The three dimensions of AI: where should an engineering firm invest?
"We have bought AI licenses for the whole staff."
"We have several AI pilots running across our business units."
"We are completely rebuilding the delivery of our core service."
All three lines can be heard in management meetings, and all three get reported as AI investments. In practice they are different things: the first evens out how AI is used across the staff, the second aims to improve selected processes, and the third rebuilds the business. Each has its own people, its own owners, its own time horizon, and its own measures of success.
We use this three-way split ourselves when we assess our own AI investments and those of our clients. It is a portfolio view, not a ladder: all three can run at the same time, and the real decision is the balance between them.
The three dimensions in one picture
Horizontal AI raises the shared floor: the skills and everyday tools of the whole staff. Vertical AI aims to build local peaks in selected workflows. Strategic AI advances through the company as a front, replacing the old way of working and creating entirely new business opportunities.
Horizontal AI evens things out
For every company that intends to keep running most of its business on the work of its people, broad skills are a necessary foundation. When people can use the tools and judge their output, many small problems get solved without a separate project.
There is a trap in the metrics, though. License counts and usage rates mostly measure activity in chatbot-style tools built around one-off prompting. The impact comes from lasting, improved routines, and it may never show up in the metrics being tracked.
A typical organization splits roughly into three kinds of users. Most people are genuinely interested, and their everyday experience alternates between small wins and disappointments. Some are cautious, usually for good reasons: quality, security, and accountability are real questions. And a small group has already automated part of its own work and tells no one.
The silence of the front-runners is understandable. Reporting your time savings easily means more work. Sharing your method means becoming the trainer. If the tool was never approved, the conversation gets awkward. A company can therefore hold more AI skill than it can see, and learn less from it than it could.
More useful measures are these: is the shared level rising, is the gap between people narrowing, and do individual discoveries become house practice?
A vertical pilot is a long way from anything permanent
Vertical AI targets a specific workflow within a business: a project delivery, for example, or the part of it that deals with documents, calculations, information retrieval, or checking. This is the most visible part of the AI discussion. A bounded process is easy to measure, and a pilot is easy to present.
A pilot is certainly a good way to learn. There is still a gap between a pilot and a permanent way of working. A demo succeeds on a clean example case; real projects come with exceptions, missing input data, and deadlines. One project may run five percent better than the previous one, but if nobody owns the process after the pilot, the gain has evaporated by the next project and the tool ends up in the archive.
The working rule is simple: the team that uses the solution helps build it, owns the result, and keeps developing it. Handing over input data for the pilot is not enough of a role for the expert, and the developer cannot deliver a system tested on one project and move on. Lasting results come from choosing pilots so that the people doing the work are trained to be users, maintainers, and further developers of the system. Everything else is wasted work.
The strategic lane: the business built from scratch
Strategic AI starts from a different question: if a new competitor got hold of our expertise and our data, how would it build this business from scratch with today's technology? The question has an immediate follow-up, often felt as an urgent one: should we not build it ourselves, and before anyone else? Why are we not doing it?
The work needs its own team with software, data, and AI skills, plus the best subject-matter experts in the house, even though they are the busiest and the most valuable for billable work. It also needs real access to data, the authority to cross organizational boundaries, and permission to challenge the current operating model.
Wherever such a front advances, earlier aids and pilots lose their meaning, because the work itself has been redesigned. This does not necessarily concern the whole company: the front moves through selected areas, and elsewhere the work continues to rest on people (who are also needed to deliver what the automated processes produce). In some companies, though, the strategic lane may be the chosen main strategy. Perhaps at exactly that competitor which is not yet even visible in the market?
Our own assessment is that long-term competitive advantage is most likely born in the strategic dimension, because the advance of technology itself will wipe out the gains from smaller investments faster than people tend to assume. Attention and money still mostly go to the horizontal and vertical lanes, because those are easier to buy, measure, and present.
The balance is the real decision
Even if a company will not or cannot focus on the strategic lane, the dimensions still support each other. A rising level of skill can also reduce the need for separate pilots. A successful pilot produces data, integrations, skills, and new ideas that strategic building later draws on. And wherever the strategic front reaches, it absorbs the other two.
What remains to decide is the ratio: how much broad skill, how much process improvement, how much rebuilding. In most organizations the weight currently sits firmly on the first two. It is also understandable that in people-centred expert work, dismantling your own operating model midstream is a radical step. That is why we will probably see separate companies set up to rebuild operations from scratch without disturbing the existing organization.
The target moves, so ownership decides
For a long time, the standard pattern of buying technology was clear: define the problem, tender the solution, deploy it, and use it for years. In AI-critical workflows the pattern breaks down, because models, interfaces, and costs change while the project is still running. These days practically every larger system project is outdated by the time it is finished, and it needs continuous further development.
The most durable part of the acquisition is the ability to keep developing the workflow: a permanent owner, a plan that keeps updating, and permission to shut down a solution that does not work. Partners, platforms, and off-the-shelf software are still worth using, and there is rarely a reason to build a deeply customized solution for every workflow. The ownership of the problem, the data, the architecture, and the know-how gained should still stay in your own hands. The riskiest combination is a rigid custom system and deep vendor dependence in an area that keeps changing.
What does not change
As technical production gets cheaper and more common, the value of the bare output falls. Collecting tacit knowledge alongside other data, specification-driven AI-based software development, and similar methods will grow in importance, but keeping the most capable experts in the house matters just as much. The customer will continue to buy judgment, accountability, defensible reasoning, trust built over time, and local presence. Even with the same tools, the deal is often decided by the expert who answers the phone, understands the situation, and takes responsibility. AI strategy is therefore best anchored in the parts of the business that are hardest to copy.
Three questions go a long way when assessing any AI initiative: does this raise the shared floor, does it build a local (and short-lived) peak, or does it change the map of the business entirely? The fourth question is the hardest: do we dare to invest enough in the third one?
Looking for expert solutions? Discover Crestia's professional services today.