Skip to content
DT Nova

AI strategy for business: six steps from goal to working systems

An AI strategy is a short plan that says where artificial intelligence will create value in your company, which use cases come first, what they need in data, people and tools, which rules apply, and how you will know it works. For a small or midsize business, a good AI strategy fits on a few pages and leads to a first working system within months, not a long report. This guide sets out six steps, with public data on how European companies use AI today and the EU AI Act dates that shape the plan.

Figures on AI use come from Eurostat's article "Use of artificial intelligence in enterprises" (2025 survey of enterprises with 10 or more employees). AI Act dates come from the European Commission's AI Act page and Regulation (EU) 2026/1744, which amended the AI Act. All were read on 12 November 2026.

Where European companies stand

Eurostat reports that in 2025, 19.95 % of EU enterprises with 10 or more employees used at least one AI technology, up from 13.48 % in 2024. Use rises steeply with size: 17 % of small enterprises, 30.36 % of medium-sized and 55.03 % of large enterprises. Denmark (42.03 %), Finland (37.82 %) and Sweden (35.04 %) had the highest shares. Among companies using AI, the most common purposes were marketing or sales (34.70 %) and organisation of business administration processes or management (31.05 %).

The same survey asked companies that had considered AI but not used it why. The most common reason was lack of relevant expertise (70.89 %), followed by lack of clarity about the legal consequences (52.52 %) and concerns about data protection and privacy (48.83 %). Those three answers are, in effect, the three things an AI strategy for a smaller company must solve: who builds and runs it, which rules apply, and how the data is protected.

Step 1: start from business goals, not from AI

Write down the three to five business goals for the next two years: grow sales in a segment, cut lead times, serve more customers with the same team, reduce errors in a process. Then ask, for each goal, where work is slow, repetitive, error-prone or dependent on finding information. Those are the places AI can help. A strategy that starts from "we need to use AI" ends with demos; one that starts from goals ends with results you can measure.

Step 2: collect and rank use cases

Interview each team for an hour. Ask what they spend time on, what they search for, what they copy between systems and what they would hand to a capable assistant. Write each idea as a use case: who uses it, what goes in, what comes out, and what changes for the business. Then rank them:

CriterionQuestion
ValueHow many hours, errors or lost sales does it address, measured today?
FeasibilityIs the data available and good enough? Does an existing model do the task well?
RiskWhich AI Act class is it? Does it process personal data? What if it is wrong?
AdoptionDoes it fit into the tools people already use? Is there a team that wants it?
Time to resultCan a first version be in daily use within three months?

Choose two or three pilots with high value, good feasibility, low risk and a team that wants them. Typical first picks are a search assistant over internal documents, drafting of quotes or replies, classification of incoming email or documents, and summaries of cases or meetings. Our RPA use cases and NLP applications guides list more.

Step 3: check the data

For each pilot, find where the data lives, who owns it, how good it is and whether it may be used for this purpose. Personal data needs a legal basis, and high-risk processing needs a data protection impact assessment. Decide one rule for company data at the start: which AI services may receive which data, under which contract, in which region. Many problems of AI projects are data problems found too late.

Step 4: decide build, buy or combine

  • Buy when a standard tool does the job well: AI features in your office suite, CRM or helpdesk. Fast, but limited to what the vendor offers, and your data goes where the vendor runs.
  • Build on existing models when the value lies in your own data and workflow: a model connected to your documents, systems and rules, inside your tools. This is where most company-specific value is.
  • Train your own only with a specific task, enough labelled data and a clear advantage.

Whatever you choose, keep ownership: the prompts, integrations, evaluation sets and code should be yours, so you can change model or vendor later. Our guide to what an AI project costs explains the cost drivers.

Step 5: set the rules and the people

The EU AI Act sorts AI uses by risk. According to the Commission, prohibited practices and the AI literacy provisions have applied since 2 February 2025, the Act became generally applicable on 2 August 2026, the rules for high-risk uses in sensitive areas such as employment, education and critical infrastructure apply from 2 December 2027, and those for AI built into regulated products such as lifts or toys from 2 August 2028. Article 4, as replaced by Regulation (EU) 2026/1744, requires providers and deployers to take measures to support the development of AI literacy of their staff and others using AI systems on their behalf.

For the strategy, that means three things: sort each use case by risk class before you choose it, train the people who use AI, and set up light governance that scales with the risk. Our AI governance guide describes the roles, policies and controls in detail. Name one owner in management for the AI strategy and give the teams a clear route to propose new use cases.

Step 6: pilot, measure, scale

  1. Measure the baseline before the pilot: time per task, error rate, response time, sales.
  2. Build the smallest version that real users can use daily, inside their existing tools.
  3. Test quality with a fixed set of real examples and a person who checks the results.
  4. Measure for four to eight weeks against the baseline.
  5. Decide: scale, change or stop. Stopping a pilot that does not work is a good result.

Review the AI strategy every six to twelve months. Models, prices and rules change quickly; the goals and the measurement should not.

An AI strategy on one page

  • Business goals the AI strategy serves.
  • Ranked use cases, with the first two or three pilots and their owners.
  • Data rules: what may go where, under which contract and region.
  • Build, buy or combine per use case; ownership of code and data.
  • Risk classes under the AI Act, governance and training.
  • Metrics, baseline and review dates.

How we help

Our AI consulting work starts with the interviews and the ranked use case list, and ends with a first system in daily use, measured against the baseline. We build on existing models connected to your data, host in EU regions by default, and you own the code, the prompts and the evaluation sets. For automation beyond AI, see business process automation.

Questions people ask

What is an AI strategy?

An AI strategy is a company's plan for where artificial intelligence will create value, which use cases come first, what data, people, tools and rules they need, and how results are measured. It is a business plan with AI as the means, not a technology plan. For most small and midsize companies it fits on a few pages and is reviewed every six to twelve months.

How long does it take to create an AI strategy?

For a small or midsize company, a first version usually takes a few weeks of part-time work: interviews with the teams, a list of candidate use cases, a data check and a decision on the first two or three pilots. The strategy becomes useful once the first pilot is live and measured, so keep the planning phase short.

Do we need a chief AI officer?

Most small and midsize companies do not need a new executive role. They need a named owner in management who is accountable for the AI strategy, a small group that reviews use cases and risks, and access to people who can build and run AI systems, in-house or from a partner.

Should a small company build its own AI models?

Rarely. Most business value comes from using existing models and services well: connecting them to your own data, building them into your workflows and measuring the result. Training a model of your own makes sense only when you have a specific task, enough good data and a clear advantage over what is available.

How does the EU AI Act affect our AI strategy?

It sets the rules for each use case. Some practices are banned outright, high-risk uses such as recruitment or credit scoring carry heavy obligations from 2 December 2027, and many common uses like drafting, search and summarising carry light or no specific obligations beyond transparency in some cases. A strategy should sort use cases by risk class before choosing pilots.

Sources