The bridge applied to SMEs

Editorial note. This chapter preserves the earlier concise English edition. A full translation of the expanded Spanish edition published on July 21, 2026 is in progress.

Part IV brings the method down to concrete ground. These are not four loose chapters: they are four lenses on the same kind of question —where does AI add value and where does it destroy it?—. This chapter speaks of small scale and cash pressure; the public sector of rights and auditability; IMIA makes the rule “measure before buying” measurable; value metrics ask whether the intervention changed someone’s day. Distribuidora Norte appears in all four with different emphases.

Here the thesis stops being theory. Behind almost every SME there is a person who risks what is theirs —the savings, a life’s project, the work of the people who go along with them— and who hears from every direction that they must “get on board with AI”, without anyone explaining how to do it without stumbling. This chapter is for that case: how the sociotechnical bridge becomes real value for a small or medium business, and why, on that ground, guarding against harm comes before promising gain.

The worst of both worlds

The SME lives a contradiction it did not choose. It has all the pressure to adopt —the AI discourse is at every trade fair, every bank, every vendor— and none of the buffers of the large company: not the data muscle, not the technical team, not the margin for a project to go wrong without it hurting. Where a corporation can afford three failed pilots and learn, the SME has a single shot and a cost of being wrong measured in months of cash.

Field evidence makes that asymmetry even more uncomfortable, because it shows that AI is not neutral with respect to who uses it. In an experiment with hundreds of entrepreneurs, generative AI helped the high performers —by around 15% more— and harmed the low performers —by around 8% less—, because the latter followed the model’s generic advice on hard tasks without the judgment to discard the bad ones.¹ The conclusion is harsh: AI without a human reading to guide it does not level, it widens the gap that already existed. Translated to the ground: the SME that most needs help can be sunk by poorly applied AI. That is why the bridge, in an SME, starts by protecting against harm before promising gain.

The regional context confirms that the problem is structural and not anecdotal. In Latin America, AI penetration does not reach 4% —against more than 20% in Europe—, and within each country the distance between large firms and SMEs is as wide as the one that separates the regions: in Brazil, 41% of large companies use AI against barely 11% of SMEs.² And when one looks closely at those that do adopt, the decisive nuance appears: a survey of Argentine SMEs found that 41.6% already use some AI tool, but concentrated in the most basic —text generation— and with very low governance and internal-capability indicators.³ In other words: surface adoption has already happened; what is missing is everything beneath it. The SME is not “behind out of clumsiness”, it is behind because no one built it a bridge to translate the technology to its scale. The regional anchor to situate that gap, today, is the Latin American Artificial Intelligence Index, which measures nineteen countries in the region and classifies them as pioneers, adopters and explorers by the maturity of their ecosystem.⁴

That configuration has a name in the foundations of the method. What happens to the SME is that it falls into the chasm of the adoption curve: pilots are designed for the enthusiast who tolerates friction, and they die crossing over to the one who only demands that the thing work —and the SME is, almost by definition, that second audience—. And what prevents it from leaping all at once is its absorptive capacity: without a prior history of data and analytics, an organization does not “buy AI” and assimilate it from one day to the next, because it lacks the prior knowledge that would make it usable. → Authors and currents.

Triple constraint in emerging economies

Beyond the large-firm/SME gap, adoption in emerging economies hits a triple constraint (Shokrollahi Yancheshmeh, 2026; Gupta et al., 2020): limited IT infrastructure, scarce specialized talent and culture that contradicts digital strategy. In Latin America, figures remain anchored in CEPAL/nadIA; the frame explains why the bottleneck is triple, not only quantitative. The bridge responds by reading culture and incentives (link 1) before licenses —especially when infrastructure and talent are already stretched.

From scarcity, not from excess

It is worth specifying where the SME starts from, because it changes the whole framing. Most do not arrive at AI from the excess of failed technology, but from scarcity: they manage with spreadsheets, have their data in silos that do not talk to each other, processes are manual and decisions are made without evidence. That, against first impressions, is the cleanest opportunity: there is order to build and little technical legacy to dismantle before starting.

But there is a nuance that inverts the intuition and that the method puts up front: starting from little or no digitalization does not reduce the risk, it concentrates it. Whoever starts from zero does not have less to resolve, they have to build the technical subsystem and the human one at the same time —the data and the trust, the system and the habit of using it— with no prior scaffolding to lean on. That is why the value path is not a leap to AI but a constructive arc: connect the silos, raise an information architecture, govern the data until it is trustworthy and only then add agents and AI where the organization is already ready. It is the concept from silo to architecture, and its rule number one is always the same: measure maturity first. → Concepts of our own.

What the method refuses to offer

A good part of what defines the bridge in an SME is what it refuses to do, because each refusal is the antithesis of a habitual way of failing. It does not offer AI: it offers, first of all, to know whether AI suits this SME, where and when —and sometimes the honest answer is “not here yet”—. It does not offer a model: it offers freed-up hours and better decisions, which is the only thing the SME can take to the bank. It does not offer a large investment: it works with modular tools and with the data the company already generates, not with a platform that has to be justified for three years. And it does not offer a two-year transformation: it offers a first bounded and verifiable result, because an SME needs to see the value before it can believe in it. Each of those refusals protects the person who risks what is theirs from the version of AI that would leave them worse off than before.

The method, in SME terms

The four links of the method do not change in nature when brought down to the SME; they change in size. Landed, they look like this:

What is distinctive is not the four disciplines, which exist loose in the market, but that they are exercised as a single path with no handoffs: the hypothesis about how the team works is not lost between the one who diagnoses and the one who builds, because it is the same practice that does both things. → The four links.

Three principles and a promise

That practice, said in the SME’s voice, is ordered into three principles. First you measure, then you install: the diagnosis exists so as not to spend on what does not pay off. You start by freeing up time, not by impressing: automating the repetitive and low-value gives talent back the hours it was squandering, and that is the first return an SME truly feels. And you work with what is already there: its data, its tools, with no large investments or hidden costs. The promise, then, is not to put AI into the company, but to find out first whether it suits it and build only what gives it back hours or better decisions, without requiring it to understand technology.

It is also worth naming the places where the pain shows up, because they are almost always the same and almost never the ones the owner expected. “I have the data but it is no use for deciding” is the data that lies, and it is solved by giving it context, not by buying a dashboard. “I tried an AI tool and nothing changed, or it was worse” is the trace of a diagnosis that was not done: the error was automated, so now it is committed faster. “We handle everything in spreadsheets and nothing connects” is the first step of the constructive arc —connecting the scattered information— and comes long before thinking about AI. In each case, the stated problem and the real problem do not coincide, and separating them is half the work.

Many SMEs are in digitization without redesign: the PDF replaces paper but handoffs stay the same. Without an end-to-end process owner with a mandate across silos, each area optimizes its piece and the customer pays the friction. Process and data with meaning first; automation and AI after.

The value beyond the SME

There is one last layer that exceeds each client. Every SME that crosses the bridge closes a piece of the gap: the value is not only one company’s gain, it is shrinking the distance between the few who already use AI with judgment and the productive fabric being left behind. In a region where that internal distance is as large as the one that separates it from the developed world, helping an SME cross is not only business —it is development, measured where it matters—. → closing the gap in Concepts of our own.

And it is worth closing the circle with what opened this chapter. The same AI that can close that gap is the one that, poorly applied, widens it: in Kenya it helped those already doing well and sank those doing badly. It is not a paradox, it is the nature of the tool —AI multiplies what it finds, and what decides the sign of that multiplication is not the model’s power but the judgment that guides it—. In the SME that most needs help, letting the tool loose without that judgment is betting that the sign comes out positive. The bridge is, without metaphor, what sets the sign.

Honesty criterion. The value of a project in an SME is not measured in the model left running or the dashboard switched on, but in the person: an hour recovered, a decision that comes out better, a piece of data that can finally be trusted. When the metric shifts from the artifact to the person, it becomes impossible to confuse activity with result —and almost all the projects that do not pay off live in that confusion—.

Mechanisms recent evidence names

Amanollahnejad et al. (2026) interviewed 27 UK SMEs through an STST lens.⁵ They do not contradict the bridge: they name phenomena the diagnosis already finds. Adoption there is recursive and episodic —partial, reversible progress— not an irreversible ladder to maturity.

When automation duplicates effort —some on the tool, some by hand— the gap is usually missing sociotechnical RACI, not model iteration. Sometimes the right move is to stop, not accelerate.

Distribuidora Norte, second pass

Synthetic case. Rebuilt from patterns observed in the field; it does not correspond to an identifiable client. It continues the story begun in The four links.

A year after the at-risk customer alert project, the same Distribuidora Norte returned with another pain: delayed deliveries, complaints that did not match the routing dashboard, driver spreadsheets the official system did not reflect. The owner named an AI opportunity in delivery time windows; diagnosis found first a scar —a routing pilot that failed and no one had named in the meeting— and Excel shortcuts that held up real operations while the dashboard showed something else. The opportunity was genuine; governance and data were not yet ready for a model to decide on its own.

The intervention first unified the data flow between complaints and delivery, co-developed a dashboard with supervisors —not for them, with them— and only then added prediction where the team validated what “on-time delivery” meant in each zone. The result that matters is not a precision percentage but the sequence: measure, connect, govern, only then automate —and whoever read the organization kept building without dropping the hypothesis at the handover. That profile, in numbers, is what IMIA summarizes; the change in the driver’s and supervisor’s day is what value metrics measure.

See also: The bridge applied to the public sector · IMIA, the maturity instrument · The four links · Authors and currents · Bibliography