AI strategy and AI programme: 5 steps to a plan

An AI strategy answers three questions: where does AI create measurable value for your business, which rules apply to data and employees, and in what order do you implement? It becomes an AI programme once phases, owners and a timeline are added. Five steps lead from the status quo there.

Strategy or programme: what it is about

Many people search for an AI programme and mean two things. One is the strategy: a short answer to what your company uses AI for and what it does not. The other is the programme: the implementation in several initiatives that build on each other. This article covers both for an SME, first the five steps of the strategy, then the framework for the programme.

The two belong together. A strategy without a programme stays a document, and a programme without a strategy is a collection of tools. The strategy supplies the priorities, the programme the implementation.

Step 1: Assessment

Capture where time is lost today: recurring tasks, media breaks, bottlenecks. This includes a look at your data (what sits where, in what quality) and at the skills in the team. This foundation prevents buying tools nobody needs.

Step 2: Prioritise use cases

Collect possible fields of application and rate each against two criteria: expected value and feasibility. Start with cases promising high value at low risk, typically routine communication, document processing or reporting. Concrete examples are in our article AI in SMEs.

Step 3: Clarify rules and data protection

Before the first tool goes live, you need simple guardrails: which data may go into which services? Who reviews AI output before it is sent? How do you ensure revDSG and GDPR compliance? A one-page AI policy is enough to start and creates confidence in the team. More in the article AI and data protection under the revDSG.

Step 4: Run a pilot and measure

Implement the prioritised use case as a pilot with a clear timeframe, and define beforehand how you measure success: hours saved, faster response times, fewer errors. A pilot that improves nothing measurable gets stopped, that is part of the strategy. Record the result in writing so the decision stays traceable later.

Step 5: Scale and enable the team

What works in the pilot is rolled out to further areas. Invest in enabling employees in parallel: short trainings, shared prompts, clear responsibilities. AI value is not created by tools but by people using them well.

From plan to programme: phases, roles, timeline

The following is a general framework, not a rule. How long each phase takes depends on the size, data situation and capacity of your company.

Four phases

• Orientation: assessment, prioritisation and rules (steps 1 to 3). Result: a short strategy and a list of prioritised initiatives.
• Pilot: one initiative with a clear timeframe and a success measure set in advance (step 4). Result: a decision on whether to continue.
• Expansion: proven pilots are transferred to further areas while the team is enabled in parallel (step 5).
• Operation: ongoing care, regular review of rules and tools, new initiatives from the list.

When a pilot continues

Set the criteria before the pilot starts. At the end you ask four questions:

• Was the success measure set in advance reached?
• Does the team like working with the solution?
• Are the data protection rules being followed?
• Is the effort for operation and maintenance sustainable?

Roles

A programme needs owners, even in a small company. Often several roles sit with a few people:

• One person from management who prioritises and decides.
• A programme owner who keeps the timeline.
• Subject owners from the areas where an initiative runs.
• A contact for data protection and IT who checks rules and data flows.

In a very small company, one person can take on several roles. What matters is that the roles are named.

Timeline

Plan in stages instead of towards a distant end date. Set a decision date for each phase on whether to continue, and review the programme at fixed intervals, for example once a quarter. That way you see early which initiatives pay off and which do not.

As an example of a rhythm, not a standard: the first weeks for orientation, then a pilot with a fixed end date, then expansion in stages with one review per quarter.

How a programme can fail

Possible stumbling blocks can be named:

• There are tools, but no rules on which data may go into them.
• A pilot starts without a success measure and cannot be judged afterwards.
• Nobody is responsible, so in the end nobody decides.
• The programme has no pause for review and keeps running even when the value is missing.

What you hold in your hands at the end

A programme in the sense of this article consists of a few documents: a one-page strategy, a prioritised list of initiatives, a short AI policy and a timeline with fixed decision dates. For a start, that can be enough. The paper is not the goal; it records who decides what by when.

If you would rather not walk this path alone: our AI consulting accompanies you from assessment to implementation. What such support costs is described in the article AI consulting: costs.