The AI Savings Gap: Why Consulting Clients Are Still Waiting

What Soren Kaplan Gets Right, and Where I'd Push Back

I read Soren Kaplan's recent Inc. column arguing that consulting disruption has finally arrived, that once AI can generate a strategic plan or synthesize market trends in seconds, the rules of what clients pay for change completely. He's right about the direction. After nearly three decades in this industry, and the last few years watching AI move through dozens of engagements alongside consulting partners of every size, I'd push back on the timeline, and on what's actually being disrupted.

Buyer Expectations vs. Consulting Realities

Procurement has been told, by vendors, by analysts, by their own boards, that AI should be cutting consulting costs dramatically by now. Consulting firms have been told to adopt AI or get disrupted. The savings clients budgeted for still haven't materialized at the scale they were promised, and that's the gap: procurement asking where the savings are, delivery explaining why the number was never going to show up this fast.

It's not that the technology doesn't work. It's that the layer AI is automating, drafting slides, summarizing transcripts, producing a first-pass model, was never the expensive part of the engagement. Clients weren't paying a premium for the slide deck. They were paying for the meetings (human discussions) and the judgment behind the outputs: which of forty possible findings actually matters to this client, in this politics, with this budget, this quarter. That judgment hasn't moved to software. It has gotten faster to produce, because the person doing it now has AI doing the first draft. Yes, the meeting notes can be synthesized into a document, but the conversation still has to happen, the relationships still need to be built, the alignment achieved.

Quality, Not Diminishment

I want to be precise here, because it's easy to hear this argument as "AI can't really do the work." That's not what I'm saying. AI is genuinely good at what it does, and I use it every day, as does every team I work with. My point is about where the value sits. Clients aren't buying speed. They're buying a recommendation someone with real experience is willing to put their name on, and the context that comes from experts and AI working the problem together, not AI working it alone.

People are still at the center of these projects, whether the market wants to hear that right now or not. The firms doing this well aren't the ones with the flashiest AI demo in the pitch meeting. They're the ones pairing AI's speed with people who've sat through the last five versions of this exact decision and know what actually survives implementation versus what only looks good in a deck. That combination, not either half alone, is what a client is paying for. Quality of delivery is the whole argument, and it's the piece that gets lost when the conversation collapses to "how much will AI save us."

What the Data Actually Shows

This isn't a hunch. McKinsey's own State of AI survey found that while 88% of organizations report regular AI use in at least one business function, only about 39% can attribute any measurable EBIT impact to it, and most of those put the number under 5%. Enterprise-wide ROI is still out of reach for the large majority of companies despite record spending. That's the client side of the ledger. On the delivery side, Gartner has been blunt that AI-driven headcount reductions create budget room without reliably delivering returns, and the firm has separately projected that roughly 30% of generative AI projects get abandoned after proof of concept, citing poor data quality, weak risk controls, and unclear business value as the recurring causes.

Translate that into consulting terms: the savings procurement is expecting live mostly in the experimental phase. They're being piloted and re-piloted, not deployed at scale. Anyone telling a client otherwise right now is selling the pilot as the finished product.

The Costs the AI Pitch Deck Leaves Out

When a client asks why the savings haven't materialized, I point to three costs that rarely make it into the AI sales pitch.

• Upskilling, and it's not a lunch-and-learn. Getting a consulting team from "I used a chatbot to write an email" to "I can validate a model's output well enough to put my name on it" is a real training investment, measured in months, not a webinar.
• IT infrastructure, privacy, and security. A client's data can't just flow into a public model. Every serious engagement now needs a governed environment, access controls, data residency, audit trails, before a single AI-assisted deliverable can leave the building. That's infrastructure work, and it's slow on purpose.
• Bespoke build-out per use case. There is no single "AI for consulting" button. A due-diligence workflow, a supply-chain optimization model, and an org-design exercise each need their own tooling, their own guardrails, their own validation approach. Every one of those is its own build.
• AI can be expensive. In my personal experience and what I have observed with organizations executing large scale tests, they are seeing costs similar to human labor. It may be faster and more consistent, but that doesn't come cheap.

None of that is a reason to wait. It's a reason to stop pricing AI-enabled work as if the hard part is already solved.

This Isn't the First "Disruption Is Here" Headline

Consulting has been declared disrupted before. Offshoring was going to gut it in the 2000s. Then it was disintermediation, clients building internal strategy teams and cutting out the middleman. Then it was low-code and RPA. Each wave compressed some layer of the work, and none of them killed the industry, because each time the prediction conflated a task getting cheaper with the value proposition disappearing. AI is the biggest wave yet, and it's tempting to assume this time the whole model finally breaks. I think it compresses the same layer harder and faster than any prior wave, but it's still one layer, not the whole stack.

What the Layoffs Actually Prove

The counter-argument I hear most: look at the headcount cuts, isn't that proof AI is already doing the expensive work? McKinsey trimmed roughly 3,000 to 4,000 roles in late 2025 on top of earlier reductions; Bain, BCG, and Deloitte have all slowed hiring or cut staff through 2026.

Look at where those cuts actually landed: back-office functions, junior research roles, and practice areas where generative AI compressed a specific, well-defined task. That's real, and it matters to the people affected. It's also consistent with my argument, not a rebuttal of it. The automatable layer is the layer getting cut. The senior judgment layer isn't shrinking; if anything, firms are shifting investment toward people who can navigate the parts AI still can't: AI strategists, implementation leads, change-management specialists who understand both the technology and the client's politics. Forbes' framing of a "broken" business model is really describing a re-priced one: fewer people on the superficial work, no discount yet on the judgment work, because nobody has found a way to automate that part (yet).

What I'd Tell a CIO Right Now

If you're the one signing off on the consulting budget, don't ask your SI or your boutique firm "how much will AI save us." Ask three narrower questions instead: which specific task on this engagement will AI actually touch, what governed environment does that task require before it can touch your data, and what does the team validating that output actually look like. Vague answers mean you're being sold the demo, not the deliverable.

Push harder on the second question than most CIOs currently do. "We use an enterprise-approved model" is not the same as "we've mapped which of your data classifications are allowed to touch it, who's audited that mapping, and what happens when a consultant's prompt accidentally includes something it shouldn't." I've watched engagements stall for weeks at exactly that gap, not because the AI failed, but because nobody had done the unglamorous security work before AI was allowed near real client data.

Over the next twelve to eighteen months, expect the savings conversation to get more honest, not less, as these pilots either graduate to production or get quietly shelved. The firms that win the next round of RFPs won't be the ones with the flashiest AI demo. They'll be the ones with a governed environment already built, a validated use case already in production, and a team that's done the upskilling, because they started that build well before their competitors stopped denying it was necessary.

And if you're a consultant reading this hoping I've talked your client out of expecting savings, I haven't. They're coming. They're arriving use case by use case, behind a security review, after a training investment, not as a line item procurement can book this quarter. What doesn't change, no matter how good the tooling gets, is that the client is still buying people who know what to do with what the AI produces.

If you're wrestling with where AI actually fits in your delivery model versus where it's being oversold to your board, that's worth talking through before the budget gets set, not after. Reach out and let's talk it through.