87% of professional services firms plan to put AI agents into their workforce. Only 12% fully trust their own data. 88% trust AI outputs enough to use them. 89% verify those outputs by hand anyway. Something does not add up.
The numbers come from Kantata's 2025 State of the Professional Services Industry Report, a survey of 200 professional services leaders conducted by Censuswide. On paper, the industry is racing into the AI era. Read past the headlines and a quieter story appears. Firms are buying AI while still doing the work twice.
This is what happens when you automate on top of a coordination layer that was never structured in the first place.
87% want AI agents. 12% trust their data. Do the math.
The Kantata report is clear about where professional services believes its future lives. 89% of leaders say revenue growth will depend more on scaling AI than on scaling headcount. 66% already turn down work because they cannot resource it. 68% cite skill availability as their biggest constraint, up from 45% a year earlier. The pressure to adopt is not ambition. It is capacity.
So firms move fast. They plug AI agents into project management. They add copilots to the inbox. They pilot generative tools in proposal writing, scoping, and review. Spending is up. Vendor slide decks promise outcomes.
Then the same report shows the other half of the picture. Only 12% fully trust their own data. 88% trust AI outputs for operational decisions, yet 89% spend time verifying those outputs. A partner reads the summary, then reads the source file. A manager accepts the AI draft, then rewrites half of it. The time saved in step one gets spent in step two. The output looks like progress. The day looks the same as it did last year.
This is the verification paradox. A firm cannot trust its data, so it cannot fully trust the AI trained on that data, so the AI saves time that humans spend earning back.
The data problem in professional services is an external collaboration problem
Most of the conversation about AI readiness focuses on internal data. Clean CRMs. Tidy project plans. Well-tagged document libraries. Those things matter. But for audit, accounting, legal, M&A, architecture, and insurance brokerage, the real data problem lives outside the firm's walls.
It lives in email threads with clients. In the WeTransfer link a partner sent on Tuesday. In the Excel request list that got forwarded twice before anyone opened it. In the scanned PDFs that arrived at 11:47 PM the night before the deadline. In the PBC request that was closed in the firm's audit software but never actually resolved on the client side.
This is the coordination layer. It is where a third of professional services work happens, and it is almost completely unstructured.
Harvard Business Review research by Rob Cross, Reb Rebele, and Adam Grant found that most managers spend 85% or more of their work time on email, meetings, and phone calls. McKinsey Global Institute research shows knowledge workers spend around 28% of their workweek on email and close to 20% searching for information. In professional services, a large share of that time is spent chasing clients for documents, clarifying what was sent, or rebuilding status manually from threads nobody tagged.
When AI is dropped into that environment, it inherits the mess. It cannot tell which version of the balance sheet is the one the client signed off on. It does not know that the partner's promise in an email on March 3rd changed the engagement scope. It does not see that the "approved" label in the audit tool does not match what actually happened in the reply chain.
So the senior professional reads the AI output and checks it against the source. Every time. Because the source was never clean enough to trust.
What HBR's research on AI readiness actually shows
This is not speculation. A Harvard Business Review Analytic Services report from 2025, sponsored by Workato and AWS, surveyed 603 business and technology leaders worldwide. Only 20% said their technology infrastructure was fully ready for agentic AI. 15% said the same about their data and systems. Just 12% felt risk and governance controls were fully in place.
Across industries, 43% of respondents said they only trust AI agents with limited or routine tasks. 39% restrict them to supervised use cases or non-core processes. Even with heavy investment, firms are running AI on the edges. Not because the technology cannot handle more. Because the ground underneath it cannot.
Professional services sits squarely in this pattern. Deadlines are hard. Regulatory exposure is real. A misfiled document or a missed deadline has consequences that show up in audit opinions, compliance reports, and client relationships. When the coordination layer is fragile, giving AI more autonomy feels reckless. So it does not get more autonomy. And the promised productivity gains stay on the slide.
Structure first, intelligence second
The firms moving past the verification paradox are not racing to add more AI. They are rebuilding how work gets requested, owned, completed, and approved between the firm and its clients, before intelligence gets layered on top.
That rebuild looks specific. A document request exists as a discrete item, not a sentence in a reply chain. It has a named owner, a deadline, and a clear definition of "complete." Status is visible to both the firm and the client without anyone asking. Approvals and sign-offs happen in the same environment as the request, so context does not get lost between tools. The audit trail writes itself, because the structure of the work makes every action traceable by default.
Once that exists, AI stops being a guess. It can extract requests from a client email because the request format is defined. It can flag a stuck item because "stuck" has a meaning in the system. It can draft a reminder because the system knows who is waiting on what and for how long. The output stops needing a full manual re-check. The AI and the human are reading from the same structured source.
This is where productivity gains compound. Not when you add intelligence to chaos. When you give intelligence something clean to work with.
What high-performing firms are doing differently
The firms that close the trust gap share a few patterns. None of them are revolutionary. All of them are disciplined.
They define what a complete deliverable looks like before the engagement starts, so AI and humans are measuring the same finish line. They centralize client requests, updates, and documents into a single shared view instead of a SharePoint folder and a spreadsheet and an inbox. They build context retention into the system, so next year's audit does not start by re-explaining the client to a new junior. They make progress visible without meetings or status emails, because the system shows who owes what by when. And they automate follow-ups only after the structure is in place, not before.
None of this requires abandoning AI plans. It requires sequencing them correctly. Fix the coordination layer. Then let AI run on top of it.
The quiet advantage
AI is going to keep accelerating. The firms that get there first will not be the ones with the biggest AI budgets. They will be the ones who fixed their data and collaboration infrastructure early, so their AI has something real to work with. The 12% who trust their data today are the 12% who will trust their AI tomorrow.
Everyone else will keep verifying by hand. Slightly faster. Slightly more tired. Wondering why the productivity graph stayed flat.
If your firm is planning its AI investments for 2026, start one step earlier. Ask what a document request looks like today. Ask where a client deliverable actually lives. Ask who knows the status of every open engagement without checking their inbox. If those answers are vague, no AI agent is going to fix it. A coordination layer will.
Alkmist builds that coordination layer for professional services firms. If you are rethinking how your firm collaborates with clients before you scale AI on top of it, it is worth a conversation.
Sources
- Kantata, 2025 State of the Professional Services Industry Report, based on a Censuswide survey of 200 professional services leaders (January 2026). kantata.com
- Harvard Business Review Analytic Services, Agentic AI: Where Organizations Stand and What Comes Next, sponsored by Workato and AWS, July 2025. Coverage: Fortune
- Rob Cross, Reb Rebele, and Adam Grant, "Collaborative Overload," Harvard Business Review, January-February 2016. hbr.org
- McKinsey Global Institute, The Social Economy: Unlocking Value and Productivity Through Social Technologies. mckinsey.com
.png)
