Introduction
Generative AI has moved rapidly from novelty to managerial utility. Small and medium-sized enterprises (SMEs) are beginning to use these systems for drafting, summarising, ideation, customer communication, and operational support. The appeal is obvious: generative AI is relatively inexpensive, widely accessible, and capable of producing useful outputs across a broad range of tasks.
Yet a more important research question sits beneath the adoption story. The issue is not simply whether SMEs are using generative AI, but what happens to managerial decision-making once these systems become part of the decision environment.
Much of the current discourse still assumes a straightforward progression: more capable AI systems improve information processing, and improved information processing leads to better strategic decisions. That assumption is too simple. Generative AI does not enter a neutral space. It enters organisations already shaped by bounded rationality, time pressure, resource constraints, uneven expertise, and competing interpretations of what the underlying problem actually is.
In that context, AI-generated output is not a decision. It is an input into a human interpretive process.
This article argues that the strategic significance of generative AI in SMEs lies less in automation alone and more in how managers interpret, trust, contest, and integrate AI-generated outputs into judgement under uncertainty. The core issue, therefore, is not merely adoption. It is sensemaking and trust calibration in AI-augmented decision-making.
Why the “better AI = better decisions” assumption is weak
A recurring assumption in both business commentary and parts of the AI adoption literature is that improved technical capability naturally produces improved managerial judgement. If a system is faster, more articulate, and able to synthesise large volumes of information, it is tempting to assume that better decisions will follow.
The problem is that decisions are not made by models alone. They are made by actors operating within specific organisational contexts.
Managers do not receive AI outputs as neutral truth. They receive them as summaries, recommendations, rankings, scenario suggestions, or strategic drafts that still need to be interpreted. The same AI-generated output may be treated by one manager as a useful prompt, by another as highly credible advice, and by a third as untrustworthy noise. This means that AI capability and AI effect are not the same thing.
Recent OECD evidence already suggests the need for caution. A 2025 OECD report based on a survey of more than 5,000 SMEs across seven countries found that generative AI was already being used by a substantial share of SMEs and was associated with benefits such as employee support and performance gains. At the same time, the report indicates that generative AI is often being used for relatively peripheral tasks rather than for deep transformation of production systems or strategic decision structures. In other words, adoption should not be conflated with strategic integration.
That distinction matters. Using generative AI to draft an email or summarise a document is not equivalent to using it in pricing, investment, hiring, supplier selection, or market-entry decisions. The managerial stakes, epistemic demands, and consequences of error are entirely different.
SMEs are not simply smaller versions of large firms
A second weakness in current discussion is the tendency to treat SMEs as scaled-down versions of large organisations. They are not. SMEs often operate under tighter resource constraints, flatter managerial structures, lower formalisation, and greater dependence on the judgement of a small number of decision-makers.
These characteristics are central to the study of AI in SMEs because they shape not only whether AI is adopted, but how it is interpreted, challenged, and acted upon.
Generative AI may offer SMEs access to analytical and communicative capabilities that were previously expensive or inaccessible. A founder-manager can now ask an AI system to draft a competitor scan, generate customer segments, compare strategic options, or structure a market-entry outline in minutes. That is significant. But democratisation of access is not the same as democratisation of judgement.
If the decision-maker lacks the time, domain expertise, or organisational routines needed to critically evaluate the output, then generative AI may amplify weak reasoning just as easily as it supports strong reasoning. In large firms, AI use may be buffered by analysts, governance processes, or specialist teams. In SMEs, the distance between AI output and strategic action may be much shorter. That makes managerial interpretation even more consequential.
Generative AI as an interpretive input rather than a strategic oracle
A more useful way to conceptualise generative AI in SME decision-making is to treat it as an interpretive input rather than a strategic oracle.
By interpretive input, I mean a resource that enters the decision process in provisional form. It can frame options, synthesise information, generate scenarios, and produce candidate recommendations, but it does not remove the need for managerial judgement. If anything, it intensifies that need.
Consider a straightforward strategic task: an SME manager is considering whether to expand into a new product category. A generative AI system can quickly produce a market summary, identify competitors, suggest customer segments, and draft a provisional go-to-market plan. At first glance, this appears to represent a major improvement in analytical capacity.
But each output immediately raises further questions:
- What assumptions has the model embedded in its analysis?
- What evidence is missing, outdated, or potentially fabricated?
- Has the output flattened sector-specific nuance?
- Does the recommendation align with the firm’s actual resources and capabilities?
- Is the output persuasive because it is analytically robust, or simply because it is fluently written?
These are not merely technical questions. They are managerial and interpretive ones. This is why the value of generative AI in strategic decision-making cannot be reduced to output quality alone. It depends heavily on the quality of managerial sensemaking around that output.
Why sensemaking is central
Sensemaking provides a stronger lens for understanding AI use in organisations than simple adoption models. In organisational research, sensemaking refers to the process through which actors construct meaning in ambiguous, uncertain, or rapidly changing environments. It is particularly relevant where information is incomplete, cues are equivocal, and action must occur before certainty is available.
That description fits many SME decision environments closely.
From a sensemaking perspective, generative AI does not simply “provide answers.” It produces cues, framings, syntheses, and possible interpretations that managers must work into a coherent understanding of the situation. The task is not to receive truth from the system, but to decide what the output means, whether it is credible, and how it should influence action.
This matters because generative AI often produces outputs that are rhetorically convincing even when the underlying reasoning is shallow, generic, or weakly grounded. A manager who mistakes fluency for validity may over-trust the system. A manager who dismisses it entirely because of occasional hallucinations may underuse a potentially valuable aid. Neither position is analytically satisfying. What matters is calibrated trust.
Trust is not binary. It is a calibration problem.
Much of the conversation around AI in management still treats trust as if it were a simple yes/no variable: either managers trust the system or they do not. In practice, trust is better understood as a calibration problem.
Managers are likely to adjust reliance on AI depending on:
- the type of task involved
- the perceived consequences of error
- time pressure and urgency
- prior experience with AI performance
- their own expertise in the domain
- the presence or absence of internal verification routines
A well-calibrated manager might rely heavily on generative AI for low-stakes drafting and exploratory ideation, use it more cautiously for analytical framing, and subject it to rigorous verification in high-stakes strategic decisions. A poorly calibrated manager may do the opposite: either defer too readily to the system because it appears authoritative, or reject it wholesale because it is imperfect.
This raises one of the most important questions for future research:
Under what conditions do managers calibrate trust in generative AI appropriately, and under what conditions do they over-rely or under-rely on it?
This is not only a psychological question. It is also an organisational one. Trust calibration is shaped by routines, norms, task design, power structures, and the availability of challenge mechanisms inside the firm.
The strategic risk is not only hallucination
Public discussion of generative AI often focuses on hallucinations, fabricated facts, and false confidence. Those risks are real. But in strategic decision-making, the deeper issue may be broader than factual error alone. The more significant risk is that AI subtly changes the shape of managerial reasoning without managers fully recognising how that change is occurring.
There are at least four ways this can happen.
1. Premature closure
A plausible AI-generated synthesis may cause managers to stop searching too early. Instead of broadening strategic analysis, the tool may narrow it by making the first answer feel sufficient.
2. Framing capture
The way a problem is posed to the model, and the way the model responds, can anchor decision-makers to a particular interpretation of the issue. Alternative framings may disappear before they are seriously examined.
3. Confidence inflation
Because generative AI outputs are often coherent, polished, and fast, managers may attribute more authority to them than the underlying reasoning warrants.
4. Outsourced ambiguity management
Managers may begin using AI not only to process information but to relieve the discomfort of uncertainty itself. That is a strategic danger. Uncertainty has to be managed analytically, not cosmetically overwritten.
The concern, then, is not simply whether AI produces factual inaccuracies. It is whether it reorganises the decision process in ways that improve, distort, or defer managerial judgement.
What current evidence tells us — and what it still does not tell us
The evidence base is still developing, but some patterns are already visible. OECD findings indicate that generative AI is spreading meaningfully through SMEs and may support employee performance and certain capability gaps. At the same time, those findings also suggest that use remains concentrated in relatively bounded tasks rather than in deep strategic transformation.
This points to an important distinction between adoption and decision integration. We know increasingly that firms are experimenting with generative AI. We know far less about how managers actually incorporate AI into pricing decisions, market-entry assessments, hiring choices, strategic pivots, or crisis responses. We know even less about how trust in AI is built, challenged, and recalibrated over time inside SMEs.
This is where management research needs to move next. The key empirical question is not simply whether a prompt was used. It is what happens between prompt and decision: the interpretation, the contestation, the verification, the shortcuts, and the moments where managers choose either to rely on or override the system.
A more useful research agenda for SMEs
If generative AI in SMEs is to be studied seriously, three shifts are necessary.
1. Move from adoption to use-in-practice
It is no longer enough to ask whether SMEs are adopting generative AI. Research needs to examine how these systems are used in concrete decision contexts such as hiring, forecasting, supplier selection, pricing, market expansion, and organisational restructuring.
2. Treat managerial cognition as central rather than residual
The real action is not only in the model, but in the manager’s interpretation of the model. Research should focus more explicitly on the cognitive and organisational mechanisms through which AI output becomes actionable.
3. Study trust calibration rather than “trust” in the abstract
The critical issue is not whether managers trust AI in general, but whether they trust it appropriately for specific tasks under specific conditions.
These shifts would move the literature beyond broad claims about disruption and toward a more realistic understanding of AI-mediated decision-making in SMEs.
Conclusion
Generative AI is likely to become a routine feature of SME management. But its significance should not be overstated in simplistic terms. Better tools do not automatically produce better decisions because decisions are not generated by tools alone. They are generated through the interaction of tools, managers, organisational contexts, and uncertain environments.
For that reason, the most important question is not whether generative AI is entering SME decision-making. It already is. The more important question is what kind of decision process it is helping to create.
If generative AI is treated as a strategic oracle, it may encourage overconfidence, premature closure, and uncritical reliance on plausible output. If it is treated as an interpretive input within a disciplined process of managerial judgement, challenge, and verification, it may genuinely expand the cognitive capacity of SMEs operating under constraint.
That distinction is where the next phase of research should begin.
Author Note
Saba Shahzadi is an emerging researcher working at the intersection of generative AI, managerial cognition, and strategic decision-making in SMEs. Her current doctoral research focuses on sensemaking, trust calibration, and human–AI interaction under conditions of uncertainty.
References / Further Reading
Use this at the bottom of the article.
- OECD (2025). Generative AI and the SME Workforce: New Survey Evidence. OECD Publishing. DOI: 10.1787/2d08b99d-en.
- OECD (2025). Organize posts with categories / tags isn’t a research source, so do not cite WordPress support pages in the article itself.
- When you later expand the article, I’d add 4–6 academic sources around sensemaking, trust calibration, and AI decision support.
shahzadisaba4z-biesm 
Leave a comment