Guide · October 2026 · 16 min
Data and AI strategy: what it should contain and how to build it
What a data and AI strategy is, what it should contain and how to build it in 12 weeks. With a scoring model, an example and the pitfalls that most often stall the strategy.
Anders Butzbach Christensen
Plan & Proof
What is a data and AI strategy?
A data and AI strategy is leadership's decision on which 3-5 data and AI initiatives the company will fund over the next 24 months, what they are worth in money, and what needs to be in place in data, platform, organisation and governance for them to succeed.
It is one strategy, not a data strategy with an AI chapter at the back. The ambition and the portfolio of initiatives are the strategy itself. Data, platform, organisation and skills are built only to the extent the portfolio requires.
A good data and AI strategy leaves you with three things:
An approved roadmap
24-36 months with investment envelope, funding model and KPIs.
An operating model
Roles, governance and sourcing, so someone owns the value.
Business cases for 3-5 lighthouses
Each with owner, price, value, KPI and start date.
That is the difference between a strategy and a report: the strategy ends in decisions, not in a document that is read and put away.
Why do companies need a data and AI strategy now?
AI is widespread in Danish companies, but the value is not. 59% of companies with at least ten employees use AI in 2026, up from 42% in 2025 and 15% in 2023, according to Statistics Denmark. In the Confederation of Danish Industry's own survey, 42% use AI ad hoc, and only 11% use it systematically in virtually all departments.
The impact is not keeping up. In McKinsey's survey from August 2026, 37% see any EBIT impact from AI, the same as the year before. Only about 6% are high performers who attribute at least 5% of EBIT to AI.
The difference rarely lies in the technology:
Workflows
Nearly three in four high performers have fundamentally redesigned workflows, against one in four of the rest (McKinsey).
Control of the money
AI spending has grown from about 1.7% of revenue in late 2025 to 3.3%, and more than 80% sits outside IT (BCG Applied AI Index 2026).
People before algorithms
About 70% of the value from AI comes from people, processes and change management, 20% from technology and 10% from algorithms (BCG, January 2026).
Data
In 2025, Gartner predicted that through 2026 organisations will abandon 60% of AI projects unsupported by AI-ready data (Gartner, February 2025).
A data and AI strategy pulls the money together, prioritises the few initiatives that show on the bottom line, and makes sure the work actually changes.
What should a data and AI strategy contain?
A data and AI strategy covers eight building blocks. Two of them are the strategy itself, four are capabilities built in step with the portfolio, and two are the foundation everything rests on.
| Building block | The question the strategy must answer | What the strategy contains |
|---|---|---|
| 1. Ambition and direction | What should data and AI do for the business over the next three years: make it more efficient, grow it or transform it? | Ambition, 3-5 goals with KPIs and a level of ambition per business area |
| 2. Value portfolio | Which initiatives create the most value, and what are they worth? | Use case catalogue, prioritisation and business cases for 3-5 lighthouse projects |
| 3. Data foundation | Which data must be AI-ready for which use cases, and who owns them? | Critical data domains, data products, ownership and quality requirements |
| 4. Platform and architecture | What do we build, buy or reuse, and how does it fit together? | Target architecture, build/buy principles and an agent and model platform |
| 5. Operating model | Who does what, and which roles are missing? | Organisation, roles, sourcing and staffing |
| 6. Governance, risk and regulation | How do we decide on, control and document the use of data and AI? | Decision forums, AI policy, AI register, the AI Act and the Data Act |
| 7. People and adoption | How do we get people to work differently, not just get new tools? | AI literacy per role, a change plan and new workflows |
| 8. Roadmap, funding and measurement | In what order, for what money, and how do we know it works? | Roadmap, investment envelope, KPI tree and review rhythm |
Five principles that keep the strategy sharp
Value first
Every foundation initiative points to the use cases it enables and their value in kroner.
Smallest necessary foundation
Data and platform are built in step with the portfolio, not everything first.
The business owns the value
Ownership of data and use cases sits in the business.
Workflows, not tools
An initiative only counts once a process has changed and the effect has been measured.
A living strategy
The portfolio is reprioritised every quarter, and the strategy is reviewed together with the corporate strategy.
See the full framework behind a data and AI strategy, with the maturity model and the process week by week.
What it looks like in practice
The principles sound simple. It is in practice that they get skipped. Four examples from Ten pitfalls in data and AI strategies:
Ownership
AI was to fill in customer profiles and save about two hours per profile across tens of thousands of customers. No one knew the time spent today, and no one in the business owned the change. After several months of analysis with nothing in operation, it had become an IT project.
Foundation
Ask the control question: if the platform work stopped tomorrow, which use case would notice? If the answer is none, you have a foundation without a portfolio.
Adoption
A lead generation solution worked well, but the salespeople never used it, because no one measured whether they did.
Running costs
A process was automated with more than 20 agents, but only about a fifth of the cases went through without errors. Most of the token spend went on cases that created no value.
How do you build a data and AI strategy?
A data and AI strategy can be built in 12 weeks if every phase ends with a decision. The decision points are what separate a strategy that gets adopted from a report that gets read.
Mobilise (week 0-1)
Kick-off with the sponsor, scope, stakeholder map, collection of existing strategies and documents, and an interview plan. Decision G0: scope and sponsor are in place.
Understand (week 1-4)
15-25 interviews across the executive team, the business, IT, legal, compliance, HR and a few super users, an online survey of 50-200 employees, a review of the data and system landscape, and a map of the AI you already use and pay for, unofficial use included. The result is a maturity heatmap, a list of pains and a long list of opportunities. G1: the steering group approves the findings.
Prioritise (week 4-7)
An ambition workshop with the executive team, 3-5 use case workshops, scoring of use cases, an AI-ready data check of the shortlist and business cases. G2: the executive team chooses the ambition and the portfolio.
Design (week 7-10)
Operating model, governance and AI governance, target architecture, an AI literacy plan and regulation. G3: the design is approved.
Embed (week 10-12)
Roadmap, investment envelope, KPI tree, a plan for the first 100 days and a presentation to the executive team or the board. G4: go and funding.
Then execution begins: the first lighthouse projects start, governance goes into operation, the first wave of AI literacy rolls out and the benefits are measured. The portfolio is reviewed every quarter.
What does it take from you?
Sponsor on the executive team
1 hour a week plus four steering group meetings.
Core team
20-30% of their time for 12 weeks.
Interviewees and workshop participants
1 hour per interview and 2-3 half days per workshop participant.
The timeline only holds if that effort is set aside from the start. Involve the works council early when AI changes job content. It takes the air out of resistance later.
The tools: maturity, prioritisation and business case
Three tools carry most of the work: a maturity model that shows where you stand, a way to prioritise use cases and a business case the CFO can approve.
A data and AI maturity model in five levels
Each building block is scored from 1 to 5 per business unit and shown as a heatmap. The target is rarely 5 everywhere. You set the level the portfolio actually requires, and repeat the assessment after 12 months.
| Level | Characteristics | Typical sign |
|---|---|---|
| 1. Ad hoc | Data and AI are used by enthusiasts without a shared direction | Shadow spreadsheets, private AI accounts |
| 2. Experimenting | Pilots and tools, but no prioritisation or measurement | Many proofs of concept, few in operation |
| 3. Systematic | A shared platform, ownership and portfolio management | A use case process and business cases exist |
| 4. Scaled | Value is measured and reused, governance works in practice | AI in core processes, benefits are tracked |
| 5. Transformative | Data and AI shape the business model | New products and services built on data and AI |
How to prioritise AI use cases
Each use case is described on a card: the problem, the workflow that changes, the value driver and KPI, the baseline, the data needed, the owner, the risk class and an estimate for development and running costs. It is then scored on six weighted criteria:
| Criterion | Starting weight | The question |
|---|---|---|
| Value | 30% | What is the annual effect in money on revenue, costs, capital or risk? |
| Feasibility | 20% | Do you have the technology, the skills and the suppliers? |
| Strategic fit | 15% | Does it support one of the 3-5 goals? |
| Data readiness | 15% | Is the data ready, can it be within three months, or does it need a foundation project? |
| Risk and regulation | 10% | Which risk class under the AI Act, and how serious is an error? |
| Need for change | 10% | How much must workflows and roles change? |
The weights are a starting point that leadership adjusts in the prioritisation workshop. Then plot value against feasibility: high value and high feasibility are quick wins. High value and low feasibility are lighthouse projects that need a foundation first. Use cases with low value are parked, however easy they are. The templates for use case cards, scoring and business cases are in our Playbook.
Also choose the right type of solution: agents where decisions are needed, automation for routine workflows and assistants for simple lookups, as Gartner recommends. An agent may only act on its own once seven controls are in place.
AI-ready data is assessed per use case
Data is assessed for the specific use case, not in the abstract. Each use case on the shortlist ends up in one of three categories:
Ready
Can join the portfolio now.
Ready within three months
Can start with a short data sprint.
Needs a foundation project
Waits for the foundation in the roadmap.
A business case the CFO can approve
A CFO does not fund “high value”. Every use case that goes forward needs a sponsor, a named owner in the business and a yes from the functions that must put resources into it or live with the solution afterwards. The case itself has five parts:
Value drivers
More revenue, lower costs, released capital, lower risk.
Baseline and assumptions
Document them and have them approved by the benefit owner.
Costs
Development, operation (licences, compute, tokens, support) and change.
Phasing
Value arrives gradually: show cautious phasing and sensitivity.
Key figures
Net present value, payback period and a named benefit owner.
Don't forget the running costs. One in five companies limits its use of AI because of operating costs, tokens included (McKinsey, August 2026). So work backwards from what the process costs manually today: what may it cost at most to run one case through the solution before the business case turns negative? That number is the ceiling for running costs and tokens. Read more about the business case for AI.
Operating model: who does what?
Three models are worth assessing. Hub-and-spoke is the best starting point for most mid-sized and large companies.
| Model | Characteristics | Fits | Risk |
|---|---|---|---|
| Centralised | One team builds and runs everything | Low maturity and few use cases | A bottleneck, far from the business |
| Hub-and-spoke | A central team provides the platform, standards and specialists, and the business owns and builds with champions | Most mid-sized and large companies | Needs clear roles and decision rights |
| Decentralised | The business units build for themselves | High maturity and strong domains | Duplicated work and weak governance |
What you are left with
A decision document of 25-40 pages, plus the tools that keep the strategy alive once the advisor has gone.
| Tool | Your owner afterwards |
|---|---|
| Strategy on one page | Sponsor |
| Use case catalogue with scoring model | Portfolio owner |
| Business case model | CFO or controlling |
| Maturity assessment as baseline | Head of data and AI |
| Roadmap | PMO |
| Governance package and AI policy | Governance owner |
| AI literacy plan per role | HR |
| Plan for the first 100 days | Programme manager |
Before the strategy goes to the executive team, three things must be in place: the strategy on one page can be explained in three minutes, no recommendation stands without a value, an owner and a price, and the difficult recommendations have been aligned with the CFO before the final presentation, not after.
Example of a data and AI strategy on one page
This is the strategy on one page for Example Ltd, a fictional Danish manufacturer with 2,500 employees and four business units. All names and figures are made up, but the structure is the same as in a real strategy.
| Field | Example Ltd |
|---|---|
| North star | Data and AI make Example Ltd the most reliable supplier in the Nordics |
| Goals for 2028 | DKK 12 million a year in benefits from data and AI, 97% delivery performance (92% today), and 80% of managers and specialists using AI daily |
| Ambition | Efficiency in production and supply chain, and growth in sales and service |
| Investment | DKK 9.5 million over 24 months |
| Lighthouse projects | A complaints agent (quick win), automatic invoice matching (quick win), demand forecasting and predictive maintenance on line 3 |
| Foundation | Data products for customer, product and production, an agent and model platform with cost control, and sensor data from production |
| Governance | Hub-and-spoke with champions, a data and AI committee every quarter, an AI policy and an AI register |
| Measurement | A KPI tree with impact metrics, and quarterly reprioritisation |
| First 100 days | Committee set up and policy adopted, two lighthouse projects in pilot and the first wave of AI literacy under way |
Notice that every foundation initiative can be traced back to a lighthouse project, and that the goals are impact in money and delivery performance, not numbers of tools or users. See the full example of a data and AI strategy, with the maturity heatmap, use case portfolio, target architecture and process mapping.
Ten pitfalls in data and AI strategies
Most data and AI strategies do not fail on analysis, but on ownership and on the missing link between the value and the data and platform work it takes.
Recommendations without owners
No one in the business owns the change and the business case.
Use cases without hard numbers
Value is rated high, medium or low instead of in money.
Foundation without a portfolio
The platform is built before any use case needs it.
Too many initiatives
Everyone gets a little, and no one gets enough to succeed.
Solutions without adoption
The solution works, but no one uses it.
Technology choices without options
The platform is chosen before the need is known.
Activity metrics instead of impact metrics
You measure completed fields instead of additional sales.
Forgotten running costs
Tokens and operations only show up after go-live.
Agents without controls
The agent acts on its own before anyone has set limits.
A strategy that freezes
No one looks at it or measures it, and with new leadership you start over.
Read the ten pitfalls in full, with examples from practice and a health check of ten questions for your own strategy.
Should you build the strategy yourselves or get help?
There are two routes: you can drive the strategy yourselves with a method, materials and sparring, or you can have an advisor drive it for you. Many combine the two and start on their own before bringing in help for execution.
| Route | Fits | Duration | Price |
|---|---|---|---|
| Plan Lab: an online cohort of 6-10 leaders | You, if you want to build the strategy yourself | 10 weeks | DKK 29,500 |
| First 100 Days: 1:1 with a senior advisor | A new CDO, CDAO or data and AI leader | 100 days | DKK 65,000 |
| Playbook: toolkit and templates | Anyone who wants to start at their own pace | Own pace | From DKK 2,500 |
| Data and AI readiness check | Those who want to know whether the strategy should be updated or replaced | 3-4 weeks | Fixed price |
| Data and AI strategy, driven by us | Leadership teams that want the strategy built without using their own people | 12 weeks | Fixed price, paid at the decision points |
Typical occasions for building a data and AI strategy
A new CDO, CDAO or data and AI leader has been hired and needs to get going in the first 100 days.
A small or mid-sized company sets up a data and AI, analytics or BI function for the first time.
An established leader needs to renew the strategy and find the use cases worth the most.
There are many AI pilots and few in operation.
A new group strategy, an ERP change or a board asking about AI.
Frequently asked questions about data and AI strategy
What is the difference between a data strategy and an AI strategy?
A data strategy is about how data is collected, governed and made available. An AI strategy is about where AI should create value. In practice they belong together: AI without a data foundation does not scale, and a data foundation without concrete use cases loses support. That is why we recommend one combined data and AI strategy, tied to the digital strategy and the corporate strategy.
We already have a data strategy. Is that enough?
It depends on when it was written. A data strategy from before 2023 rarely covers AI, agents, the AI Act and the running costs of AI. A readiness check shows in 3-4 weeks whether it should be updated or replaced.
How long does it take to create a data and AI strategy?
A typical process takes 12 weeks across five phases with five decision points. Execution of the first lighthouse projects is agreed afterwards.
How many use cases should a data and AI strategy contain?
3-5 lighthouse projects: a few larger initiatives with measurable impact. Choose a domain, such as sales or operations, and a theme within it, such as cross-selling or demand forecasting. Each lighthouse can hold many smaller deliveries. The rest goes on a waiting list with a clear rule for when something can move up.
Who should own the strategy?
The executive team owns the ambition and the portfolio, usually through a sponsor. For each initiative, a business leader owns the change and the business case, while IT owns the implementation. The data and AI leader holds the portfolio together and measures maturity again after 12 months.
Does a data and AI strategy suit a small or mid-sized company?
Yes. The structure is the same, but the scope is smaller. Many smaller companies start when they set up a data, analytics or BI function for the first time. If you want to drive the work yourselves, Plan Lab is a cheaper place to start.
What does a data and AI strategy cost?
It depends on the scope, and on whether you drive the work yourselves. At Plan & Proof the strategy engagement has a fixed price, and you pay at the decision points: 20% at G1, 30% at G2, 30% at G3 and 20% at G4. If you want to build the strategy yourselves, Plan Lab costs DKK 29,500.
Should the AI Act be part of the strategy?
Yes, as part of governance, but not as the main argument. The AI literacy requirement already applies and is enforced from August 2026, and the high-risk rules are postponed to December 2027, not cancelled. The strategy should therefore include an AI policy, an AI register and an AI literacy plan per role. Read more about the AI Act timeline.
How often should the strategy be updated?
As often as the corporate strategy. Tie it to the same rhythm, reprioritise the portfolio along the way and track the impact metrics continuously. If new leadership arrives, revise it and replace only what no longer makes sense, instead of starting over. An annual update before the budget is a fixed way to do it.
About the author
Anders Butzbach Christensen is the founder of Plan & Proof. He has bought and owned data and AI strategies as a data and AI leader in several large companies, taken part in three strategy processes in international groups and built the data and AI strategy in two groups. He knows what survives contact with day-to-day operations.
Experience: Head of Data Engineering Global at The LEGO Group, Head of Data & Analytics at Terma, Director of Enterprise Architecture, Data & AI at Bunker Holding and Senior Director, Enterprise Architecture, Digital Innovation & AI at DOVISTA.
Read the interview with Anders at Hyperight too: Don't Build AI, Build Organisations That Can Use It, or see more interviews and talks.
Sources: Statistics Denmark via Ritzau, 15 Sep 2026; DI, Sep 2026; McKinsey, The state of AI, Aug 2026; BCG Applied AI Index 2026, Sep 2026; BCG, Jan 2026; Gartner, Feb 2025; Gartner, Jun 2025.
The service behind this article: Data and AI strategy →