Digital × Human: Digital Transformation Between Technology and People
The Thing Nobody Demos
There is a particular silence that follows a good AI demo. The model works, the room is impressed, and then someone asks the only question that matters: so who actually uses this on Monday morning, and what do they stop doing to make room for it?
That question is where most AI investment quietly goes to die. The technology was never really the bottleneck. The bottleneck is the messy, human, organisational distance between a capability that exists and a capability that gets used. We named the firm Digital × Human (DxH) because that distance is the entire problem we work on.
AI transformation rarely fails due to technology, it fails at the intersection of digital capability and human behaviour.
The Numbers Are Worse Than the Press Releases
It is easy to assume AI adoption is going well. Vendor keynotes and LinkedIn certainly give that impression. The data from companies actually running these systems tells a different story.
MIT's 2025 report The GenAI Divide: State of AI in Business found that roughly 95% of corporate generative AI pilots delivered no measurable return on the profit-and-loss statement. Not 95% that were technically broken. 95% that never moved a number anyone in finance cared about. S&P Global's Voice of the Enterprise survey reported that 42% of companies had scrapped most of their AI initiatives in 2025, up from 17% the year before. Back in 2024, Gartner had already predicted that at least 30% of generative AI projects would be abandoned after proof-of-concept by the end of that year.
Now hold those failure rates next to the budgets. The same MIT report describes tens of billions of dollars already committed to enterprise generative AI, and spending is still climbing. So it is not about lack of funding, and it is not about weak models. The models in these failed pilots are the same models powering the 5% that worked. What separates the two groups is not the technology they bought, it is what they did with the organisation around it.
Why the Gap Is Human, Not Technical
When we analyse failed or stalled initiatives, the cause of death is very rarely the algorithm. Three patterns come up again and again, and none of them are about model quality.
The first is ownership. A pilot is built by an innovation team or external consultants whose job is to start things. When the proof of concept ends, nobody whose performance review depends on the system actually inherits it. We have written before about how governance starts with accountability rather than policy, and the same logic applies to adoption. A system without a named owner does not survive its first quiet month.
The second is governance, or rather the absence of it where work is already happening. McKinsey's 2025 workplace research surfaced a revealing gap: executives estimate that around 4% of employees use generative AI for a meaningful share of their daily work, while the real figure was closer to 13%. The trend only runs one way: the share of employees using AI at work rose from 30% in 2023 to 76% in 2025. People want the help and are not waiting for permission, which means the practical choice is not whether AI gets used but whether it gets used inside guardrails. Governance that arrives after the fact is no longer governance. It is cleanup.
The third is workflow. A model that is technically excellent but bolted onto the side of someone's real job adds workload rather than reducing it. Adoption is most fragile in exactly the period when a new system is not yet polished, which is why we argue that workflow design matters more than raw model performance.
A PoC proves the technology can work. Adoption asks whether your organisation can make it work on a Tuesday, with real data, for people who did not ask for it.
What "Digital × Human" Means in Practice
The cross in Digital × Human is the point where the two things have to meet. Digital capability on one side, the people and processes that have to absorb it on the other, and the cross is where they are made to work together rather than in parallel. Plenty of programmes deliver only the "Digital", or the two halves separately: a tool is procured, a training session is booked, and everyone hopes the seam between them closes by itself. It does not work like this. What creates value is the joining of both aspects, digital and human, and that is the work we do at the crossing.
So we organise around six things that have to move together, not in sequence:
| Where the gap opens | What it actually requires |
|---|---|
| AI Strategy | A use case tied to a process and a number, not a technology looking for a home. |
| Operational Design | Workflows rebuilt around the tool, so it removes steps rather than adding them. |
| Rollout & Adoption | The sceptics engaged early, when their objections are still cheap. |
| AI Governance | Named owners, decision logs, and a clear authority to switch a system off. |
| Data Architecture | Readiness treated as a property of the plumbing, not a slide in the deck. |
| Program Delivery | The unglamorous second half funded and owned before the first demo. |
None of this is exotic, and that is the point. The reasons digital and AI transformations fail are mostly old, well-understood change-management reasons in a new environment.
The mistake is treating an AI rollout as a pure procurement, when it is really an operating-model change that happens to involve AI.
The Test We Apply Before Anything Else
Before we touch a tool, we ask a client three questions, and the quality of the answers usually predicts the quality of the outcome.
- Which business process is this meant to change, and by how much?
- Who owns the result at 3, 6 and 12 months, by name, with it in their objectives?
- What does done actually look like, and what does failure look like, clearly enough that we would both recognise either one?
If those answers do not exist, no amount of model selection will save the project. If they do exist, the technology choice should become secondary.
Why This Is the Half That Matters
The consulting market around AI is loud, and most of it sells the easy half of the problem. Selecting a model is the easy half. Standing up a tool like Cowork is close to trivial. The difficult half is getting a real organisation, with real incentives and real fatigue from the last three transformation programmes, to change how it works and to keep working that way once the launch energy fades. That is the Human half, and it is where the value in the MIT 5% actually came from. It is the half we chose to build a firm around precisely because it is the one the market keeps skipping.
So if your own AI effort has produced impressive demos and disappointing change, read that as information rather than failure. It tells you exactly where your gap is, and your gap is almost never the model. The organisations pulling ahead are not the ones that bought better technology. They are the ones that treated adoption as something to be designed, owned and measured, with the same seriousness they would give a balance sheet. Capability is now a commodity you can buy in an afternoon. Follow-through is the rare thing, and it is the thing we build.
Curious where your own adoption effort is strong and where it will stall? Our ADKAR AI Readiness Diagnostic gives you a structured read in about ten minutes.