You were sold a ladder: spreadsheet, connected system, dashboard, generative AI, agents that act on their own. Each rung raises maturity; you do not climb down. In consulting decks and vendor pitches it sounds convincing. In the field, in real small businesses, it almost never works that way —and in emerging economies the ceiling is lower: limited IT infrastructure, scarce talent and a culture that sometimes contradicts what strategy declares (Shokrollahi Yancheshmeh, 2026; Gupta et al., 2020).
A recent qualitative study of 27 UK SMEs (Amanollahnejad et al., 2026), grounded in the sociotechnical tradition of Trist and Mumford, describes AI adoption as recursive, episodic and reversible: progress in bursts when there is sponsorship, reliable data and someone pushing; rollback when the key person leaves, backup fails or the patch stops holding. That is not method failure: it is how resource-constrained organizations behave. Geographic caveat: the study is UK 2023–2024; the mechanisms help you read an SME, not predict it in detail —regional surveys such as nadIA and CEPAL still anchor Latin America.
Three signals the ladder does not explain
1. When the person who pushed leaves, the project dies. It is not lack of team will: it is champion dependency. Digital momentum was tied to one MD or IT lead with no second operational champion and no handover ritual. Measuring maturity without an actor map is photographing a house of cards.
2. Part of the work lives in exports and re-keying. Export to spreadsheet, copy between systems, artisan integrations because the vendor does not scale to SME size: that is not technical shame, it is sociotechnical bricolage —the real architecture. Promising agentic AI on top of that scaffolding without connecting first is automating the error faster.
3. Important decisions happen at the table, not on the dashboard. In many SMEs the coffee economy —conversation, relationship, context— is legitimate. If AI is perceived as a standardizer that erases the business differentiator, resistance is not “anti-technology”: it is defense of value.
There is more —uncertainty about which number to trust, turnover that destroys training, copying a large-firm playbook and making things worse— but the pattern is the same: maturity is not a destination, it is episodic alignment that must be sustained.
The snapshot matters, but you must repeat it
That is why I measure maturity with IMIA (seven dimensions, six levels) and repeat it: it gives direction and a snapshot in time, not a certificate of arrival. Without sociotechnical hygiene — periodic governance, living documentation, transfer of judgment— today’s score unravels tomorrow. A score that drops is not always disaster: sometimes it signals missing backup, reliable data or clear roles.
Sometimes post-implementation friction should not be hammered away either. When automation duplicates effort —part in the new tool, part in the same spreadsheet as always— the gap is usually missing sociotechnical RACI, not model iteration. That is productive misalignment: friction as a design signal, not moral failure of the team. Sometimes the right move is to stop and return to diagnosis, not accelerate the chatbot.
What to do differently
- Measure before installing. A sociotechnical diagnosis separates where AI pays off, where it does not yet and where it destroys value.
- Name backup before scaling. If digital momentum dies with one person, there is no adoption: there is champion dependency.
- Inventory bricolage. If the critical flow depends on manual export, connect before automating.
- Ask which number governs. Which report would you use for an irreversible decision? If no one agrees, the blocker is not the model: it is trust in evidence.
- Pilot one own workflow. Reject enterprise templates and toxic benchmarks; first bounded, verifiable result.
The regional gap remains huge: in Latin America AI penetration does not reach 4% against more than 20% in Europe (CEPAL, 2024). The updated regional picture is no longer a loose figure: the ILIA 2025 (CEPAL/CENIA) measures 19 countries in the region and classifies them as pioneers, adopters or explorers by ecosystem maturity. Closing it is not climbing rungs blindly. It is building capacity that survives people, turnover and hype —and using AI only where human judgment can filter it. In Kenya, an experiment with 640 entrepreneurs showed generative AI improved high performers (around +15%) but worsened low performers (around −8%): the multiplier has a sign, and for the SME that needs help most the sign is not left to chance.
Full framework in The Sociotechnical Bridge and My approach. Related essay: Digital transformation is not a technical problem · In the public sector, what cannot be audited cannot be used.