AI Next Best Actions in Sales: What Gartner's 2.6x Finding Does and Does Not Say

A figure landed on the sales internet in May 2026 and has been riding vendor slides ever since. It is a real finding from a real survey. It is also doing far more work in those decks than the survey can support, and the gap between the two is where budget goes to die.

Here is what was measured, what the same release says that nobody quotes, and the test a sales leader can run before signing anything.

What the Survey Actually Measured

Gartner put the question to chief sales officers and published the answer at its sales leader conference. The release, carried in full on Business Wire, reports a 2.6x greater likelihood of commercial growth among sales organisations that hand their sellers AI-generated recommendations on what to do next, drawn from 227 CSOs fielded from August through September 2025. The same sample produced a companion figure: prioritising AI upskilling for sellers was associated with being 2.4x more likely to see strong revenue growth. Gartner's own site blocks automated access, so these are quoted from the syndicated copy of its release rather than read off the research itself.

Two things follow immediately. First, the respondents are sales leaders describing their own organisations, so both numbers are associations between self-reported practices and self-reported growth. An organisation that funds AI guidance for sellers is usually also an organisation with cleaner data, settled territories and a functioning enablement team, and no survey of this shape can pull those apart. Gartner does not claim otherwise.

Second, the release never defines the term. A next best action could be a ranked list of accounts for Monday morning, a drafted follow-up, a prompt to bring a second stakeholder in, or a notification that an account opened an email. Those are different products at wildly different prices, and every one of them will cite this statistic.

Name the Action Before You Shop for a Tool

Write one sentence per role, and make it specific enough to be wrong. For an SDR: which ten accounts do I contact today, and what is the reason for each. For an account executive: who is missing from this deal, and what do they need to see. If nobody on your team can write that sentence, no platform will write it for them, and the demo will simply fill the silence with features. Our approach in ranking the week before working it is a next best action built from a spreadsheet, which is the cheapest possible version of the test.

Swap, Never Stack

Greg Hessong, the Gartner analyst behind the research, gives the line worth taping to a monitor. The winners, in his framing, are not the leaders who hand their team the same job list and ask for it "just faster". They are the ones redesigning the workflow itself, so that execution and recommendation move to the system and sellers keep the moments where judgement earns its keep.

Operationally that means a rule: for every step a tool adds, name the step it removes, before the pilot starts. If a system now produces the account brief, the manual research block comes off the calendar that same week, and our template for the brief a rep will actually read tells you what the output has to contain to earn that swap. Teams that skip this step usually discover their adoption problem six weeks in, and misdiagnose it as resistance to change.

What the Same Release Says About Human Sellers

The buyer half of the research is more interesting than the headline and gets quoted far less. In a parallel survey of 645 B2B buyers over the same two months, the gap favouring human sellers over generative AI ran to 39 percentage points on grasping what the buyer needed, 32 points on leaving them confident about the decision, 28 points on moving them to the next stage of the purchase, and 21 points on putting a number against the benefit.

One further line from that release deserves its own pipeline review. Buying groups with low dysfunction were 13x more likely to report a high-quality deal, and buyers who spent more time with supplier reps reported the lowest dysfunction. Read alongside the headline, that is not an argument for replacing contact with automation. It is an argument for using automation to buy back the hours that contact requires. Our piece on why buyers take AI answers back to a rep covers the same dynamic from the buyer's side.

A Number We Took Out, and Why

Earlier versions of this argument leaned on adoption figures attributed to a 2025 Harvard Business Review piece. Chasing them back, the underlying numbers come from McKinsey's ninth annual B2B Pulse survey of 3,942 decision makers across thirteen countries, reported by Digital Commerce 360 in September 2024: 19% of respondents were already implementing generative AI for B2B buying and selling, with a further 23% building or experimenting. The attribution we inherited pointed at the wrong publication, and the figures are now two years old, which is a long time in this particular subject. They still make the one point that matters here: plenty of companies are implementing, far fewer are measuring, and the published survey that tells you which group you are in does not exist.

Buy the Workflow Change, Not the Recommendation Engine

The position: this research is a reason to redesign a role and a very poor reason to buy a product, and the two get conflated because only one of them has a salesperson attached. Everything in the release that explains the result points at work design rather than software. Hessong's recommendations, as summarised by Phillip Britt for DestinationCRM in July 2026, are to audit the workflow, rewrite seller role descriptions around AI fluency and pilot agents in a sandbox. None of those is a purchase order. All of them are management work that most teams would rather pay to avoid.

There is a measurable version of the same claim. Gartner forecasts that AI will be the starting point for 95% of seller research workflows by 2027, against fewer than a fifth of them in 2024, and that is a prediction rather than an observation. If it holds, the recommendation engine becomes a commodity and the differentiator is what your reps do with the hours it returns. If a team has not decided what that is, buying the engine earlier simply produces the same activity with better formatting.

A team drowning in accounts with no prioritisation of any kind reads this differently. Where reps genuinely open the CRM each morning and guess, a ranked list is worth real money immediately, and the workflow argument is a luxury you can afford after the bleeding stops. The test is whether your reps can already tell you who they are working today and why. If they can, buy the workflow change. If they cannot, buy the list, then do the workflow work anyway.

A Two-Week Test That Costs Almost Nothing

• Read the original release before the next demo, and ask each vendor to define a next best action in one sentence and say how they measured the effect.
• Pick one team, run two weeks, and hold a second team as a control. Compare meetings booked per rep, not tool usage.
• Pair the pilot with exactly one task removed, and tell the reps which one.
• Record the baseline before you start. Most pilots fail at this step rather than at the technology.
• Keep a named person accountable for every message that goes out, whatever produced the draft.

The structured version of this is in our two-week pilot for prospecting agents, and the judgement question behind it is covered in moving a team from instinct to evidence without discarding the instinct. A survey association is a reason to test something on your own numbers. It is never a result you can borrow. No tool can promise growth, the research quoted here does not claim to, and any vendor who does has told you something useful about themselves.