By Mark Walker, AI GTM Studio. Published 9 October 2026.
01The short version
Between March and October 2026 I recorded 261 calls with revenue teams: clients, prospects, founders and operators in my network, most of them talking openly about where AI was and wasn't working for them. I coded every friction, workaround and win in those calls, which came to 3,710 moments across 107 companies, to see what actually gets in the way. Most of those companies came up in a single call, so wherever I count by company I use the 17 with at least 10 coded moments each, which is where there's enough material to tell whether a problem is really there.
Most of it isn't AI. 69% of the friction involves no AI at all. Missing rules, too few hours, bad data and indifferent buyers (or lack of real PMF) do most of the damage.
A few frictions show up nearly everywhere. Market/buyer indifference, missing specification and skill gaps each appear at 13 of the 17 companies with the deepest coverage in the data set, and they're also the three most widespread across all 107.
Problems get named far faster than they get fixed. When a gap came up, 1.5% had already been fixed and 36% were "in progress". That's the status before anyone dedicated had worked on it.
The problem moved inward over the summer. Bad data went from 42% of calls to 63% while market/buyer indifference fell from 55% to 37%, as teams moved from using AI to building with it.
Maturity changes the problems, not the amount. AI work produces about twice the wins of non-AI work, but the share of weaknesses barely moves between tier 1 and tier 3.
Where you sit decides what you see. ICs report missing capability at 3 times the rate leaders do, while leaders talk about capacity and specification.
Where an AI GTM engineer was embedded, teams moved well past autocomplete. 78% of their AI work was team-wide or custom-built (tier 2 or 3) against 52% everywhere else, and custom workflows made up 22% of it, more than 3 times the 7% elsewhere.
Market/buyer indifference, missing specification and skill gaps are the closest thing in the data to universal problems.
I've counted by company rather than by mentions, because one large client talking a lot would otherwise tilt the whole ranking. Every one of the 17 companies has at least one of the top three, and 9 have all three of them.
Across all 107 companies, one-call conversations included, the same three still come out on top:
- market/buyer indifference at 39 companies,
- skill gaps at 33, and
- missing specification at 30.
Market/buyer indifference covers prospects who don't care enough to act, can't tell you apart from the next vendor, or don't know they have the problem yet. Half of it turns up in outreach and in moving live deals forward.
“Usually they don't even know they have that pain before you're talking to them, because everyone does it the same way and uses the same companies.
Prototype to production is the only rare one, at 3 companies, and that's probably because few teams had got far enough with their builds to hit it. It doubled over the period, so I'd expect it to keep climbing.
2,549 of the 3,710 moments involve no AI at all. Having run revenue teams as a CRO three times, I'd recognise most of that list from long before AI turned up.
The frictions split fairly cleanly in two. The figure in brackets is the share of that friction's moments where AI was involved:
AI-native, where most moments involve AI: output quality (81%), prototype to production (81%), unreliable tools (54%), access and permissions (51%) and usage caps (49%).
Human, where AI barely features: market/buyer indifference (3%), no measurement (10%), missing specification (12%) and leadership change (12%).
Buying more AI helps with the first list and does very little for the second, and the second list is the bigger one.
When AI is involved, the top problem is the output itself. 16% of the 1,161 AI-related moments are about output that's wrong, generic or off-voice, followed by missing capability at 13%. Usage caps get a lot of airtime but account for 6%.
“The outbound retro skill gets the numbers wrong quite a lot. I was trying to work out where it gets its data from and tell it where its source of truth should be, but I keep running out of tokens.
“Everybody is testing, but everyone is testing different things. There's no baseline.
04Gap or opening
Every moment was also classed as a strength, weakness, opportunity or threat for the company in the call, and that split says most of this is self-inflicted.
The one I'd pick out is usage caps. The token ceiling feels like the provider's problem, yet 75% of usage-cap moments are internal gaps (the wrong plan, no thought about which model does which job, one tool being asked to do everything) and only 7% are a limit the company can't move.
“From the moment I logged on I had no credits left. Every time I close and reopen it I can ask it something, and then I'm out of tokens again.
“It completely consumed our credits, something like 50,000 per month, within two days.
The untaken upside sits mostly with skill gaps, where 11% of moments are an opportunity nobody's acted on, then usage caps at 10% and missing capability at 9%.
05Where the work breaks
Pairing what people were trying to do with what stopped them, the most widespread failure is strategy without specification: 85 moments across 20 companies, where the plan gets set and the rules that would make it runnable never get written.
| What they were trying to do | What stopped them | Moments | Companies |
|---|---|---|---|
| Plan strategy and territory | Missing specification | 85 | 20 |
| Run outreach | Market/buyer indifference | 56 | 16 |
| Progress and close deals | Market/buyer indifference | 57 | 15 |
| Roll out tools | Skill or awareness gap | 44 | 12 |
| Roll out tools | Adoption or discipline gap | 53 | 11 |
| Plan strategy and territory | Leadership or org change | 53 | 11 |
| Report and forecast | No measurement | 66 | 8 |
| Maintain the CRM | Bad or fragmented data | 85 | 5 |
Sorted by how many companies each pairing shows up at. CRM upkeep against bad data is the deepest single pit at 85 moments, but it's concentrated in 5 companies, so it's a common client problem across multiple stakeholders, rather than a market-wide one.
“Every month there are 10 new custom fields, but fast forward six months and only one of them is being used. We've got all that tech debt.
“There are old sequences, workflows and automations built by a previous growth team, where no one in the business still works, and no one knows how to untangle them.
Of 2,365 internal weaknesses with a recorded response, 35 had been fixed at the point they came up. 857 were planned or in progress and 384 were being worked around by hand.
That's a snapshot of each problem at the moment it was raised, before a dedicated AI GTM engineer had worked on it, and problems come up on calls precisely because they're still unfixed. So read it as how far teams get on their own rather than as a final score.
The second-biggest bar is the one to sit with. "No next step mentioned" means the problem was named and then the conversation moved on, with no plan, workaround or owner attached to it. Some of those will have been picked up later, but a third of problems leaving the call with nothing attached goes a long way to explaining why missing specification sat at the same level all year.
The split by friction is where it gets useful:
| Friction | Worked around | In progress | No next step | Fixed | Gaps |
|---|---|---|---|---|---|
| Output quality | 33% | 26% | 27% | 2% | 129 |
| Missing capability | 31% | 31% | 26% | 1% | 206 |
| Usage or cost ceiling | 31% | 36% | 17% | 4% | 72 |
| Prototype to production | 6% | 65% | 23% | 0% | 31 |
| Missing specification | 8% | 38% | 44% | 1% | 347 |
| Skill or awareness gap | 5% | 37% | 44% | 1% | 187 |
AI problems get worked around, with about a third of output-quality and missing-capability gaps already carrying a manual workaround. Missing playbooks and skill gaps don't get workarounds, and in 44% of cases nobody said anything about what would happen next.
Prototype to production is the starkest row, with 65% in progress and none fixed.
“It takes no time at all to get to a large chunk of the work and then it takes a long time to turn it into something you can put into production.
I use a simple three-tier model for AI maturity in revenue teams. Tier 1 is AI as autocomplete (drafting, summarising, individual time saved), tier 2 is AI scaling the team through shared tools and repeatable workflows, and tier 3 is custom workflows built around a company's own process, with guardrails and human checks defined.
Across the whole data set, tier 1 fell from 24% of moments in March to 3% in September, while tier 3 rose from 1% to 12% and peaked at 20% in August.
Almost all of that climb happened at the 3 companies with an embedded AI GTM engineer, covered below.
Everywhere else, AI work stayed mostly at tier 1: it was 45% of their AI-related moments in spring and 61% in summer, and tier 3 never got above 8% (those are mostly different companies in each half, so treat it as a read on the wider market rather than the same teams standing still).
| Tier | Top problems | Top wins | Strengths | Weaknesses |
|---|---|---|---|---|
| No AI | Capacity, missing specification | Outreach, lead routing | 15% | 67% |
| 1. Autocomplete | Output quality 21%, skill gaps 17% | Writing, account research | 29% | 60% |
| 2. Scaling the team | Missing capability 14%, output quality 12%, data 11% | Tool roll-out, research, routing | 30% | 55% |
| 3. Custom workflows | Output quality 18%, missing capability 13%, data 10% | Research, routing, outreach | 32% | 56% |
Every moment was also tagged as a strength (something working) or a weakness (something broken), which is what the last two columns show. Two things come out of that.
AI work goes better than non-AI work. When AI is involved, about 30% of what people describe is working, against 15% when it isn't, so conversations about AI are twice as likely to be about a win.
Climbing doesn't make the problems go away. 60% of tier 1 moments are weaknesses and 56% of tier 3 moments still are. The problems change (output and skills early on, then missing capability and data) but there are roughly as many of them, so expect a fresh set at every step up.
What a tier 3 win sounds like, from a founder who built one:
“When we get a positive disposition, it automatically creates a transcript from the dialer, writes that into a follow-up email, puts it in my outbox so I can review and check it, and then creates a deal for me in [the CRM].
Where an AI GTM engineer was embedded
Three of the companies in the data had a dedicated AI GTM engineer working inside the revenue team, and that's where the climb was concentrated.
| Company | Window | Tier 2 and 3, first month | Tier 2 and 3, last month | Tier 3, first month to peak |
|---|---|---|---|---|
| Company A | Mar to Sep | 15% | 40% | 0% to 23% |
| Company B | May to Sep | 18% | 26% (58% in Aug) | 0% to 18% |
Company C had too few calls after April to show a trend. Across all 3, 78% of the AI work discussed was tier 2 or 3, against 52% at the other 104 companies. Custom workflows alone were 22% of it, against 7% elsewhere, while autocomplete-level use was 23% against 48%.
Company A: production apps for events, signals and warm-path introductions, plus tech-stack enablement and CRM syncs.
Company B: a new inbound qualification and routing system, fully live.
Company C: 5 AI skills built, several of them used daily.
As teams moved from using AI to building with it, the blockers moved from the market and the skills gap to their own data and plumbing.
That holds without the largest client in the data, too: bad data still goes from 31% of calls to 53%, and market/buyer indifference from 64% to 39%.
Missing specification didn't move at all, sitting at 57% of calls in spring and 58% in summer.
“What I've always had to do with board reporting is go through and really validate each deal.
“I opened up my seven spreadsheets, isolated what the real CPA trend was, and then figured it out and fed it to the LLM.
09Who feels it
The higher up someone sits, the more their problem is the plan, and the closer they are to the work, the more it's the tools.
| Role | Moments | Companies | Top frictions (share of theirs) | Main job | Gaps worked around |
|---|---|---|---|---|---|
| Founder or CEO | 175 | 41 | Market/buyer indifference 21%, capacity 11%, output quality 10% | Writing content (22%) | 22% |
| Commercial leadership | 1,189 | 21 | Capacity 17%, missing specification 12%, data 10% | Lead routing, strategy | 14% |
| Individual contributor | 1,044 | 9 | Capacity 14%, market/buyer indifference 13%, missing capability 12% | Outreach (13%) | 20% |
Founders live at tier 1, with 21% of their moments there against 8% for leaders. They're often writing their own content with AI while worrying about the market.
Leaders own the backlog. 40% of their gaps are "in progress", the highest of any group, and capacity and specification dominate what they talk about.
ICs hit the tool gaps and route round them. They report missing capability at 3 times the rate leaders do, and work around 1 in 5 of their gaps by hand.
Roles here come from names, call titles and context rather than being confirmed one by one, and 35% of moments couldn't be assigned, so treat this cut as directional.
The pattern across all of this is that the unglamorous layer underneath (rules, data, ownership, enablement) decides whether the AI work lands. These are the 8 changes I'd make first, each tied to what the calls showed.
By role:
Founders: hand off the tier 1 work so your time goes on the market problem.
Leaders: turn "in progress" into shipped by cutting the backlog to 3 things with owners and dates.
ICs: log your workarounds, because each one is a missing capability that someone above you probably can't see.
11Score your own team
15 statements, one per friction. Score each one: strongly disagree 0, disagree 1, agree 2, strongly agree 3. The higher your total, the more friction your team is carrying.
12What next
If any of this looks familiar, the useful next step is working out where your own team sits on the ladder and what it would take to get to tier 3.
Book a private consultation with me and we'll lay that out together:
Where your team sits on the three tiers today, and which of the frictions in this report are holding it there.
The roadmap from there to tier 3, in the order I'd tackle it, starting with the layer underneath.
What getting there should change, in results, in new capabilities your team doesn't have yet, and in the hours it gives back.
13Methodology
261 calls recorded between 16 March and 1 October 2026, worked from full transcripts rather than AI summaries. 236 calls from 107 companies had codeable content.
Every friction, workaround and win was coded five ways (the job, the constraint, the response, SWOT and AI maturity tier) against a 16-constraint codebook that was frozen before the full pass.
A stronger second model re-coded 434 moments blind, agreeing 76% of the time on job, 78% on constraint and 90% on SWOT.
Before writing anything, a recall test checked the results against 11 findings I knew were in the calls, and all 11 came back.
My own voice is excluded: 1,016 moments where I was speaking, or talking about my own businesses, are out.
This is my network, so it skews towards teams already thinking about AI in go-to-market. Company names are removed throughout.
Mark Walker, "What actually stops AI in go-to-market", AI GTM Studio, October 2026 and link to this page. Download the PDF.