An efficiency drive isn't a strategy, instead look for your real differentiator
Home • About • Projects • Blog • Resources • Contact
I keep noticing the same pattern at companies: when growth slows, or margin comes under pressure, strategy quietly narrows to efficiency. This translates as cutting cost, protecting margin, then finding another percentage point of margin next quarter. Nobody sits down and decides this is now the whole strategy, but it becomes that; because efficiency is available and measurable, right up until there's nothing left to cut and the company is exactly as replaceable as everyone else who's been chasing the same efficiency gains.
I've also noticed something closely related.
When I ask what actually makes the business different, I get one of three answers: either silence, because nobody's actually asked the question in years; or an answer that sounds confident but doesn't survive being poked at, and doesn’t resemble the answer given by the next person asked; or finally, one that’s probably correct but that isn't reflected anywhere in the strategy or in what is actually getting prioritised.
In one previous role, I'd noticed that leadership had no real answer to a question that should have been foundational: ‘who is our customer?’ Not in the marketing sense, in the sense of who the business actually exists to serve when two priorities conflict and someone has to pick. Nobody had asked, so nobody had to defend an answer, so nothing forced the org to be coherent about what it was actually optimising for. Working in product, this made it really difficult to strategise and prioritise. So I decided to take it upon myself to work this out.
I carried out a value-mapping exercise using Itamar Gilad's Business Value Loop Model, deriving a North Star metric from what the business gave customers versus what it captured back, and sketching an opportunity solution tree that would let the organisation be restructured around real problems instead of existing functional lines. I presented it but it didn't land. Leadership's attention stayed on efficiency and tech leverage, which is a comfortable place for attention to stay, because it's legible and it's what everyone else's dashboard already measures.
I could have left it there, but it niggled at me. Instead I decided to treat the rejection as information rather than a verdict, and went back to first principles: I read ‘Understanding Michael Porter’* in order to understand business strategy properly, and used a conversation with a departing executive (someone who had nothing left to protect and no reason to be diplomatic) to pressure-test what I thought I knew about the business's economics. The company's claimed differentiator was that it was the only business in its category doing two things itself, service and technology, rather than buying one and outsourcing the other the way the rest of the market did. Whilst this was true, I felt that nothing was stopping a technology company from acquiring or building its way into the service side. Equally, nothing was stopping another service provider from doing the same in reverse, buying or building the technology it had been doing without.
This differentiator was not in any way an entry barrier, as described by Porter, but instead a capability gap, and capability gaps are exactly what a large enough acquisition budget closes overnight. A genuine barrier to entry has to survive someone writing a cheque for it and nothing the company was currently focussing on could do this.
What nobody in the room was talking about was what was actually sitting underneath all of it: having been the only company combining both sides for long enough, the business had built up a depth of structured data on a specific, underserved customer base that neither kind of competitor could simply buy or replicate quickly, however much capital they had behind them.
A year later I built a sharper version of the same argument and took it to the CEO instead and this time it landed. The business has since changed direction because of it, shifting real investment away from the vertical-integration story that had never actually been defensible and toward the data asset that was. This has since shaped how the company thinks about building AI products at all, because you cannot build a defensible AI product on top of a dataset anyone could assemble in six months, you can only build one on top of something that took years, and would still take years even now, to accumulate properly.
What changed between the first version and the second wasn't volume or seniority. What changed was that I'd spent a year letting the argument get shot at by someone with no reason to be kind about it, instead of just waiting to be vindicated. I feel there's a useful discipline buried in that, one philosophy, maths and science got right long before anyone was writing AI strategy decks: a claim that can't survive a serious attempt to break it isn't a strong claim, it's an untested one wearing the costume of conviction.
These days, businesses need to ask the question: ‘Does this get more valuable as AI capability increases, or does AI make it easier for someone else to close the gap?’ As a departure from original list of ‘7 Powers’**, I believe there are really only five candidates worth taking seriously as an answer:
- Proprietary data that can't be replicated, with the operative word being proprietary (data you can buy access to, or that a competitor could assemble with enough budget, isn't this);
- Compliance that functions as a genuine structural barrier, not paperwork, but a regulatory or safety requirement dense enough that a competitor can't simply move fast and build around it;
- Network effects, where the product gets better for existing users specifically because new users join, not just bigger;
- Accumulated judgement and relationships: the kind of hard-won, embedded understanding of a problem that no model has been trained on, because it never got written down anywhere a model could read it; and
- Brand and trust: as AI makes content, reviews, credentials, and even conversations cheap to fake, the cost of establishing genuine, verifiable trust goes up, not down.
Vertical integration isn't on that list, and neither is efficiency, however hard either one is currently being sold as a strategy; both are features. Features are good but they're not moats, and confusing the two is how a business ends up racing to defend something AI is actively designed to make cheap.
One thing worth being explicit about: this list answers the question for a business built on top of AI, not for the businesses building it. Hamilton Helmer, whose 7 Powers this departs from, has argued the opposite case applies one layer down, that the sheer capital cost of training a frontier model is itself becoming a scale-economy moat, one that gets stronger rather than weaker as AI capability increases. I think he's right, and it doesn't actually contradict the argument above so much as sit beneath it. Raw capital intensity is a real answer to the test if you're a frontier lab or a chip manufacturer. It stops being one the moment you're an ordinary business simply using what they've built, which is where almost everyone reading this actually sits.
So for these companies racing to make AI part of their strategy, particularly in the name of efficiency, I’d argue the bigger question is rather: does your business get better because AI exists, or does AI just make it easier for somebody else to do what you do, faster, cheaper, and without your permission? If you don't know, that's not a failure but it means you need to start thinking hard about this, because the honest answer tends to be more useful than the story you've been telling yourself.
*’Understanding Michael Porter: The Essential Guide to Competition and Strategy’, Joan Magretta (2011)
**’7 Powers: The Foundations of Business Strategy’, Hamilton Helmer (2016)