# The SME lag on AI is a window > With no heavy legacy to dismantle, a smaller company can build around AI what large groups struggle to rebuild. But the window has a date, and it only pays off with a new frame of mind and an architecture decided before the tools pile up. Article · By Élie Sader, Network Effect · Published on November 4, 2026 “Even the big groups can't get it to work, so how could we…” Owners of small and mid-sized companies say it often, to justify caution about artificial intelligence (AI). It points the wrong way. What large groups do badly, injecting AI into what they already have without touching their organisation, is exactly what that legacy forces on them. A smaller company that has not even finished digitising has no such legacy to take apart, and for a few years its lag is a better starting position than theirs. Here is why large groups are stuck, why having no legacy is a freedom, why that freedom has an expiry date, and under what conditions it becomes an advantage rather than the purchase of an off-the-shelf package. > Being behind on digital is not a fate. It is a starting position. ## Why can't large groups, with all their resources, make it work? The pattern repeats. A pilot, then another, then a committee, then a study; months, sometimes years; and a verdict that comes back again and again: “promising, but not mature yet”. The existing circuits, the silos, the multi-level sign-offs stay as they are, and AI goes on top, like a turbocharger bolted onto an engine nobody wants to open. There are two reasons. The first is technical. A large group rests on twenty years of layers: enterprise software that has turned into monoliths, systems that do not open up, tangled integration chains. Before building around AI, you have to dismantle, and dismantle without stopping what runs. The second reason is political, and it weighs more. Building around AI means touching the organisation: if AI moves information and orchestrates the work, the roles that used to filter and coordinate lose their purpose, and process experts lose their role as gatekeepers. Nobody has the mandate to dismantle those layers, and the people who occupy them have every interest in the project going nowhere. This is not incompetence; it is the rational defence of a position. A failed pilot can be a success for someone. The objection deserves to be heard. Part of that slowness is risk management, and it is legitimate: a bank, an insurer, a manufacturer running plants does not rebuild its systems lightly. Caution is not always bureaucracy. But on both counts, technical layers and the keepers of those layers, the smaller company is somewhere else. ## Why is having no legacy an advantage? There is a name for this: leapfrogging. In Kenya, a large share of the population had no access to a bank; when mobile money arrived in 2007, it was adopted on a massive scale, without going through a branch or a card. Those who did not have the old system had nothing to undo. Landlines, retail banking: stages that could be skipped because they had never been reached. A company that has not invested much in IT is in the same position. It has no assets to write off, no sunk costs weighing on every decision, no tangled layers to unpick, no roles built around the old circuit. What looked like an absence becomes design freedom: it can think of AI as a way of organising the company, not as one more tool to plug into what exists. Bitcoin has shown the mechanism. Its adoption was not, at first, a matter of conviction but of exit cost: you leave a system when leaving costs less than staying. The first to switch were those for whom leaving cost least, because they had nothing to write off and no position to defend. The shift did not come through revolution but through quiet migration. Faced with AI, the company with no legacy is in exactly that situation. It is the junior in the story: not smarter than the senior, simply with less to unlearn. Still, you need to know what “building around AI” means for a smaller company. ## What does building around AI mean for us? Not “making my teams more productive with AI”. That is pouring more volume into the same processes, and adding AI to a circuit nobody has rethought mostly produces friction disguised as speed. The question is rather: which roles exist here because we did not have AI, and which exist to set the intent and check the result? An organisation designed today would keep two families of functions. Those of intent: strategy, what we want, how to specify it. Those of validation: quality, security, final judgement. A growing share of the rest can be handed to AI agents, programs that act on their own: they read, prepare, carry out. One public example shows the scale. In May 2026, an American software company cut 22% of its staff while announcing a “100x” organisation built on AI, with very high salaries for those who stay. It is a signal, not a model. A company whose value lies in its software can restructure that way; a services or manufacturing business has a relationship layer, the field, the customers, the suppliers, that does not shrink like that. And an organisation that is too thin concentrates risk on a few people: when one of them leaves, a lot leaves with them. So the smaller company does not need to imitate. It needs to ask the question before stacking tools, because it still can. As for how to rebuild, and which layer to start with, that is the subject of [our article on layering or rebuilding](https://networkeffect.fr/en/layer-or-rebuild/). “Still” is the important word. ## Why talk about a window? Because AI architectures will set. Enterprise software was, when it arrived, a promise of flexibility; it became the monolith that is so hard to take apart today. The same will happen to stacks of AI tools. Once everyone has installed their layers, agents, subscriptions and integrations, large and small will have the same inertia, and the leap will no longer be possible. Our estimate is that the window will stay open for a few years, three to five perhaps; it is an estimate, not a measurement. Surer than the number is the mechanism. Infrastructure, once in place, persists by default, because taking it down costs more than maintaining it. It is a ratchet effect: once the technical rails are laid, nobody has to choose them again for them to keep running; it is enough that nobody switches them off. Bitcoin's advocates have long described this for infrastructures of control; the lesson applies just as well to a stack of AI tools installed without thought. Whatever a smaller company plugs in today for convenience will still be there in ten years, and it is on top of that that it will have to build. The real competitor of the smaller company is not today's large group. It is the window closing. But having no legacy is not enough. ## Is having no legacy enough? No, and this is the most important part. Having no legacy is not the same as having competence: no past IT investment often also means no technical team, no habit of designing an architecture, no access to the right experts. The leap only works when three things come together. **No debt.** Many smaller companies already have this: few heavy systems, few layers to dismantle. **A new frame of mind at the top.** Seeing AI as a way of organising the company, not as a tactical tool to add to the list. It is a change of outlook, and it cannot be bought. **Advice that backs that architecture.** Not advice that reproduces the usual split, where large groups get expertise and smaller companies get the packaged product. Without the last two, the company does what the companies around it do: it buys off-the-shelf, because everyone buys it. It then depends on the vendor for every change, and reproduces a smaller copy of the large-group model, with less legacy but more dependency. The window closes for it without its noticing. And its most valuable asset, what its people know, leaves for other people's tools before it has even been written down. One question needs to be asked plainly: who pays for building the skills? If the answer is “nobody”, the advantage is cosmetic. The leap will therefore be highly selective: it will not work for every smaller company, and it does not depend on size, but on those three conditions. So where do you start, to seize the window? ## How do we seize the window? Three moves, in this order. **Start from long-term goals, not from the tool.** Where does the company want to be in five years, and what stands in the way? That is where you work out where AI comes in, in what order, and to do what. The tool comes last. **Write down what the company knows.** Its customers, its rules of thumb, its decisions and the reasons behind them, what the field says before the figures do. In a smaller company, that knowledge is still concentrated, in a few heads, close to the ground; in a large group, it is scattered across silos and twenty tools. It is the smaller company's intangible capital, what an AI must read to serve it, and it is the raw material of the leap. The smaller company has no legacy to dismantle; it has knowledge to write down, and that is where the leap begins. **Decide the architecture before stacking tools.** What stays in-house, in files readable by a person and by any model, so you can change model and provider without redoing everything. What can be rented, and on what terms. Deciding this now costs some thought; deciding it in five years will cost a dismantling. The window is seized by an architecture decision, not by a subscription. ## The lag is not a fate, it is a deadline The lag of smaller companies on AI is not a fate: it is a window, and a window has a date. It is seized by deciding the architecture before stacking the tools, and by writing down what the company knows while it still knows it. If you are taking the turn now, [the VIRAGE workshop](https://networkeffect.fr/en/services/strategy/#virage) (French for “the turn”) starts from your long-term goals and works out where AI should come in, in what order, and with what context. For that context, the knowledge to write down, [intangible capital](https://networkeffect.fr/en/intangible-capital/) sets out what it contains, and [CORPUS](https://networkeffect.fr/en/services/corpus/) how to hold it in-house. ## Further reading - [What strategy for AI and Bitcoin?](https://networkeffect.fr/en/services/strategy/) - [Zero times a thousand: what AI amplifies, and what it will never give you](https://networkeffect.fr/en/zero-times-a-thousand/) - [All resources](https://networkeffect.fr/en/resources/)