By: Robert Nelles
As AI compresses billable work from weeks to seconds, a former consulting executive says the industry’s real problem is older than the technology. Firms forget what they know.
Every professional services firm sells the same underlying product: accumulated expertise. Decades of engagements, proposals, contracts, and hard-won judgment. Yet ask a partner at almost any consulting, accounting, or advisory firm to produce the details of a similar project from six years ago, and the search begins. Someone remembers who ran the account. That person is on vacation. The files are on a drive nobody opens anymore.
Daniel Cohen-Dumani spent three decades inside that reality, first as a consultant in Switzerland at the firm now known as Accenture, then as founder of Portal Solutions, a technology consultancy he grew from a team of one to 60 before its acquisition by a large accounting firm in 2017. His diagnosis of the industry is blunt.
“Your firm knows more than it can find,” he says. “Decades of expertise, scattered across drives, inboxes, and people’s heads. Everyone reinvents the wheel because nobody can find what the firm already knows.”
The problem is not new. Knowledge management systems have promised to solve it for thirty years, and Cohen-Dumani built plenty of them. What changed, he argues, is the physics. Large language models can finally read unstructured information at scale, understand context, and retrieve meaning without armies of people tagging documents. “LLMs didn’t improve knowledge management,” he says. “They replaced its physics.”
That conviction led him to found Experio Labs, a company building what it calls organizational memory for high-stakes professional services firms. Its intelligence layer, IQ1, connects to a firm’s institutional knowledge and answers questions that general-purpose AI tools cannot reach. The company’s benchmark example: show every active contract signed in the last ten years with a general-liability clause in excess of $2 million. Generic assistants, Cohen-Dumani notes, go quiet on queries like that. Experio returns the matching contracts in seconds, with the exact clause and source document cited.
The same gap shows up in more mundane moments. A managing partner needs a full briefing on a client before a meeting that starts within the hour. In most firms, that request sets off a chase. Someone has to find out who ran the account, get on their calendar, and sit through a conversation that begins with “honestly, that project was six months ago,” then wait while the history is reconstructed from old decks and memory. Days pass, sometimes weeks. Experio’s answer to the same request is the full engagement history, the people involved, the work product, and the open risks, assembled in seconds and cited to its sources, reliable enough to walk into the room with.
The citation requirement is not a detail. In work where an error becomes a liability, Cohen-Dumani holds a hard line the industry is only beginning to adopt. Answers must be traceable to their source, and a system should say “I don’t have that” rather than guess. “An answer without a source is a guess wearing a suit,” he says. The company describes its retrieval as highly accurate with zero tolerance for fabricated answers, a standard it argues should be table stakes for any AI operating in professional services.
He is equally direct about why impressive demonstrations so often collapse in production. Run a language model across a large knowledge base at scale, he warns, and meaning quietly erodes. Context drifts, similar concepts blur, and the system that dazzled on day one degrades by document one hundred thousand. He calls the phenomenon semantic decay, and much of Experio’s engineering, from knowledge graphs to retrieval discipline to keeping humans in the loop, exists to fight it. Accuracy, in his framing, is not a launch-day number but a property a firm must actively defend.
Cohen-Dumani describes the knowledge available to AI as three layers. The first is the internet, what every generic tool knows, impressive and identical for everyone. The second is the industry, its vocabulary, its regulatory weight, and the way work actually gets done in a vertical. The third is the firm itself, the engagements, precedents, and judgment nobody else possesses. The uncomfortable truth for buyers, he argues, is that competitive advantage lives almost entirely in the layer generic tools cannot see. Nor does he expect the frontier laboratories to close that gap on their own. Some problems live so deep inside one industry that a general model never reaches them, he says. A better base model makes vertical products stronger. It does not make them unnecessary.
The market timing is uncomfortable for the industry’s dominant business model. When work that took eight hours takes seconds, firms that bill by the hour face what Cohen-Dumani calls an existential question rather than a productivity upgrade. Firms that pool and retain their knowledge, he argues, will compound the advantage. Firms that let expertise walk out the door with every retirement will pay for the same lessons twice.
Where the value shows up, he says, is in areas firm leaders already watch. A firm can prove its experience in the room instead of asserting it. Obligations buried across ten years of contracts surface before they become surprises. And a junior hire can reach the firm’s full memory on day one instead of year five, without interrupting the partners who have become, in his phrase, the firm’s only search engine.
The onboarding math alone tends to catch the attention of managing partners. The most expensive part of a new hire, Cohen-Dumani observes, is not the salary but the years of context they lack, a ramp firms have simply accepted as the cost of growing a team. Give the first-year associate the firm’s memory on the first morning and the apprenticeship accelerates rather than disappears. Juniors move like veterans, and veterans stop fielding the same questions for the hundredth time in their careers.
He is careful, however, to draw a line under the automation narrative. Cohen-Dumani estimates AI will eventually handle 80 percent of consulting work: the research, the drafting, the finding. The remaining 20 percent: judgment, relationships, and the instinct that reads a room, stays human. “The goal isn’t to remove the human,” he says. “It’s to give the human their time back.”
None of this strikes him as radical, and that is precisely his point. For an industry built on knowing things, the next competitive edge may be deceptively simple. It is being able to find what you already know. Cohen-Dumani’s bet is that the firms that solve it first will own the decade, and that the ones waiting to be convinced will spend it catching up.
