Between April and June 2026, Clarity Recruitment interviewed senior leaders at sixteen organizations and asked them, candidly and off the record, where AI is actually working inside their companies, where it’s stuck, and what kind of people they wish they had.
Nobody said the problem was tools. Almost everyone described the same hiring profile: a domain expert in their own function who is also fluent in AI, sitting inside their team. Almost nobody had found that person.
This page summarizes what he heard and the findings that are consistent and actionable. The full report, What We Heard, and What the Frontier Is Saying, is available at the bottom of the page.
The findings in brief
The most common primary bottleneck was data and tools. Skills almost never topped the ranking.
In eleven of sixteen conversations, leaders described a comprehension gap rather than a skills gap: people can’t picture what AI would do for their own job. Here, company size predicted almost nothing. The two most AI-advanced organizations were a company of about fifty people and one of more than a hundred thousand.
More than ten of sixteen leaders asked for the same hire, unprompted. That hire is three people, not one.
The clear majority if leaders told us they want transitional capability of six to twelve months, not permanent headcount.
The teams most successful with AI implementation were not doing old work done faster. The success is in work that wasn’t happening at all.
Who Clarity spoke with
The sample was sixteen organizations and more than sixteen senior leaders, interviewed across roughly ten weeks between April and June 2026.
By seat: seven finance leaders, four in people, HR and operations, three founder-operators, two in technology and analytics.
By size and stage: from a company of about fifty people to one of more than a hundred thousand. Public, private-equity-backed, venture-backed, and founder or family owned.
By sector: financial services, software, real estate, manufacturing, environmental services, media, medtech and IT services. No single vertical dominated.
Why isn’t AI delivering value in most companies?
Because the blocker is rarely the tools. Every leader was asked to stack-rank four layers for their own organization: process, data and tools, skills, and will.
Data and tools came first, in roughly seven of sixteen conversations. Dirty or fragmented data, reconciled by hand, with no single version of the truth to point AI at. The problem was the worst in acquisitive companies. One finance leader described thirty-three companies, each with different data and its own flavour of Salesforce, where it had Salesforce at all.
There was a further twist several leaders raised: cleaning the data isn’t the finish line. Until a human decides which field is authoritative, AI still can’t interpret it reliably. Half of every data problem is a decision backlog, not an engineering backlog.
Will came second and turned out to be two different problems wearing one name. In some organizations leadership had built no mental model of what AI concretely does, so nothing got prioritized or funded. In others, the frontline was afraid of being replaced. One is solved above the top of the org chart, the other below it. The fixes point in opposite directions, which is why the diagnosis matters.
The organizations that had solved the will to use AI all solved it the same way: permission rather than mandate. Founders and CEOs who used the tools themselves, visibly, and framed AI as career-making rather than career-ending. None of the high-adoption organizations ordered anyone to use AI.
Process came third, but the leaders who named it were the most sophisticated in the group. Several others discovered mid-conversation that their data problem was an undocumented-process problem in disguise.
Skills came last, almost every time. Nobody believes the problem is that people can’t learn the tools.
What is the comprehension gap?
The comprehension gap is the distance between being willing to use AI and being able to picture what it would do for your specific job. It appeared in eleven of the sixteen conversations and it doesn’t fit any of the four layers we walked into the study with: process, data and tools, skills, and will.
The willingness is there. The tool-operation skills are learnable. What’s missing is use-case imagination.
This matters because the two obvious responses both fail. Training courses teach tools, not use cases. Mandates supply pressure, not imagination. What closes the gap, in the organizations that closed it, is a person inside the function showing people what’s possible in their own work.
The outside view sharpens the point: KPMG’s reading of leaders naming the skills gap as their top barrier moved from 25% to 62% in a single quarter in 2026. If the research holds, a large share of that reskilling budget is aimed at one layer too shallow.
Why does the order of operations matter so much?
Because process redesign is not a delay before the AI platform. It is the foundation of the work your AI tool will do, and skipping it means paying for the technology twice.
One financial services organization inherited a multi-million dollar fleet of RPA bots installed on processes nobody had redesigned. The bots faithfully automated the chaos. The company hired additional people to manage the failures falling out of them, then spent months decommissioning the program.
The counter-example came from one of the largest organizations in the sample, built through years of acquisition. It has spent more than a year standardizing data, systems and processes across acquired entities before asking anyone to build AI on top. From outside, that discipline reads as slow. From inside it is methodical.
Does company size predict AI success?
No. The two most AI-advanced organizations in the research were a company of about fifty people and one of more than a hundred thousand. Organizations in the tens of thousands were among the most stuck, not for lack of willingness but because decision rights were spread so widely that no single seat could move alone.
Two things predicted movement instead:
Concentrated decision rights
Every fast-moving organization had one person with both conviction and the authority to act. A founder who feels the pain personally, a CFO who owns the budget, an owner who has declared the direction.
The working unit is a functional team, not the entire company
Wherever AI capability lands, it lands in a group of two to nine people. The central finance team consolidating dozens of acquired entities inside a several-thousand-person group is a handful of people. The finance team at a twenty-five-person startup is three. You are never transforming twenty thousand people at one time. You are transforming one small team, then the next.
What AI roles should a company hire?
More than ten of the sixteen leaders described the same hire unprompted: a person inside the function who knows the work and is genuinely fluent in AI. Not a central data science group. Not a consultant who visits.
They were equally clear about what they didn’t mean. In nine conversations, leaders drew a sharp line between people who can talk about AI and people who can operate it, and said they’d had their fill of the first kind.
The person they describe is actually a cluster of three unique roles, which makes finding an all-in-one candidate particularly challenging:
The Functional AI Expert: the domain expert in accounting, finance, HR, IT or operations who designs AI-enabled workflows inside their own function. Deep domain expertise, moderate technical depth.
The Forward-Deployed Engineer: builds and ships the AI in production. The platforms, integrations and reliability behind it. Deep technical depth, moderate domain expertise.
The AI Enablement and Change Lead: sets direction and makes adoption stick through governance, training and workflow redesign. Strong on organizational change, lighter on technical depth.
Together they cover the four capabilities AI transformation work requires: business and strategy, technical depth, domain expertise, and organizational change. No single hire covers all four. Organizations that hunted for the all-in-one person stayed stuck.
Screening for these profiles is behavioural, not credential-based: what hiring leaders now look for, and what candidates can show.
Are these permanent hires?
Mostly not. When we asked directly, the clear majority wanted transitional capability. One CFO at a venture-backed software company wanted six to twelve months of help automating month-end workflows, after which the person moves on and the team owns the muscle. Permanent headcount was reserved for the platform layer.
The dosage varies as much as the diagnosis: one person part-time, a pair, a team for a season, or nobody yet. Hiring before an organization can absorb someone produces the next “great at thinking, couldn’t execute” story.
What were the biggest wins?
Not speed. Capacity. The most impressive results across the sixteen conversations had one thing in common: they weren’t old work done faster, they were work that previously didn’t happen at all.
- Board and investor reporting moved from around the tenth of the month to two or three days after close.
- A company of about fifty people that historically barely forecasted now runs a rolling daily revenue forecast.
- A seven-day financial forecasting job was rebuilt in roughly two hours, by a curious staffer rather than a project team.
- A residential real-estate operator now does 40% of its leasing after hours, demand that used to ring into a closed office and go unserved.
Efficiency asks what we already do that could be faster. Capacity asks what isn’t happening at all because nobody has the hours. Which customers go unserved, which analysis never gets run, which scenarios never get tested. Every step-change result in the research came from the second question.
The work that produced these results is consistent, and it isn’t specific to finance: messy data in, judgment out, and years of being deferred.
What the rest of the market is doing about it
In roughly the same weeks these interviews were taking place, about $9 billion was committed by OpenAI, Anthropic, Microsoft and AWS to stand up the same model: engineers embedded inside customer teams, redesigning the workflow and shipping a working system, then handing it off.
It is almost word for word what the sixteen leaders described.
Almost all of that money is aimed at the largest global enterprises. 83% of firms with more than 5,000 employees have deployed AI, against 42% of firms between 50 and 499. Only 25% of organizations have moved even 40% of their pilots into production, and 7% report established ROI (Deloitte, January 2026; KPMG Global AI Pulse, June 2026). BCG’s 2026 research puts roughly 70% of AI’s value in people and process rather than the technology.
The independent mid-market is largely being left to build this capability on its own.
What Clarity takes from this
AI doesn’t change what makes a team work, or what makes a good hire. It makes the same things matter more: culture, mandate, curiosity, clean process, clean data, functional depth.
Those always separated a good team from an average one. What’s different now is the size of the gap. Teams that have those qualities aren’t doing the old work faster. They’re doing work the other teams can’t do at all.
Which makes the hiring decision more consequential than the platform decision. You can’t train curiosity. Noticing that AI could fix a problem everyone stopped seeing is something people either do or don’t do. You can’t buy a culture of experimentation. And you can’t teach a technologist the functional dynamics of a finance team, an HR function, or an operations floor. That comes from sitting in the seat.
Hire people who go looking. Put them on a team that rewards it. Use the tools yourself, where people can see you.
The tool follows. People come first.
The complete research runs in two parts. Part I is what the sixteen leaders said, aggregated and anonymized, including the full roles map, a self-diagnostic you can run on your own organization, and eight patterns we didn’t expect. Part II sets those findings against what the consultancies, the frontier labs and the most-cited voices in AI published in the first half of 2026.
About this research
Clarity Recruitment is a specialist recruitment firm in Toronto working with mid-market and growth-stage companies across accounting, finance, HR, IT and operations. The listening tour was conducted between April and June 2026 across sixteen organizations. All interviews were off the record; every finding published here is aggregated and anonymized. The tour is ongoing.
Frequently asked questions
Why isn’t AI delivering value in most companies?
Because the blocker is rarely the tools. Across sixteen senior-leader interviews conducted by Clarity Recruitment in 2026, the most common primary bottleneck was data and tools, followed by leadership conviction and frontline fear. Skills almost never topped the ranking.
What is the comprehension gap in AI adoption?
The gap between being willing to use AI and being able to picture what it would do for your specific job. It appeared in eleven of sixteen conversations. Training closes skills gaps. It does not close this one. What closes it is someone inside the function showing people what’s possible in their own work.
What AI roles should a mid-market company hire?
Three profiles: a Functional AI Expert who designs the workflow inside the function, a Forward-Deployed Engineer who builds and ships it, and an AI Enablement and Change Lead who makes adoption stick. An AI Process Engineer is needed wherever the process itself has to be redesigned first.
Are AI roles permanent hires or transitional?
Most leaders interviewed wanted transitional capability of six to twelve months with a hand-off, not permanent headcount. Permanent roles were reserved for the platform layer.
Does company size predict AI success?
No. The two most advanced organizations in the research were a fifty-person company and a hundred-thousand-person one. What predicted movement was concentrated on decision rights and small functional teams of two to nine people.
Is AI reducing finance and accounting headcount?
The 2026 evidence does not show that. The Yale Budget Lab found no meaningful change in unemployment for AI-exposed workers through early 2026. What the research shows is repricing rather than replacement: the people who adapt become more valuable.
