On May 20, Gartner published survey findings that should reframe how revenue leaders think about their AI spend. The survey, fielded between August and September 2025 across 227 chief sales officers and presented at the Gartner CSO & Sales Leader Conference, found that sales organizations providing sellers with AI-enabled next-best actions are 2.6x more likely to achieve commercial growth. Organizations that prioritize AI upskilling for their sellers are 2.4x more likely to see strong revenue growth.
Here's the uncomfortable part. Almost every sales team already believes it has next-best-action AI. Salesforce's State of Sales 2026 research, published in February, found that 87% of sales organizations use some form of AI. If 87% of teams have AI and only a fraction see a growth advantage, the difference isn't whether you bought the technology. It's whether the technology you bought actually changes what sellers do next.
"Next best action" has become the most abused phrase in sales tech. Every vendor claims it. Most implementations amount to a dashboard widget that suggests "follow up with this account," which reps glance at once, distrust, and never open again. The Gartner finding is not about that widget. It's about a much harder thing: recommendations grounded in complete and current data, explainable enough that reps trust them, and one click away from execution. Miss any of those three properties and the 2.6x evaporates.
What the 2.6x teams have that the others don't
Strip away the vendor language and a next-best-action system does one job: it looks at everything known about a deal and tells the seller the single highest-leverage thing to do right now. That job has three dependencies, and each one is a place where most implementations quietly fail.
First, the recommendation is only as good as the data underneath it. Second, the recommendation only matters if the rep believes it. Third, the recommendation only produces revenue if it gets executed. Data, trust, execution. Most sales AI gets one of the three right, calls it done, and then the CRO wonders why the tool with "next best action" on the pricing page didn't move the number.
Gartner's related projection makes the stakes clearer. The firm projects that by 2027, 95% of sellers' research workflows will begin with AI, up from under 20% in 2024. Sellers will start their day by asking a system what to do. The teams that win are the ones whose system gives a correct, trusted, executable answer. Let's walk through the three ways that answer goes wrong.
Failure mode one: stale data produces wrong recommendations
A recommendation engine that reads from a nightly sync is a recommendation engine that reasons about yesterday. If the champion emailed this morning to say the budget got cut, and the engine's copy of the deal still shows "Negotiation, 80% confidence," the "next best action" it produces is not just useless, it's actively harmful. It tells the rep to send a pricing follow-up into a deal that just changed shape.
This is the quiet killer in most next-best-action deployments. The scoring model might be excellent. The recommendation logic might be sound. But the model runs on a replica of the pipeline that lags reality by hours because the data has to travel from the CRM through an ETL job into a warehouse or intelligence layer before the AI ever sees it. We've written before about why AI-ready data is the new competitive moat: AI output quality is bounded by data completeness and freshness, and no amount of model sophistication compensates for reasoning over a stale snapshot.
Ask your vendor one question: when a rep finishes a call, how long until that call's content can change a recommendation? If the answer involves a sync schedule, an overnight batch, or "typically within a few hours," your next-best-action engine is a yesterday's-best-action engine. The deals that need intervention most are precisely the ones changing fastest.
Staleness also compounds. A wrong recommendation on Monday erodes trust, which means the rep ignores a correct recommendation on Thursday. One bad suggestion built on old data costs you more than the deal it touched.
Failure mode two: black-box scoring produces ignored recommendations
The second failure mode has nothing to do with the data and everything to do with the rep. A recommendation that arrives as "Priority: High. Action: Re-engage account" with no reasoning attached is asking the rep to outsource their judgment to a system that won't show its work. Experienced sellers won't do it, and honestly, they shouldn't. Their skepticism is a feature.
Compare two versions of the same recommendation. Version one: "Follow up with Meridian Health." Version two: "Follow up with Meridian Health: the proposal was viewed 3 times yesterday by two new stakeholders, but no meeting is scheduled and the last call flagged a competing vendor. Suggest a pricing-objection call before Friday." The first is noise. The second is a colleague who did the homework. Reps act on the second version because they can verify the reasoning against what they know about the deal.
This is the same trust problem we covered in our analysis of explainable AI in sales forecasting. A forecast number without visible drivers gets overridden by gut feel. A recommendation without visible evidence gets ignored the same way. In both cases the fix is identical: every AI output must cite the specific signals that produced it, in language a rep can check in ten seconds.
Black-box recommendations follow a predictable arc. Week one: reps try a few suggestions. Week three: one suggestion is visibly wrong and there's no way to see why. Week six: reps stop opening the panel. Quarter two: the CRO reports low AI adoption, buys an enablement program, and the cycle restarts with the same opaque engine underneath. Gartner's 2.4x finding on AI upskilling matters here, but training reps to trust a system that won't explain itself is training them to ignore their own judgment. Explainability has to come first.
Failure mode three: recommendations without an execution path are decoration
The third failure mode is the most common and the least discussed. The recommendation is correct. The rep believes it. And then executing it requires leaving the tool that made it: open the email client, find the thread, check the calendar tool for availability, update the CRM stage in another tab, log the activity manually. Five context switches to act on one suggestion. By the third switch, half the reps have been pulled into something else.
This is where the read-only architecture of most revenue intelligence tools breaks down. They sit beside the systems of record, observing and suggesting, structurally unable to act. We made the longer version of this argument in From Revenue Intelligence to Revenue Action: insight that terminates in a dashboard is a cost center, and the industry has spent a decade perfecting the art of describing problems it cannot fix. A recommendation panel bolted onto that architecture is dashboard decoration with better copywriting. It joins the graveyard of widgets we described in The Death of the CRM Dashboard.
The teams in Gartner's 2.6x cohort closed this gap. The recommendation and the action live in the same system. "Send the pricing follow-up" is not a suggestion to go do work somewhere else. It's a button. The email is drafted from the deal context, the send is logged automatically, the sequence adjusts, and the deal record updates, all in the same motion. The distance from recommendation to execution is one click, not five apps.
Why the three failures are really one failure
Stale data, opaque scoring, and missing execution look like three separate product gaps. They're actually one architectural gap. If your recommendation engine is a separate product from your system of record, all three failures follow automatically. The data is stale because it has to be copied over. The reasoning is opaque because the engine only sees fragments and can't cite sources it doesn't have. The execution path is missing because the engine has no write access to the systems where work happens.
That's why bolting a next-best-action feature onto a fragmented stack keeps disappointing. The fix isn't a better model. It's collapsing the distance between the data, the reasoning, and the action until all three run in the same place.
Three questions separate real next-best-action from the widget version. Does the recommendation read live operational data, or a synced copy? Can a rep see exactly which signals produced it? Can the rep execute it without leaving the screen it appeared on? Teams that answer yes to all three are the 2.6x cohort. Teams that answer no to any of them bought a suggestion box.
How PipeLance wires recommendation to action
PipeLance was built around this exact loop. Everything runs on a single Supabase (Postgres) operational database with zero ETL, so when Deal Intelligence generates a recommendation, it's reasoning over the deal as it exists right now: the call that ended twenty minutes ago, the proposal view from this morning, the stage change from an hour ago. There is no synced copy to go stale because there is no copy.
Recommendations arrive as deal playbooks generated from live deal context, and each one cites its evidence: which signals fired, which activities are missing, what changed since the last touch. The rep can check the reasoning against the deal record because they're the same record.
Execution happens through the same 119 native AI tools, spanning 18 categories, that read the data in the first place. The intent-to-action pipeline turns "re-engage the champion before Friday" into a drafted email, a scheduled task, and an updated deal, and every action lands in a full audit trail with rollback. If the AI does something a rep disagrees with, it's visible and reversible, which is what makes reps comfortable letting it act at all. Deal Intelligence and Forecasting ship in the Pro tier at $149 per user per month; the Core tier starts at $69.
The gap compounds from here
Zoom out and the Gartner numbers describe a divergence, not a feature comparison. If 95% of seller research workflows begin with AI by 2027, then within two years the recommendation layer becomes the front door of the entire sales motion. Teams whose front door opens onto correct, explainable, executable guidance will compound the 2.6x advantage every quarter: better actions produce better data, which produces better recommendations. Teams whose front door opens onto a widget will keep paying for AI, keep reporting 87%-style adoption numbers, and keep wondering why the growth never showed up. The 2.6x isn't a bonus for buying AI. It's the return on wiring it all the way through.
Recommendations that ship as actions, not widgets.
See deal playbooks generated from live data and executed in one click, with every action audited.
Request Access