November 12, 2026
–
0 minutes to read

AI in consulting: what's actually changing (and what isn't) in 2026

Ellen Darbyshire

Senior Content Marketing Executive

Ellen leads the marketing team’s content strategy. With a background in B2B and SaaS marketing, she specializes in crafting compelling content that enhance brand messaging and drive audience engagement across multiple platforms.

AI in consulting has moved from pilots into everyday delivery. SPI Research found generative AI in 27.1% of professional services projects in 2025, and 40% of firms now sell AI-related services. Yet firms still get paid, staff work and carry risk the way they did before. That gap is where the margin pressure comes from.

Where AI is genuinely changing consulting delivery today

Most firms now use AI. The question is how deeply.

Organization-wide AI use in professional services nearly doubled in a year, from 22% to 40%, according to the Thomson Reuters Institute's 2026 AI in Professional Services Report. In the UK, the MCA found 77% of consulting firms have either integrated AI or enabled their people to use it.

Consultancy BenchPress 2026, the annual benchmark from The Consultancy Growth Network and The Wow Company, shows how far UK firms have got. Only 1% aren't using AI at all. Just over a quarter (28%) are running pilots. More than half (55%) are using it successfully in parts of the business, and 15% have rolled it out across the whole firm.

The commercial stakes are bigger than the adoption figures suggest. SPI Research's 2026 Professional Services Maturity Benchmark projects that AI will account for 22% to 37% of future professional services revenue. A new line of business is forming inside the industry while AI changes how existing work gets delivered.

Four changes stand out.

Research and first drafts are much faster

Desk research, synthesis, first-pass analysis and slide production used to fill a junior consultant's week. Now they can take an afternoon. Clients have the same tools, and they know this. Firms tend to underestimate that part.

Internal knowledge is becoming a usable asset

Several firms have built retrieval systems over their own material. Scott Walter, VP of Technology at SDG, described the thinking on Consulting Magazine’s webinar with CMap, “AI isn't killing consulting, it's crowning new champions”:

‍"We recognized very quickly that we have what we like to call tribal knowledge. We have lots of internal knowledge around engagements that we've done, practices that we run, partnerships that we have."

The knowledge was always there. Now people can ask it questions.

Smaller firms are moving faster than large ones

‍ Professor Joe O'Mahoney of Cardiff Business School has researched the consulting industry for well over a decade. He's watched the big firms' proprietary models fail to stick:

‍"The use initially was very high because everyone was very excited and they found out it was useless, and now pretty much nobody uses them regardless of the proprietary LLM that they've built."

New entrants are competing on price‍

O'Mahoney again, on who's coming into the market: "AI first companies coming in at very low price points, still with the human in the loop." They're starting with smaller, well-defined pieces of work. That's enough to reset what buyers expect to pay.

‍

Where the hype outpaces the reality

Some things have barely moved.

Measurement

‍ Only 18% of professional services organizations track the ROI of their AI tools, according to Thomson Reuters. That's roughly the same share as the year before, even though adoption nearly doubled. SPI Research reports that time to ROI on AI investment has nearly doubled in some cases. Firms are running into the complexity of adoption, and the technology itself is rarely what holds them back.

UK consultancies show the same gap. In BenchPress 2026, the UK consulting benchmark CMap sponsors, 56% of consultancies say AI has cut their cost of delivery. Only 6% report a clear, quantifiable positive impact on their finances. Another 40% call the impact positive but hard to put a number on, and 32% don't know.

Quality control

‍Tristan Grinnell of Crosstide (formerly 101 Ways) described the risk this way:

‍"Because it's such a powerful technology, the answers it gives look plausible, but plausible doesn't necessarily mean that they're right. And there's an interesting paper about the number of times people will check diligently the output of something that looks coherent, looks plausible, and it's only about five times. After five times, it tails off."

Whatever the exact number, people check carefully at first and then start trusting output that looks right. On a client deliverable with your name on it, that's a real risk.

What clients are buying

‍Christian Barnard, COO and Chief Growth Officer at KungFu.AI, is about as pro-AI as a consulting leader gets: ‍

‍"In the midst of all the AI frenzy, it's actually reassuring to me how consulting fundamentals have not changed. At its heart, consulting is still a people business. It's about trust and problem-solving and delivering results for our clients."

The industry numbers back him up. If AI were improving firm economics across the board, you'd expect to see it in utilization and margin. SPI's 2026 benchmark, covering 509 firms, found billable utilization fell to 66.4% in 2025. That's the lowest in the survey's 19-year history, down from 73.2% in 2021. Revenue per billable consultant rose to $210,000 from $199,000, and revenue growth improved to 5.2%. EBITDA sat at 9.9%.

Faster delivery of work you haven't repriced shows up as a smaller invoice. The client keeps the saving.

‍

What this means for staffing, pricing and utilization

Pricing is the pressure point

Most AI commentary skips pricing. It's also the issue most likely to decide whether 2027 is a good year.

If your fee is anchored to effort and effort is falling, your revenue falls with it. IBM's Institute for Business Value surveyed 1,000 senior technology, operations and finance leaders across 13 markets for a report on application management services, published in August 2026. Of those leaders, 83% said AMS contracts still reward labor. And 71% said contract structure is now a bigger barrier to AI than the technology itself. The same pattern applies to consulting: pay for effort, and AI savings flow straight to the client.

Buyers want change, but few have acted on it. HFS Research and EY surveyed 304 Global 2000 executives in July 2026. Most (72%) are comfortable with outcome-based pricing for AI-led delivery, but only 39% have used it in the past two years. On average, they expect AI to cut their costs by around 19%.

Both sides are waiting for the other to move. The firms that go first will need evidence of what their work really costs to deliver.

Micheline El Housseini Timbrell, Founder and CEO of Phi Management, has already made the call on part of her portfolio:

‍"In that OD you have a solution like, from scratch, job analysis, job descriptions, policies and procedures. All that, I believe the clients can do them now with AI, with all these platforms, so no need to outsource or to pay for this kind of service. So in a few months, I think we will not even have that as part of our offerings."

Retiring a service line is a big decision. It needs numbers behind it.

Staffing: the pyramid is thinning from the bottom

The Big Four have cut UK graduate intakes. According to analysis by The Telegraph published in June 2025, KPMG's intake fell 33%, Deloitte's 18%, EY's 11% and PwC's 6%. McKinsey's headcount fell by more than 10% over 18 months to around 40,000, as reported in May 2025. The firm put that down to attrition and performance exits rather than layoffs.

The wider picture is mixed. PwC's 2026 Global AI Jobs Barometer looked at US entry-level job openings. Openings for AI-exposed entry-level roles grew 35% since 2019, while other entry-level openings fell 10%. The squeeze hits some roles much harder than others.

Juniors have always learned judgment by doing the work AI now does. Firms that stopped hiring them will still have to solve that later.

Utilization stops being the number that tells you anything

Falling utilization alongside faster delivery makes sense. The same demand gets met in fewer hours, with the plan unchanged.

That makes utilization a weak headline metric in 2026 and a useful diagnostic one. Look at it by person and by practice, over a rolling forward view. It shows where freed-up capacity has gone. A single firm-level average in a quarterly pack won't give you anything to act on. ‍

‍

How firms are adapting their operating model

The firms handling this well tend to share four habits.

They start with one or two named problems. Each problem has a cost attached, plus a clear before and after.

They point AI at their own operations as well as client work. That means looking at scope creep patterns, pricing by engagement type, and who delivered the work that made money. It's unglamorous, and the gains build over time.

They treat the data underneath as the real project. AI amplifies whatever it's built on. Point it at a resourcing spreadsheet and a month-end pack, and it'll give you confident answers about a picture that's three weeks old. That's why the order matters more than the tool. Connected, governed data comes first. Workflows with AI built in come next. Open access for the rest of your stack follows.

SPI's maturity model puts a number on what that foundation is worth. Level 5 firms grew revenue 23.2% in 2025, against 1.7% at Level 1. Their project margins were around 250% higher and billable utilization 42% higher. Around 30% of firms sit at Level 1 and only 5% reach Level 5.

They keep a human accountable. Jeff Plowman of Collective 54 has worked with around 250 boutique founders. He describes where things are heading:

‍"We're calling this next era, era three, or the age of AI-delivered professional services, where about 80% of the service and the whole operations of the business is delivered by technology. But the real critical part is the 20% of the human in the loop."

Steve Hoeft of Pivotal Advisors explained why that human part matters on our podcast. When a person gives you an idea, "there's a human indebtedness". When a model gives you one, "it's just text on the screen". Clients pay for that sense of obligation.

That's the layer CMap builds. CMap is the firm intelligence platform for consulting firms. Pipeline, resourcing, time, delivery and billing are connected in one place, with AI that flags what needs attention while you can still act. Your data stays yours. It works inside your permissions, and nothing you store trains anyone else's models.

If you're deciding where to start, start with the number you most often get late.

FAQs

What is AI in consulting?

AI in consulting covers two things. One is AI that firms sell to clients as a service. The other is AI that firms use to do their own work. Most of the measurable change in 2026 is in the second group, in research, drafting, analysis, knowledge retrieval and internal operations.

How many consulting firms are using AI?

Organization-wide use in professional services nearly doubled from 22% to 40% between 2025 and 2026 (Thomson Reuters Institute). In the UK, 77% of consulting firms have integrated AI or enabled employee use (MCA, 2026). SPI Research found generative AI in 27.1% of projects delivered in 2025.

Is AI making consulting firms more profitable?

Not yet, on average. Industry EBITDA sat at 9.9% in 2025 and billable utilization fell to a 19-year low (SPI Research). In the UK, 56% of consultancies say AI has cut their delivery costs, but only 6% can point to a clear financial impact (BenchPress 2026). The gap comes down to pricing and measurement.

Will AI reduce consulting fees?

It already has at the commoditized end, where clients can now get an adequate answer themselves. Work priced against outcomes is holding up better than work priced by the hour. That's why the pricing conversation has become urgent.

What should a mid-sized firm do first?

Pick one internal problem with a cost attached. Fix the data underneath it, and train people properly. Firms that start with a tool usually end up with several tools and no change they can point to.