AI Strategy for SMEs: The Under-€100K Playbook

Why SMEs Face Different AI Strategy Challenges

In short

SMEs face three constraints enterprises don't: a budget typically under €100K, a thin or absent internal AI talent bench, and infrastructure built for operations, not data science. The right SME strategy is shaped by these constraints — not by copying the Fortune 500 stack.

Most AI strategy advice on the open web is written for Fortune 500. That advice assumes a CoE, a data platform, and a multi-million euro budget — none of which apply to a 50-person SME.

Three constraints define the SME context. First, budget: typical SME AI spend runs under €100K per year. Enterprise programmes routinely spend €500K+ on governance alone.

Second, talent: most SMEs have zero in-house data scientists. AI gets delivered by an owner, a tech-savvy operator, or an external partner — not by an internal squad.

Third, infrastructure: SME tech stacks are built for running the business (ERP, CRM, accounting), not for analytics. A "data lake" is not on the roadmap, and shouldn't need to be for the first AI use case.

Eurostat 2025 confirms the resulting gap. Enterprises with 250+ employees adopt AI at roughly 3–4× the rate of small firms (10–49). McKinsey 2025 puts overall adoption at 72%, but the bulk of that is large enterprise — the SMB adoption gap is the biggest structural divide in EU AI use.

The SME Advantage: Speed and Focus

In short

SMEs beat enterprises on three execution dimensions: no committee paralysis (owner decides), no portfolio fragmentation (one or two use cases get full attention), and shorter time-to-value (no platform overhead). Used well, these advantages outweigh budget gaps.

The same constraints that limit SME AI also create real advantages. The owner-operator pattern is the most underrated of them.

At Fortune 500 scale, an AI use case typically requires sign-off from IT, security, legal, procurement, and a business sponsor. Calendars alone add 6–12 weeks. RAND RR-A2680-1 (Aug 2024) names the missing business owner as the #1 cause of AI project failure — magnified inside matrix organisations.

An SME owner can make the same decision over coffee. Budget, risk appetite, and outcome accountability sit with one person. When that person is in the room, decisions compress from quarters to days.

Focus is the second advantage. Enterprise portfolios spread budget across 15–50 use cases. BCG/MIT 2024 link the ~26% GenAI value-capture rate partly to over-fragmentation. SMEs running one or two use cases get the full attention of the operator — and full attention is what turns a pilot into production.

Speed is the third. Across 100+ Alice Labs engagements, the Implementation Index 2026 records a 14-week median from kick-off to production. SME engagements typically land at or below that median — less bureaucracy, faster decisions.

The 3-Use-Case SME Approach

In short

Pick three use cases tied to the operator's biggest weekly headaches: one quick-win automation, one revenue lever, one cost lever. Sequence them — do not run them in parallel. Each use case gets a named owner (usually the operator), a baseline metric, and a 90-day production target.

Enterprise portfolios run dozens of pilots simultaneously. SMEs should not. Three is the right number — and even three should be sequenced, not parallel.

The pattern that works in 100+ Alice Labs engagements:

• Use case 1 — quick-win automation. A repetitive workflow the operator personally dislikes. Document drafting, inbox triage, internal Q&A. Goal: ship in 4–6 weeks, prove the model and the partner.
• Use case 2 — revenue lever. Lead qualification, content for top-funnel marketing, sales-rep enablement. Goal: measurable lift on pipeline or conversion within one quarter.
• Use case 3 — cost lever. Production planning, demand forecasting, ops scheduling. Larger payoff, longer build, more integration. Goal: documented annual saving in euros or SEK.

Each use case gets a named owner (usually the operator), a baseline metric agreed before kick-off, and a 90-day production target. If the baseline does not exist, the use case is not ready to fund.

The trap to avoid: starting use case 2 before use case 1 is in production. SMEs lack the bench to run things in parallel. Sequencing is the discipline that makes the budget feel larger than it is.

Buy vs Build for SMEs (Almost Always: Buy)

In short

For SMEs the default answer is buy. Off-the-shelf SaaS AI delivers faster time-to-value, removes platform overhead, and lets the team focus on adoption rather than infrastructure. Build only when the use case is core to the business model and no SaaS option exists — typically <10% of SME use cases.

Enterprises debate build-vs-buy because they have the engineering capacity to build. Most SMEs do not. The decision is simpler — and the default should be buy.

Buy reasons that dominate for SMEs:

• Time-to-value. Most SaaS AI tools install in days, not months. For a 14-week production target, that matters.
• No platform overhead. A SaaS vendor runs the model gateway, evaluation, and updates. The SME team focuses on adoption and process.
• Predictable cost. Per-seat or per-usage pricing is forecastable. A custom build has unknown maintenance costs after launch.
• Vendor accountability. When the model drifts or the API breaks, the vendor fixes it under SLA — not your overworked head of ops.

Build reasons that occasionally apply:

• The use case is core to the business model — i.e. it is the product, not a support function.
• No SaaS option exists at acceptable quality, after a real market scan (not a quick Google).
• Data residency, IP, or competitive sensitivity makes SaaS unworkable — rare outside regulated sectors.

In practice, fewer than 10% of SME use cases satisfy a real build test. The other 90% should buy, integrate, and ship.

EU AI Act for SMEs (Risk-Based, Not Size-Based)

In short

EU AI Act (Regulation 2024/1689) applies regardless of company size — but obligations scale with system risk classification, not headcount. Most SME use cases (chatbots, drafting, automation) fall outside Annex III high-risk categories. The compliance burden is real but manageable on SME budgets.

A common SME misconception: "the AI Act is for big companies." It is not. The EU AI Act applies regardless of company size. The right question is which risk class the use case falls into.

The Act sets four risk tiers: unacceptable (prohibited), high-risk (Annex III), limited-risk (transparency), and minimal-risk (no specific obligations). Most SME use cases sit in the bottom two tiers.

Typical SME use cases and their likely tier:

• Customer chatbots. Limited-risk: disclose AI to the user.
• Document drafting and summarisation. Minimal-risk in most cases.
• Production planning and demand forecasting. Usually minimal-risk unless tied to critical infrastructure.
• Recruitment screening, employee evaluation, credit scoring. High-risk (Annex III). Avoid unless you can carry the FRIA and documentation load.

The practical SME approach: classify each use case in week 1, prefer limited-risk and minimal-risk applications, and treat any high-risk system as a project in its own right with budget for documentation and oversight.

The 2 August 2026 deadline for high-risk obligations is real. An SME running recruitment screening AI must be ready — but most SMEs simply choose not to run such systems.

Real Alice Labs SME Case: Ljusgårda

In short

Ljusgårda — a Swedish food-and-cannabis production company — deployed AI in production planning with Alice Labs. Documented annual saving: 2.5M SEK. The pattern: one focused use case, owner-led decisions, off-the-shelf model where possible, and a tight weekly review cadence. Exactly the SME playbook described in this article.

Ljusgårda is a Swedish SME production company. Alice Labs delivered an AI engagement focused on production planning — a classic SME cost-lever use case (use case type 3 from the section above).

Documented outcome: 2.5M SEK in annual savings, traced to better planning decisions and reduced waste in the production cycle.

What made it work — and why the same pattern travels to other SMEs:

• One use case, not five. Production planning got full attention. Other ideas were parked until this one was in production.
• Named operator owner. The production manager owned the outcome and was in the room every week — exactly what RAND RR-A2680-1 (Aug 2024) identifies as the #1 success factor.
• Tight metric baseline. Saved euros (and SEK) were measured against a documented pre-deployment baseline, not vibes.
• Buy where possible, build where it mattered. Off-the-shelf components handled what they could; the planning logic specific to Ljusgårda was the only piece that warranted custom work.
• Weekly review cadence. Owner, operator, and Alice Labs delivery met every week. Decisions that would have taken an enterprise a quarter were closed inside a week.

Across the broader Alice Labs Implementation Index 2026 portfolio of 50+ engagements, this pattern recurs — 96% production rate, 14-week median to production. Against an industry baseline of roughly 26% per BCG/MIT 2024, the differentiator is operating discipline, not budget size.

Source: Alice Labs analysis of 100+ Nordic engagements, BCG/MIT 2024, Eurostat 2025