73% Use AI Regularly, 10% Call It Core: Why Adoption and Transformation Split

73% of the AI decision-makers Publicis Sapient surveyed in 2026 say AI is in regular use, and only 10% say it is core to how the business runs. Deloitte found the same shape a few months earlier: 66% report productivity gains, 20% report revenue growth. Enterprise AI adoption is real. It is also a different measurement from transformation, and the two surveys show exactly where one stops.

This post is written for the founder or director who has to tell a board whether the company is "behind on AI", because that reader converts on knowing which benchmark measures what, and which rung their own company stands on.

The short answer. The gap between 73% and 10% is not a contradiction. Regular use measures how widely AI tools are switched on. Core measures whether the business would stop working without them. Deloitte's 66% and 20% split the same way: productivity gains come from internal use, revenue needs a change customers can see.

Two enterprise AI adoption surveys, one consistent pattern

Publicis Sapient released its Global Enterprise AI Report at VivaTech in Paris on 17 June 2026. The headline was blunt: most enterprises use AI, but only 10% say it is core to their operations.

Deloitte published its State of AI in the Enterprise report at Davos on 21 January 2026. Its own summary says revenue growth "largely remains an aspiration", with 20% already growing revenue through AI and 74% hoping to.

Each number gets quoted alone, usually to make one of two arguments. The first is that AI adoption is nearly universal and laggards are in danger. The second is that AI is a hype cycle with little to show for itself.

My position is that both arguments misread the same data. The surveys are not measuring one thing badly. They are measuring two things, adoption and transformation, and reporting both in the same percentage format.

Read together, they describe a ladder. Many firms stand on the bottom rung, about a third have climbed into real process change, and a small minority sit at the top. That shape holds across both surveys, which is the most useful thing either of them tells you.

Who the Publicis Sapient and Deloitte surveys asked

Before comparing any two percentages, check whether the denominators describe the same population. These two are close in kind, but not identical, and the differences push in a predictable direction.

Sources: Publicis Sapient press release, 17 June 2026. Deloitte press release, 21 January 2026, and Deloitte report landing page. Neither sponsor published the full questionnaire alongside these releases.

Both samples are selected for AI involvement

Every Publicis Sapient respondent holds responsibility for evaluating, influencing or selecting enterprise AI technologies. Deloitte describes its panel as leaders "with direct involvement in their companies' AI initiatives". Neither survey asks a random manager whether AI is in use.

That screen matters most for the adoption figure. A person whose job is choosing AI platforms works at a company that has AI platforms to choose. The 73% is therefore a reading of adoption among firms where someone already owns the question.

The fieldwork windows are eight months apart

Deloitte's survey closed in September 2025. Publicis Sapient's ran in late April and May 2026. In a market where Deloitte says sanctioned worker access to AI rose from fewer than 40% to around 60% of workers in one year, eight months is long enough to move adoption numbers.

It is less likely to move transformation numbers. Rebuilding a process takes longer than rolling out a licence. That asymmetry is part of the argument below, and it is also part of the case against it.

Enterprise AI adoption measures breadth, transformation measures depth

The 73% and the 10% sit in the same press release and come from the same 1,550 people. They cannot both be accurate unless they measure different properties. They do.

Regular use is a frequency answer

Publicis Sapient's wording is that 73% say AI is "used regularly or across most business processes". That phrase joins two claims with "or". Regular use by some teams qualifies. So does wide coverage of processes.

Frequency is the easy bar. A sales team drafting emails with an assistant every day is regular use. A finance team summarising contracts each week is regular use. Neither changes what the business sells or how it is organised.

This is why adoption numbers rise quickly. Buying seats, switching on a copilot in an existing suite, or approving a chat tool for staff all count. Each is a procurement decision, and procurement moves on a quarterly cycle.

Core is a dependency answer

"Core to how the business operates" asks something different. It asks whether the operating model now depends on AI. Remove the tools and a core business stops, while a business with regular use just slows down.

I think "core" is the most honest single question in either survey. It is hard to answer yes without evidence, and the respondent knows it. Only 10% did, and I would treat that as a ceiling on true AI dependency among large firms rather than a floor.

The country detail supports the reading. In Germany, 35% describe AI as a "colleague" used by teams, while 10% say it is fully integrated enterprise-wide. In the UAE, 60% report AI coordinated across teams and only 5% report full integration. Team-level use and enterprise dependency diverge in every market where both are reported.

Deloitte's 66% and 20% measure two different AI outcomes

Deloitte's gap is about results rather than usage, but it has the same structure. On its report page, Deloitte says improving productivity and efficiency tops the list of benefits achieved so far, "with two-thirds (66%) of organizations reporting gains". Increasing revenue sits at 20%.

Productivity gains arrive inside the building

Productivity is the outcome closest to the tool. A drafting assistant saves an analyst an hour, and the analyst notices the hour. No customer has to change behaviour, no price has to move, and no product has to ship for the gain to register.

That makes productivity gains easy to report in a survey. They are also easy to overstate, because time saved is often not time redeployed. A firm can report productivity gains for years without a matching change in output, a pattern covered in the analysis of AI time savings against hours worked.

Revenue needs a change customers can see

Revenue only moves when something reaches the market. That means a new product, a faster service, a better price, or more customers reached per salesperson. Each of those requires the deeper change Deloitte calls reinventing core processes or business models.

So the 20% is not a separate finding from the 34% who are deeply transforming. It is roughly the share you would expect to see revenue, given that only about a third are making the changes that produce it. The 74% hoping to grow revenue later is a statement of intent, and I would not put it in a forecast.

This is a different gap from the one between revenue and profit. A firm can grow revenue with AI and still lose money on the programme. That separate question, covered in the comparison of NVIDIA's revenue figure with McKinsey's EBIT figure, sits one step further along the same chain.

The AI transformation ladder: where both surveys agree

Lay the two reports side by side and an agreement appears that neither headline mentions. Deloitte sorts its respondents into three groups, and the boundary between the bottom two and the top one sits close to Publicis Sapient's transformation figure.

Deloitte reports that 34% are using AI to deeply transform, 30% are redesigning key processes, and 37% use AI at a surface level with little or no change to existing processes. Publicis Sapient reports that 38% say AI is fundamentally changing how their business operates today.

Two different panels, two different question sets, fieldwork eight months apart, and both land between 34% and 38%. I take that convergence more seriously than any single headline in either report. Independent instruments agreeing is the closest thing to corroboration a survey reader gets.

The 10% is the top rung, not the whole story

If one third is transforming, why do only 10% call AI core? Because "deeply transforming" describes a direction and "core" describes an arrival. A company can be rebuilding three processes around AI while its other forty run as before.

Deloitte's production data shows the same lag. Only 25% have moved 40% or more of their AI pilots into production, and 54% expect to reach that level within three to six months. Most firms are still converting experiments into running systems, and a firm cannot depend on what it has not shipped.

The production step is where most programmes stall, and the specific tests that separate a pilot from a production system are set out in the guide to AI proof-of-concept readiness tests.

Why the AI transformation gap is organisational, not technical

Both sponsors locate the blockage in the same place, and it is not the model. Publicis Sapient reports that 47% believe AI is already capable of meeting today's business needs, and 42% say AI is capable but their organisations are not set up to capture its value.

22% name the way their organisation operates as the primary barrier to AI success. In the US, 34% point to organisational structure. In France, 51% cite internal data. Those are three different blockers, and none of them is solved by buying a better model.

Deloitte's findings rhyme. Its report calls the AI skills gap the biggest barrier to integration and notes that education, not role or workflow redesign, was the top way companies adjusted their talent strategies. Training people to use tools inside unchanged workflows is precisely what produces adoption without transformation.

Governance is lagging in the same way. Among companies planning agentic AI, only 21% report a mature model for agent governance, while close to three quarters plan to deploy agents within two years. Adoption intent is running ahead of the controls that would let AI carry real operational weight.

My reading is that the 73% to 10% drop measures organisational design debt. The tools have been bought, the training has been delivered, and the org chart is unchanged. Where AI change programmes stall on people rather than technology, the analysis of passive resistance in AI change management covers the behaviours to look for.

Where this reading of the surveys is weakest

I have argued that the gap is a ladder of different measurements, not a contradiction. Here is where that case is thinnest, on the same standard I applied to the surveys.

Neither questionnaire is public in what I could open

Publicis Sapient's release gives the 73% and 10% but not the question text or the full answer scale. If both figures come from one single-choice maturity question, they are rungs of one ladder and my reading is strong. If they come from separate questions, respondents may have interpreted "core" in ways I cannot see.

Deloitte's methodology is similarly summarised. The press release names the sample, the countries and the AI-involvement screen, but I could not extract the full methodology appendix from the report PDF during research. The 34%, 30% and 37% groups are Deloitte's own segmentation, and the cut-offs behind them are not stated in the release.

The samples start high up the adoption curve

The US Census Bureau's Business Trends and Outlook Survey asks firms whether they used AI "in any business function" in the past two weeks. In data up to 3 May 2026, the national rate was 19.8%, and 37% among firms with at least 250 employees. That is an official statistic with a random business sample.

The Census measure is not comparable to the Publicis Sapient one. It is a different question, a different respondent, and a two-week window. It does show how much the panel screen does: even the largest US firms report AI use at roughly half the rate of the Publicis Sapient panel. If you are a mid-sized company, neither consultancy's adoption figure describes your peer group.

Both sponsors also sell transformation services. A persistent gap between adoption and transformation is the problem their consulting arms exist to close. I do not think that makes the findings wrong, but it does make the framing convenient, and a reader should hold both thoughts at once.

How to place your company on the AI transformation ladder

The surveys are most useful as a measuring stick for your own company. Each rung below is defined by evidence you can check internally, not by how a survey respondent feels.

The ladder does not argue that every company should reach rung 6. For many businesses, solid productivity gains at rung 2 or 3 are the right place to stop. The mistake is reporting rung 1 activity to a board as if it were rung 4 progress.

The ladder also changes what a sensible AI metric looks like at each stage. Usage counts suit rung 1, task timings suit rung 2, and only financial measures make sense from rung 5 upward. The shift between those metric types is traced in the analysis of how AI ROI metrics are moving toward productivity.

If your company is stuck between rungs 2 and 3, the blocker is usually ownership. Somebody has to own a redesigned process end to end, with authority to change roles and budgets. A central AI team rarely has that authority, a pattern described in the piece on AI centres of excellence as a bottleneck.

Frequently asked questions

Where to start this week

Take the ladder table to your next leadership meeting and ask each function head one question: which rung is your best AI use case on, and what is the evidence? Write the answer and the evidence side by side. Any rung claimed without the evidence column filled in goes down one level.

Then pick the single use case nearest to rung 3 and name one person who owns its process redesign, with a date. Adoption will keep rising on its own. Transformation only moves when someone is accountable for it.

Related reading