minute call with the economic buyer, and offer the mutual action plan as the reason" is a play someone can run on a Tuesday afternoon.
• A clock, meaning how fast the owner is expected to move (24 hours, 48 hours, whatever fits your cycle), and who notices when it doesn't happen. The clock is what turns an alert into an expectation.

None of this needs AI, and that's sort of the point. If you can't fill in those three things for a given signal, no tool can fill them in for you either. And the companies that do fill them in tend to find that their existing tools suddenly look a lot more useful, because there's finally a person on the other end of the alert who knows what to do with it.

This is also where enablement either earns its keep or doesn't. A play nobody has practiced doesn't get run under pressure. Having led AI GTM and revenue enablement at Commerce, the way I think about it is that rolling out the tool is the easy part, and making sure the people receiving the output know exactly what's expected of them when it lands is most of the work (and, admittedly, the boring part).

Why an Agent That "Acts" Won't Fix This Yet

The natural next pitch, and you'll hear it a lot this quarter, is an agent that closes the gap for you: it spots the risk, drafts the email, books the meeting, updates the forecast. Some of that will be genuinely useful eventually, and I don't think anyone can say yet exactly how far it goes. But if today nobody knows who owns a single-threaded deal or what the play is, an agent that acts will mostly automate the confusion faster. It'll send the email nobody agreed on, to a contact nobody vetted, on a deal the manager had already written off.

Salesloft's 2026 U.S. Revenue Benchmark Report, a survey of 500 sales and revenue decision-makers, found that every organization surveyed uses AI somewhere in its revenue process, but only 20.6% describe their AI strategy as production-ready with measurable outcomes, and 28.2% are still experimenting. (Salesloft is a vendor too, so the same grain of salt applies.) I'd read that gap the same way I read the Everstage numbers. Adoption isn't the constraint anymore. The constraint is whether the operating rhythm underneath can turn what the tools surface into something measurable.

So I'm not anti-agent here. I'd just sequence it. Get the owner, play and clock working by hand for a handful of signals, measure how fast you move, and then let automation take over the steps you've already proven. At that point you're automating a decision you understand, which is a very different bet from automating one you don't.

Where to Start This Quarter

For most revenue teams I'd keep this small and specific. You don't need a transformation program, you need a few weeks of honest measurement.

• Pick the three to five risk signals that most often show up in deals you end up losing. Your own closed-lost notes are a better guide here than any vendor's list.
• For each one, write down the owner, the play and the clock. If the team argues about who owns it, good, that argument was going to happen eventually and it's better to have it now than in the forecast call.
• Measure time to action for each signal for a month: when the signal appeared, and when the owner actually did something. You'll likely find the number is uncomfortable, and that's the baseline.
• Change what pipeline review is for. If a risk was handled within the clock, it doesn't need airtime. Use the meeting for the deals where the play didn't work, so it stops being the place you first find out.
• Only then look at where automation or an agent would shorten a step that's already working.

That last one matters for the budget conversation too. If your board or CFO is asking what the AI spend is buying, "our median time from risk signal to action went from four days to one" is a sentence a board understands. "We rolled out deal intelligence to the whole team" isn't.

Write Down Who Acts Today

If I had to boil it down: before you buy an agent that acts, write down who acts today and how fast. If you can do that cleanly for your top signals, you're in a great spot to put AI on top of it and actually see the result. If you can't, that's the project, and it's a much cheaper project than the one you were about to fund.

If you want a second set of eyes on where the time is leaking between signal and action in your org, I'm happy to take a look. Grab a 30-minute working session here.