Quick answer
Gartner's 2026 survey of 210 CSOs and senior sales leaders found a bimodal split on AI ROI: 25% of sales organizations report 50%+ positive return on their AI investment, while 20% report 50%+ negative return. There is no fat middle. What predicts which side a team lands on isn't the tool, it's whether saved time gets reinvested, how deep usage actually goes, whether the surrounding process got redesigned, which task got automated first, and whether the data underneath was ready before the tool went live.
Gartner's bimodal AI ROI split
I'm Hlib Storchak. I build outbound systems for B2B founders and sales teams, and I've booked 2000+ meetings for B2B clients doing it. Most of what follows comes from watching the moment a client's AI rollout either compounds or quietly stalls, not from a slide about what it should do in theory.
Gartner surveyed 210 CSOs and senior sales leaders between January and February 2026 and presented the results at its CSO & Sales Leader Conference in Las Vegas that May. Buried inside a broader report on AI adoption is a specific finding worth sitting with: 25% of sales organizations report a 50% or higher return on their AI investment, while 20% report a 50% or higher negative return. Gartner VP Analyst Dan Gottlieb, who presented the data, put the takeaway plainly: AI value depends less on access to the technology and more on how a sales organization redesigns the systems around it.
That is not a distribution with a comfortable middle. Roughly 45% of organizations sit at one of two extremes, either clearly ahead or clearly behind, and only the remaining slice reports something closer to a moderate, unremarkable return. If you are budgeting for an AI SDR, a sequencing platform's new AI layer, or any GTM AI tool this year, the honest planning question isn't "what's the average ROI on this," it's "what determines which of these two groups I end up in."
Why the "modest average" takeaway is wrong
Most coverage of AI ROI surveys collapses everything into one blended number, something like "companies report X% average return," and that number is almost always unhelpful precisely because it averages a group that is genuinely split into winners and losers rather than clustered near the mean. A blended average from a bimodal distribution tells you nothing about your own odds. It's the sales equivalent of being told the average temperature in a room where one half is on fire and the other half is in a freezer.
The more useful frame, and the one Gartner's own data supports, is to treat AI ROI as a coin with a heavily weighted outcome on each side rather than a normal curve. Once you accept that, the actual work is figuring out what tilts the coin, not what the "average" outcome supposedly is.
Tip. If a vendor or a roundup quotes you a single average ROI percentage for AI in sales, ask whether the underlying data is bimodal like Gartner's. A blended average from a split distribution is close to meaningless for predicting your own outcome.
Factor 1: reinvested or pocketed
The same Gartner survey found AI saves sellers an average of 4.8 hours a week, but 72% of sales organizations report low reinvestment of that saved time into high-value selling activities. I've written separately about which GTM tasks to keep human as the other half of this problem, but the reinvestment question stands on its own: organizations that do put the freed-up hours into account research, live discovery, or multi-threading are 2.2x more likely to exceed their customer growth goals and 3.1x more likely to exceed their lead-to-opportunity conversion goals. An hour an AI tool frees up is not automatically a productive hour. It's an hour that has to be claimed by someone, on purpose, or it gets absorbed into admin and nothing changes.
Factor 2: usage depth, not headcount
A separate piece of Bridge Group's 2026 AE and quota research, unrelated to the ROI survey but measuring a similar underlying idea, found organizations in the highest "AI Engagement Score" tercile reported 57% of reps at quota, against 39% in the lowest tercile, an 18-point gap tied to how much a team actually uses its AI tools day to day, not whether it bought a license. Rollout counts, seats provisioned, and dashboards configured are adoption metrics. They are not usage metrics, and adoption without depth is exactly the kind of AI spend that shows up in Gartner's negative-ROI cohort: paid for, technically live, barely touched.
Factor 3: redesigned, not bolted on
Gottlieb's framing of the Gartner data is worth repeating in full because it names the actual mechanism: sales productivity stalls because the surrounding system quietly caps what a rep can do, not because reps forget how to sell. Dropping an AI SDR or an AI drafting tool into an unchanged process, same territory assignment, same manual handoffs, same reporting cadence, tends to just move the bottleneck rather than remove it. The teams that land in the positive-ROI cohort are usually the ones that changed a process step to make room for what the tool actually does well, not the ones that expected the tool to fix a process they never touched.
This is the audit I run before adding any AI tool to a client's stack: name the specific process step that has to change once the tool goes live, in writing, before it goes live. If nothing in the surrounding workflow is scheduled to change, the tool is being bolted on, and bolted-on AI spend is where I've seen the negative-ROI outcomes concentrate.
Factor 4: which task went first
Not every automatable task carries the same risk if the rollout goes badly. A first task with cheap, reversible mistakes, like drafting a first-touch email a human still reviews, gives a team room to learn the tool's failure modes before anything expensive is on the line. A first task with account-specific memory and real consequences, like autonomous multi-step sequencing into named enterprise accounts with no review step, front-loads the downside before the team has learned how the tool actually behaves. Teams that start with a low-stakes, high-volume task and expand from a working baseline tend to compound toward the positive side. Teams that start with the highest-stakes task because it looked like the biggest win on paper carry more of the risk that shows up in the 20% negative-ROI group.
Factor 5: was the data underneath ready
An AI SDR or scoring model is only as good as the contact data, intent signals, and CRM hygiene feeding it. A tool layered on top of a stale, duplicate-heavy, badly enriched database doesn't get a fair test, it gets amplified garbage at higher volume than a human ever sent manually. This is the quiet failure mode behind a lot of "the AI SDR didn't work" post-mortems I've been asked to diagnose: the agent was rarely the actual problem, the list and the CRM state underneath it were.
The five factors, side by side
| Factor | Positive-ROI pattern | Negative-ROI pattern |
|---|---|---|
| Reinvestment | Freed-up hours named and redirected on purpose | Hours saved, absorbed into admin, untracked |
| Usage depth | Deep daily use by fewer, committed reps | Wide seat rollout, shallow actual use |
| Process redesign | A workflow step changed to fit the tool | Tool bolted onto an unchanged process |
| First task chosen | Low-stakes, reviewable, high-volume task first | Highest-stakes task first, no review step |
| Data readiness | Clean, enriched data before scaling up the tool | Stale or duplicate-heavy data feeding the tool |
A 5-question self-audit
Before trusting whatever ROI number your own AI rollout is currently reporting, or before you sign for a new one, run it through five questions built directly off the factors above. First, can you name, specifically, what a rep now does with the hours the tool freed up. Second, what share of licensed seats used the tool in the last seven days, not the last quarter. Third, what changed in your actual process, not just your tool stack, since the rollout. Fourth, did the first task you handed over have a human review step while the team learned the tool's failure modes. Fifth, when was the underlying contact and CRM data last cleaned and verified before the tool started acting on it. A confident answer to all five doesn't guarantee the positive-ROI outcome, but a team that can't answer two or more of them is describing, almost exactly, Gartner's negative-ROI cohort.
Where this hits outbound and AI SDR pilots first
Outbound is usually the earliest place this split shows up, because it's the part of the GTM motion most founders automate first and measure fastest. An AI SDR that drafts, sequences, and triages replies at volume can be a genuine positive-ROI move, but only once the five factors above are actually in place. The AI SDR stack I run for clients, Salesforge's Agent Frank, is a case in point: it earns its keep on the repeatable half of outbound, research, drafting, and reply triage at a volume no rep sustains solo, and it only pays off once the data feeding it is clean and someone has explicitly claimed the hours it frees up for higher-value work. Handed a messy list and an unchanged process, the same tool produces the same negative-ROI story Gartner's 20% describe, just faster and at higher volume than a human would have.
What the wrong side of the split actually costs
Here's a way to size what a negative-ROI AI rollout actually costs, using inputs you should swap for your own rather than trusting this exact figure.
Assume a monthly AI SDR or sales-AI tool spend of €2,000 to €5,000, a team of 5 reps at a fully loaded cost of roughly €70,000/year each including on-costs, and that the rollout consumes about 10% of each rep's time in the first quarter (learning the tool, reviewing its output, fixing what it gets wrong). That's roughly €8,750 to €9,000 a quarter in rep time alone, on top of the tool spend, so €14,750 to €18,750 in the first quarter before counting a single meeting booked.
If that spend lands in Gartner's positive-ROI cohort, a 50%+ return means the tool is generating meaningfully more in booked-meeting value than it costs, a real win worth scaling. If it lands in the 20% negative-ROI cohort, the same quarter has produced a real cash cost, real rep time spent on review and cleanup, and no offsetting pipeline to show for it, which is a materially worse outcome than simply not automating that quarter at all. The point of the model isn't the exact euro figure, it's that "we're not sure if this is working" is not a neutral state. It is either compounding or actively costing you, and the five factors above are what decide which.
My take, after watching this with clients
The founders who end up in Gartner's positive-ROI quarter aren't the ones who picked the flashiest AI tool. They're the ones who could answer the five-question audit above before they signed anything, and who treated the tool as one piece of a system they were willing to change, not a fix for a system they weren't. Everyone else is rolling a weighted coin and hoping it lands their way. Run the audit first. The tool matters less than people assume.
Key takeaways
- Gartner's 2026 survey of 210 CSOs found a bimodal AI ROI split: 25% of orgs report 50%+ positive return, 20% report 50%+ negative return, with no fat middle.
- A blended "average ROI" figure is close to meaningless when the underlying distribution is this split. The real question is which side you're likely to land on.
- Five factors predict the split: reinvesting saved time, usage depth over seat count, redesigning the process rather than bolting the tool on, choosing a low-stakes first task, and clean data underneath the tool.
- A negative-ROI rollout isn't neutral. It's a real cash and rep-time cost with no offsetting pipeline, which is why "let's just try it and see" is a riskier default than it sounds.
- Outbound and AI SDR pilots are usually where this split shows up first, since they're the earliest and most measured part of the GTM motion most founders automate.
FAQ
What did Gartner's 2026 AI ROI survey actually find?
Gartner surveyed 210 CSOs and senior sales leaders in January and February 2026 and found a bimodal ROI split: 25% of sales organizations report 50% or higher positive return on their AI investment, while 20% report 50% or higher negative return. Gartner VP Analyst Dan Gottlieb presented the finding at the company's May 2026 CSO & Sales Leader Conference.
Why isn't the average ROI on AI in sales a useful number?
Because the underlying distribution is split into two extremes rather than clustered near a middle value. A blended average from a bimodal split tells you little about your own likely outcome, which is why the more useful question is what predicts which side of the split a team lands on.
What actually predicts whether an AI sales tool pays off?
Five factors, based on Gartner's own data and adjacent research: whether saved time gets reinvested into high-value work, how deep actual daily usage goes versus how many seats were provisioned, whether the surrounding process got redesigned around the tool, whether a low-stakes task was automated first, and whether the underlying contact and CRM data was clean before the tool started acting on it.
Does a negative ROI on an AI tool just mean no gain?
No. A negative-ROI quarter still carries the tool's subscription cost plus real rep time spent learning, reviewing, and fixing its output, with no offsetting pipeline to show for it. That is a worse outcome than not automating that task yet, not a neutral one.
Where does this split show up first in a GTM motion?
Usually in outbound and AI SDR pilots, since they're typically the first and most closely measured part of the funnel a founder automates. The same five factors apply: clean data, a reviewable first task, a redesigned process, deep usage, and reinvested time are what separate a genuine win from a faster, higher-volume version of the same negative-ROI story.