PERFORMANCE · Business & Performance Coach

AI Adoption Doubled. Only 29% of Using SMEs Put It in the Core Business

AI becomes strategic when it compresses a complete cycle, improves a decision or increases the capacity to sell and deliver.

François Assock· · 10 min read

Direct answer: using generative AI to produce more text is not yet a durable operating advantage. AI becomes strategic when it reduces the cycle time of a core process, removes handoffs, improves a consequential decision or increases the company’s capacity to sell and deliver without proportional cost growth.

20.2%of firms across available OECD countries reported using AI in 2025, up from 8.7% in 2023.
29%of generative-AI-using SMEs reported using it in core company activities.
65%reported improved employee performance.
26%reported increased revenue, far below the share reporting productivity benefits.

Adoption doubled. Advantage did not.

OECD data shows firm-level AI adoption rising from 8.7% in 2023 to 20.2% in 2025 — roughly a 132% relative increase in two years.

But a second number matters more. Among SMEs using generative AI, only 28.7% report using it in core activities: the activities that produce the company’s main goods, services or revenue.

Most firms are still using AI at the edge: drafting, summarizing, brainstorming or polishing communication. Those uses are helpful. They are also easy for competitors to copy.

The four levels of AI leverage

01

Content assistant

AI drafts, summarizes and generates ideas. The gain is individual and widely available.

02

Workflow assistant

AI receives structured inputs, produces a standard output, updates a system and removes recurring manual work.

03

Decision system

AI consolidates sources, detects anomalies, prepares a diagnosis and helps a human make a higher-quality decision faster.

04

Revenue system

AI directly increases the capacity to prospect, personalize, convert, serve or retain customers without proportional cost growth.

The metric that matters: AI leverage ratio

Coach François framework AI leverage = (hours removed + errors avoided + margin or revenue enabled) ÷ (tool cost + human supervision + maintenance)

The formula is not designed for perfect accounting precision. It forces a company to move from “we use AI” to measuring economic and operational impact.

Five metrics for every AI workflow

1. Cycle time

Measure the elapsed time between an input and a usable output. Track hours or days, not a subjective feeling of speed.

2. Number of handoffs

Every transfer between people or tools adds delay, context loss and error risk. Removing two handoffs often creates more value than writing one document 30% faster.

3. Rework rate

FormulaRework rate = outputs requiring material correction ÷ total outputs

An automation that accelerates production while doubling corrections has not improved the system.

4. Founder supervision time

If the founder must verify every output, move information and resolve every exception, AI has not reduced dependency. It has relocated the work.

5. Margin or revenue per cycle

For sales and delivery workflows, measure conversion, average value, retention or cost-to-serve. The OECD survey found that 65% of AI-using SMEs reported employee performance gains, but only 26% reported increased revenue. Productivity does not automatically become customer value.

Hypothetical example.

Before: a lead submits a form, an assistant copies the answers, the founder analyzes them, a copywriter drafts a proposal and a salesperson follows up. Five handoffs, four days of elapsed time and ninety-five minutes of human work.

After: the form automatically builds the account file, AI prepares a diagnosis and structured proposal, and the founder validates the strategic risks. Two handoffs, one day and twenty-eight minutes of human work.

The advantage is not the generated prose. It is the compression of the complete cycle.

How to move AI from the edge to the core

  1. Choose a workflow connected to revenue, delivery or a critical decision. Do not begin by searching for a use case for a tool.
  2. Measure the baseline. Time, people, errors, cost, delay and economic result.
  3. Remove unnecessary steps first. Automating a bad process only makes the bad process faster.
  4. Define the human role. Judgment, approval, relationships, exceptions and accountability must remain explicit.
  5. Test on thirty cases. Compare performance before scaling.

Performance still starts with the human

AI does not turn unclear direction into good direction. It does not automatically know which promise deserves to be sold, which compromise protects the brand or which risk should be refused.

It amplifies the system it enters. A clear founder uses AI to expand reach and decision quality. A scattered founder often uses it to produce more non-priority work.

The durable advantage is not access to the model. Access will spread. The advantage is knowing which workflow deserves to be rebuilt, which data it needs, where human judgment belongs and which metric must move.

Frequently asked questions

What percentage of firms use AI in 2025?

Across OECD countries with available data, 20.2% of firms reported using AI in 2025, up from 14.2% in 2024 and 8.7% in 2023.

Do SMEs use generative AI in their core business?

Not often yet. An OECD survey found that only 28.7% of generative-AI-using SMEs used it in core activities.

How should a company measure AI ROI?

Track cycle time, human hours, rework, tool cost, supervision and the effect on margin or revenue. Compare before and after across a meaningful number of cases.

Can AI replace ten people?

AI can compress skills, remove tasks and eliminate handoffs in a small organization. It does not automatically replace ten complete roles, human accountability, customer relationships or management judgment.

Sources and methodology

  1. OECD — AI use by firms reached 20.2% in 2025
  2. OECD — How SMEs are using generative AI
  3. OECD — AI adoption by small and medium-sized enterprises

The thresholds and examples presented as the “Coach François framework” are practical decision rules, not universal statistical laws. Hypothetical examples are explicitly identified.

About François Assock

François Assock is a business and performance coach. He helps entrepreneurs launch new ventures or restart businesses that have stalled by combining mindset, strategy and execution.