Ask most people in AI Enablement where buyer budgets are going and they'll say tokens, tools and maybe training. The buyer data tells a different story. When we asked 100 mid-market and enterprise buyers where they would allocate a hypothetical unlimited budget, 50% said technical integrations and 31% said workflow redesign. That's 81% pointing to infrastructure. Workforce training came in at 15%. Change management, arguably the hardest part of AI adoption, at 4%.

That is not where the money has gone. It's where buyers wish it had gone. The gap between the two is where the next wave of services opportunity lives.

Fully 87% of enterprises we surveyed report buying AI training services. But those same buyers point to technical integration and workflow redesign as their biggest unmet needs. 

What happened, in most cases, is that an event was purchased when a practice was needed. They ran workshops, completed modules, and ticked the right boxes. But capability doesn't emerge from a point-in-time intervention. It accumulates from repetition and feedback inside the actual workflow - the kind of practice that changes the defaults people reach for when they're under pressure. That change mostly didn't happen.

The consequence is predictable. Organisations who have nominally "done training" arrive at the next phase without the necessary foundation. Integration work assumes AI fluency in the workforce. Workflow redesign assumes people have actually changed how they approach tasks. If neither is true, more expensive interventions land on unstable ground.

Anecdotally, we see two types of organisations making genuine progress. The first are the "really committed": mid-size companies led by a CEO who has made AI adoption a top priority. They scope out different programs, are prepared to experiment to see what works,  and they treat AI fluency as a job requirement. The second are the lucky ones: smaller businesses where an enthusiastic manager has been tinkering with AI and has enough organisational capital to get experiments taken seriously. Both types are the exception. Vendors who can identify them early can deliver referenceable projects from this pool.

When 50% of buyers say they'd like to spend more on technical integrations, it seems to be a clean market signal. I suspect it's more complicated than that.

For most organisations, the integration wall is not a technology problem. It is where AI adoption collides with lack of preparation: data quality issues that hands-on training could have surfaced, or organisational behaviours that should have shifted before anyone tried to build integrations on top of them. The wall is real. But it is often a symptom of what wasn't resolved earlier.

This creates risk for providers selling AI integration. The rollout will likely underdeliver, leaving a frustrated client looking for another way to fix the problem. If you want to  build a genuinely defensible business, you need to be the one willing to challenge clients earlier, and be honest about the operational foundations that need to exist before any integration can succeed.

When we speak with AI enablement firms focused on implementation, many (often the best ones) sheepishly admit that half their projects are not actually AI but basic data cleansing and preparation. Work that got deprioritised for years is now finally getting done because AI has made it politically palatable. But it’s these types of projects, coupled with on-going training, that are most likely to lead to positive ROI.

31% of buyers in our survey would allocate the largest share of unconstrained budget to workflow redesign. This finding didn't get the attention it deserved when we published it.

Workflow redesign is not a more expensive version of training or a more complex version of integration. It is something structurally different: rethinking how tasks are structured, how decisions get made, how outputs are defined, how roles interact - with AI as a core input rather than an add-on. It requires domain understanding deep enough to know what the work actually is and change management capability to navigate human resistance. Brett Taylor, CEO of Sierra and Chairman of OpenAI, spoke about it extensively in his Cheeky Pint interview. 

Almost no vendor in the current market is purpose-built for this. Training providers lack implementation resources. System integrators touch the technical layer but not the human one. Strategy consultants diagnose but rarely stay to implement. After speaking with more than 100 founders of AI enablement businesses, we have yet to find a firm covering this requirement fully. Some firms are moving in this direction. None has mastered the transition yet. Those who get there fast will have an unfair advantage.

Change management came last at 4% of unconstrained budget allocation. The obvious read is that buyers don't value it. The more useful read is that buyers don't connect their AI adoption problems to change management as a category.

This matters because the problems buyers are actually pointing to - difficulty translating AI into practical use cases (28%), lack of internal capability (12%), underperformance of existing AI tools (6%) - are fundamentally change management problems that appear to be technical ones. Organisations that have tools, have done training, but are now stuck are stuck because their people still work in the “old way”, not because of what systems they have. But buyers are consistently reaching for technical answers to what are essentially organisational and behavioural challenges.

If I were running an AI Enablement services firm today, I would definitely be building a change management practice (or partnering with someone who delivers this service). Buyers won't come looking for change management. They'll come looking for integration help or workflow redesign. You can call it change management or simply customer success.  If you can deliver the organisational layer while your client believes they're buying the technical layer you will build the kind of relationships that generate the 89% expand-with-existing-vendor loyalty we see among the market's best performers.  

You can download the full survey for free here.

We track approximately 650 AI enablement services companies. The majority are AI-native, bootstrapped or seed-strapped, and built around the service layers where demand has been easiest to capture. Very few have developed genuine capability to deliver workflow redesign or the kind of embedded integration work that buyers say they need most.

The reason is understandable. Demand has been strong enough that branching out felt unnecessary. Moving upmarket is painful. It means different people, a different way of selling, and a period of slower growth while new capability matures. When existing projects are closing easily, that trade-off is a hard one to make.

What also helped is that AI purchasing has largely bypassed normal procurement. A CEO hears a compelling speaker at a conference, gets excited, and signs off without an RFP. We have heard this story many times. It will not last. As AI spend matures it migrates back into standard procurement routes, and the vendors already inside an organisation become very hard to displace. Getting a new vendor approved takes months. Staying in just takes doing good work.

The firms that remain stuck at the training layer inside clients who now need integration and workflow capability will hit a profitability ceiling. They end up spending heavily on new customer acquisition rather than growing with existing ones. The relationship asset they worked hard to build stops paying dividends.

The best firms we speak to are looking to acquire or partner their way toward upstack capability rather than hire slowly toward it. How very different from the typical SaaS playbook! But it’s definitely the right move in this market based on how fast it’s growing and the fact that they are many founders out there who got into the services business by accident (and would rather sell and move on to their next product company).

The unconstrained budget data is less a prediction of future spending and more a signal of where buyers are feeling the most friction. Integration and workflow redesign sit at the top of the list not because training no longer matters, but because training on its own has left many organisations staring at a barrier they still don’t know how to move past.

The providers best positioned for the next wave of AI enablement spend will be the ones that help buyers connect the dots across training, integration, and workflow redesign. The opportunity lies in framing these challenges as parts of the same transformation problem, rather than treating them as separate purchasing decisions.

– Daria