Product1 distinct publisher3 min readPublished
A survey of 1,800 professionals finds 91% saying their firm still falls short on AI. The remedy its chief operations officer names is fewer tools with owners attached, plus a date on which a failed pilot dies.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The line that will sound familiar to anyone who has run a rollout is the one Kirsty Roth says she heard from staff: "I've been given all these things. But what am I meant to be doing?" [6] That complaint and the tool-shortage complaint are the same complaint. The gap between handing out access and handing out a usable tool is what the 41% figure records [4]. The license arrives with nobody's job description attached.
It is worth being precise about what this survey measures, because none of it is value delivered. The headline number is a self-assessment of the firm, and fewer than one in ten respondents said their employer was doing fine [9]. The three requirement figures are stated preferences about what tools must do, not audits of what deployed tools do [3]. The 35% figure measures whether people can find the strategy in their own week [5]. The research reports no retention curve and no time-to-value. Hold it lightly for one more reason: this is Thomson Reuters' own research, interpreted by Thomson Reuters' chief operations officer [12].
The most portable thing in the interview is a calendar. Roth describes giving teams a license to try nearly anything for about six weeks, testing the results, spreading what worked and killing what did not [7]. Six weeks cuts the year into roughly eight windows [8]. The number is the useful part, because it caps how long an unproven tool sits on the books and puts a named person on the hook to end it on a date rather than at renewal. The cost of that rule is real: a tool whose payoff depends on rewriting a workflow will show almost nothing in six weeks, so the kill list will contain some genuine losses.
For sorting which tools earn the production step Roth says separates the leaders from the laggards [10], two questions do most of the work. Does the tool clear the confidentiality, grounding and explainability bar as configured in your own tenant, rather than as described in the vendor deck. And is there a specific process step, with an owner, that disappears if the tool works. Clears the bar and removes a step: production candidate. Clears the bar and removes nothing: that is the software line growing while the work stays identical. Removes a step but fails the bar: the job is process redesign or a different supplier, and the pilot should not be live. Neither: it lapses at week six.
This is written for the person who signs the renewal in March and explains the number in April. Steve Lucas of Boomi told ZDNET the field is hyper-fragmented, carrying a vocabulary of frontier models, agentic harnesses and agentic loops that did not exist a few years ago [11]. That churn is a procurement problem more than a research one, which is why the durable question is a small and dated one: name the step that goes away, and the week the license expires if the step is still being done by hand.
Ranked by verification strength, evidence, and original report placement.
Research based on a global survey of 1,800 professionals from multiple sectors found 91% of professionals say their firm still falls short on AI, and shows a widening gap between AI ambition and reality.
At Thomson Reuters, Roth said teams could get a license to pretty much anything they wanted and try it for about six weeks; results were tested, good results were rolled out to other teams, and poor results led the firm to kill the tool and move on.
Kirsty Roth, chief operations officer at Thomson Reuters, told ZDNET that people are working out these technologies cost lots of money without seeing the value yet, and that the conversation has turned into "the classic change management one" about changing processes and ways of operating.
Respondents said their AI tools must safeguard confidential data (96%), ground outputs in authoritative content (94%), and produce explainable and defensible reasoning (90%).
Two in five professionals (41%) who use AI at work said they do not have access to high-quality tools.
Just over a third (35%) of professionals in firms with a named AI strategy say the approach is not visible in their day-to-day work.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 28, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
Most AI-justified layoffs cost more than they saved, and half the cutters may rehire by 20271 distinct publisher
build
Notion's agent stack is live, not slideware, and it only changes one of your decisions1 distinct publisher
product
Thomson Reuters spent $40M to make a $450K training run worth doing2 distinct publishers
product
Deloitte counts one orchestrated agent deployment for every three expanding ones1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One house's numbers, one voice
Every percentage in this story — 91% falling short, the 96/94/90 trust requirements, 41% without good tools, 35% who cannot see their firm's strategy — comes from research Thomson Reuters ran and is interpreted by a Thomson Reuters executive, with no methodology, fielding dates, sector breakdown or published report to check against. The claim that the gap is widening needs a previous wave to mean anything, and none is shown. ZDNET reports the interview accurately and quotes Roth at length; that is careful stenography, not corroboration, and the one outside voice speaks to market fragmentation rather than to any of the numbers.
Deep in one building, shallow everywhere else
Inside Thomson Reuters the practice is specific and traceable: 87% of employees said to use AI daily, support staff working through Open Arena with Claude against Salesforce data, production effort fenced into five named functions. Outside it, the same story reports diffusion that stops at the desk — two in five AI users without decent tools, a third of people at firms with a strategy who cannot see it in their work. So real deployment exists and is described in operational detail, but the only place we can see it end to end is the company doing the talking, and its usage figure counts logins-worth of activity rather than outcomes.
De-hyped diagnosis, unpriced cure
The diagnosis undersells nothing and overstates little — a supplier's operations chief saying customers cannot see value for their spend is the opposite of a puff piece, and the 'most firms are still in the playground' line cuts against the market's own marketing. The stretch is on the cure. Firms that productionise are said to see improvements and savings with no number behind either word, 87% internal usage is offered where value evidence belongs, and a six-week kill rule earns its headline place without a single example of a tool that actually died. The remedy is asserted at roughly the confidence of a case study while resting on one executive's recollection.
The survey and the cure share an author
Thomson Reuters sells authoritative professional content and the tools built on it, and the research it commissioned reports that 94% of professionals need AI outputs grounded in authoritative content and 96% need confidential data protected — the precise ground on which that business competes. The same company then supplies the executive who names the problem and the remedy. None of this makes the numbers wrong, and Roth's admission that software costs rise faster than visible value is not a comfortable line for a vendor to say. But readers should see that the diagnosis, the survey behind it and one candidate for the cure all come from the same balance sheet, and that the story's second voice is another vendor CEO describing the market his product integrates.
Trust the practice, hold the percentages
We are fairly confident about what was said and by whom: the quotes are long, attributed and internally consistent, and the Thomson Reuters practices are described in enough operational detail to be recognisable. We are much less confident that the survey numbers generalise, because no one outside the company has seen the instrument, and we cannot verify a single outcome claim. Read this as one large firm's account of its own method, reported faithfully by ZDNET, and hold the statistics loosely until a second source fields them.