A Consulting Firm’s Edge Is Its Own Knowledge

Key takeaway: In the AI era a consulting firm's edge is its own accumulated knowledge — and the best of it is locked inside confidential client work. Making that work genuinely safe to reuse is the deciding constraint, and it takes more than stripping client names: leave one confidential line in and your own AI can later surface it to a competitor's team with no identifier ever attached. Manually sanitised decks are routinely still re-identifiable by AI. Firms that solve this stop rebuilding work they have already done and compound an advantage every quarter; the rest carry a growing tax on their own growth.

Last updated: July 2026

In the AI era, a consulting firm's edge is its own accumulated knowledge: the frameworks, sector reads and past analysis a general model has never seen. AI multiplies whatever knowledge a firm can safely put behind it, and the deciding constraint is confidentiality, because the most valuable work sits in client-confidential decks. Firms that sanitise that work properly get far more usable IP behind their AI. Firms that cannot are stuck grounding AI in the safe scraps.

At the Knowledge Summit 2026 in Dublin this June, someone asked the room for a show of hands: who feels more important now than they did before AI? About one hand in twenty went up.

That is exactly backwards. Every consulting firm has made AI its defining priority, and AI runs on exactly one thing: the knowledge the firm has already built. Its frameworks, its sector reads, the way it structures a problem — the raw material sitting in twenty years of engagements. The people who look after that material should feel indispensable. Instead most feel written out of the story, because in most firms the AI programme is run from the CIO's office and the knowledge team was never in the room.

The 5% are wrong about their own value. But they are right that the job has changed, and the firms that understand how are already pulling ahead.

AI runs on what your firm already knows

A consulting firm's product has always been its accumulated thinking: the frameworks, the sector reads, the hard-won way of structuring a problem that a client cannot get anywhere else. The catch was that this knowledge only ever worked in fragments. A partner drew on the handful of past engagements they happened to remember; the rest decayed in folders no one reopened. For the first time, AI can in principle put the whole back-catalogue to work on every engagement at once — decades of the firm's judgement, available on demand.

That is what makes a firm's own knowledge the thing that matters now. A general model produces general work: it has read the public internet, and so has everyone. It cannot tell you how your firm sizes a market or frames a board conversation, because none of that was ever public — it lives in your past decks. A model grounded in that work produces something unmistakably yours. A model without it produces something a client could increasingly have generated themselves.

The research points the same way. McKinsey's work on AI adoption finds that the organisations capturing real value customise models with their own proprietary data rather than using off-the-shelf tools as they come — capturing the full value "often requires significant customization — for example, training models on proprietary company and customer data." (McKinsey, 2024.) An MIT study in 2025 found roughly 95% of enterprise generative-AI pilots delivered no measurable impact, with generic tools stalling because they "don't learn from or adapt to workflows." (MIT Project NANDA, reported by Fortune.) The edge is not the model. It is the knowledge you put behind it.

AI can only use the knowledge you have actually gathered

Here is where the knowledge team's job quietly became load-bearing. Grounding a model in your firm's work only helps if the AI can reach that work — and in most firms, it cannot, because the work is scattered across project folders, personal drives and the memories of whoever ran the engagement.

Before AI, scattered knowledge was a tax on productivity but a survivable one. A consultant who needed a prior market model could ask around, remember a deck existed, walk to a partner's office. AI cannot ask around. It can only reason over what has actually been gathered, connected and made available to it. Knowledge that has not been assembled is not slow to find — it is invisible. It may as well not exist. The unglamorous work of collecting, curating and connecting a firm's knowledge — the thing knowledge management has always struggled to justify on efficiency grounds — stops being a nice-to-have and becomes the precondition for the firm's AI to work at all.

You can see how firms are ducking this. Most AI initiatives are chosen to minimise the blast radius: point the model at the safest possible content — the staff handbook, published thought leadership, sanitised case studies — so that if it says something odd, nothing confidential is at stake. It is safe, it demos well, and it knows nothing that matters. The AI ends up grounded in the least valuable slice of the firm's knowledge while the proposals, interim analysis and primary research — the work that actually reflects how the firm thinks — stay locked in project folders. Moving AI off the safe scraps and onto the valuable, confidential work is the defining knowledge problem of the next few years. It is exactly the problem the knowledge function is built to own.

And it is a data problem before it is a model problem. Gartner expects 60% of AI projects lacking "AI-ready" data to be abandoned by 2026, and found 63% of organisations either lack, or are unsure they have, the data practices AI requires. (Gartner, 2025.) The models are ready. The knowledge feeding them is not — and getting it ready is not an IT task, it is a knowledge task.

"Clean" doesn't mean deleting the client's name

There is a catch that stops most firms before they start, and it is the reason the valuable content stays locked away. You cannot simply gather the archive and point AI at it, because the archive is full of client confidences. This is where most sanitisation efforts, and most AI-governance conversations, go wrong: they treat confidentiality as an identifier problem. Strip the client name, the logo, the brand colours, and the deck is "safe." Removing identifiers is now the easy part — a script or a general AI tool can do it — and it makes a deck look clean while the confidential substance walks straight out the door.

The information that actually does the damage usually carries no identifier at all:

• The benchmark leak. Client A's real cost-to-serve, margin or price point gets reused in Client B's deck as an "anonymised industry benchmark." No name attached — but you have handed A's confidential number to a competitor. If B can infer whose number it is, that is a serious breach.
• The confidential view. "Their leadership plans to exit Germany" becomes market colour in the next deliverable. No identifier, a straight leak of something shared in confidence.
• Strategy, unpublished financials, deal terms and management views — all confidential long after the client's name is gone.

This is exactly why identifier-only scrubbing is worse than it looks once AI is involved. An AI knowledge base does not need the name to leak the secret. Ask it for pricing benchmarks in telecoms and it will surface Client A's confidential figure as context for a Client B team — identifier-free, and a breach all the same. Detecting non-public information means understanding what a slide actually says in context, not matching a list of names. It is also what a CIO or risk officer will be asked to defend: when an auditor wants to know why an AI agent surfaced a particular number, "we stripped the logos" is not an answer. Getting this right is the hard part, and it is the part that decides whether your knowledge is genuinely safe to reuse or merely safe-looking.

Get this right and the advantage compounds

Now the part that matters at board level. A firm that makes its knowledge safe to gather and safe to feed AI does not just work faster. It starts building an advantage that grows on itself — one a competitor cannot copy, because it is made of that firm's own engagements. You are not putting a better tool in front of the client; you are putting decades of the firm's judgement to work in a way no one outside the firm can. That is what a premium fee is for, and it is what separates you in a competitive pitch.

Three things make that asset compound rather than simply accumulate:

• The base grows on itself. Every safely-reusable engagement re-enters the knowledge base, so the next team starts further along and adds to it in turn. The firm's institutional intelligence advances every quarter instead of resetting to a blank page. The laggard's stays at the starting line.
• AI lifts the ceiling. The old limit on drawing down past work was whether anyone remembered it existed. Point clean knowledge at AI and it surfaces the relevant prior work automatically, so far more of the firm's accumulated thinking is actually in play. The laggard's AI runs on the safe scraps and cannot be widened without leaking — it is structurally capped, and no budget fixes that.
• It converts into revenue, not just saved cost. The compounding base is what lets the firm hold its fees and win competitive pitches on something a rival cannot match. And the capacity it frees gets reinvested into the two things that create more future advantage: deeper client work — higher win rates, more engagements, more proprietary IP — and more reusable IP still.

The efficiency case is real too, and worth naming as proof rather than headline. Take one engagement — eight weeks, a team of four, roughly a thousand consultant-hours. A meaningful share of that, call it 30%, is re-derivable analysis the firm has almost certainly built before somewhere. Today firms draw down less than 5% of their past work, so nearly all of it is rebuilt from scratch. A firm that gets even part of that back recovers around a hundred hours on a single engagement — roughly $37,000 at a blended rate — and, crucially, that recovered work becomes raw material for the next one. We size the full cost of leaving it locked away — the knowledge tax — separately; for a 2,000-person firm it runs to tens of millions a year.

The laggard, meanwhile, does not hold steady. The tax on locked-away knowledge grows as content volume outpaces manual sanitisation, so a firm standing still slides backwards in relative terms. The gap widens from both ends. It is the same dynamic the wider evidence keeps finding: BCG reports that 74% of companies have yet to show tangible value from AI while a small group of leaders pull clearly ahead (BCG, 2024), and McKinsey now measures the leader-to-laggard gap widening through a loop in which "faster experimentation generates more data, more data improves models and decision quality, and better performance attracts more users and activity" (McKinsey, 2025).

The one variable that decides it

Strip it back and the deciding variable is simple: how much of the firm's accumulated knowledge is actually working — on the next engagement and behind the firm's AI — rather than sitting locked in folders. And that is gated by one thing: whether the knowledge is clean enough to move. Not clean as in name-stripped; clean as in free of the non-public substance that turns reuse into a leak.

Which brings the knowledge team's role back into focus. The 5% who felt more important had it right; the rest are measuring the old job — files tagged, searches served — instead of the new one. The new job is bigger: to gather the firm's knowledge, make it safe, and put it to work behind the AI so the firm's output stays unmistakably its own. That is not a support function. That is what the firm's fees increasingly rest on.

AI does not create the advantage. It multiplies whatever knowledge a firm can put behind it. The firms that can safely put their real knowledge to work compound a lead every quarter. The firms that cannot repeat work that already exists, feed their AI scraps, and watch their edge thin out. The gap between them is already widening.

We have written more on the specific pieces of this: the confidentiality wall that blocks consulting AI, why generic tools can't sanitise consulting decks, and the knowledge tax firms pay when their best work cannot be reused.