How long should an AI strategy document be?

Ten pages is generous. The portfolio appendix can be longer; the strategy itself should be readable in one sitting.

Do mid-market companies need an AI strategy at all?

They need the five decisions above, written down. What they usually do not need is a multi-year transformation programme structure borrowed from a much larger organisation.

AI Strategy

An AI strategy framework that maps to delivery

A five-part AI strategy framework — economic thesis, portfolio, operating model, governance and measurement — designed for mid-market companies rather than global enterprises.

Miguel Torres, Founder, Merjora · Published 10 August 2026 · Updated 3 September 2026

In short

A workable AI strategy has five parts: an economic thesis stating where AI is expected to change unit costs or capacity; a prioritized portfolio of specific workflow opportunities with value ranges; an operating model defining who builds, who reviews and who owns outcomes; a governance position covering EU AI Act classification, data protection and human oversight; and a measurement system with pre-agreed baselines. Everything else — tooling, platform selection, model choice — is downstream implementation detail.

Key takeaways

Strategy is a set of decisions, not a set of ambitions.
Name what you will not do this year. That is the part that creates focus.
Choose the operating model early; it determines delivery speed more than tooling does.
Governance belongs in the strategy, not in a later compliance review.

1. Economic thesis

One paragraph: where does the company expect AI to change economics, and through which mechanism — reduced handling time, higher throughput per person, faster cycle time, or fewer errors? If the thesis cannot be stated without the word 'transformation', it is not yet a thesis.

2. Opportunity portfolio

A ranked list of specific workflow steps with value ranges, confidence levels and readiness constraints, plus a parked list with reasons. This is the output of an opportunity assessment, not a brainstorm.

3. Operating model

Model	Works when	Fails when
Central team builds	Few workflows, deep integration needs	Demand outgrows the team and a queue forms
Federated with central platform	Multiple functions, shared plumbing	Standards are advisory and drift
Vendor-delivered	Commodity workflows, limited internal capacity	Nobody internally owns the outcome after handover

4. Governance

Classify each use case under the EU AI Act and record the reasoning
Define the human oversight point for every automated output
State the data protection basis and retention position per workflow
Set a review cadence and name the accountable owner

5. Measurement

Define, per opportunity, the baseline metric, the target range, the measurement owner and the reporting cadence. Strategies without this section cannot be evaluated, so they get replaced rather than corrected.

Common strategy failures

Tool selection before opportunity definition
A pilot portfolio with no promotion criteria, so pilots never end
Value claimed in hours saved with no conversion to cash or capacity
Governance treated as a gate at the end rather than a design input

Frequently asked questions

How long should an AI strategy document be?: Ten pages is generous. The portfolio appendix can be longer; the strategy itself should be readable in one sitting.
Do mid-market companies need an AI strategy at all?: They need the five decisions above, written down. What they usually do not need is a multi-year transformation programme structure borrowed from a much larger organisation.

Where would this apply in your business?

Merjora maps your workflows, scores each opportunity and quantifies the likely value range before you spend budget.

Discover your AI opportunities