# Data Strategy as a Prerequisite for AI

Author: Marin Niehues, Niehues Consulting · Published: 2025-05-12 · Updated: 2026-11-16 · AI Strategy & Business Case
Source: https://www.niehuesconsulting.com/blog/data-strategy-as-a-prerequisite-for-ai/

For every AI application that works with company data, a data strategy is the prerequisite, because it defines which data is reliably available for which decision. General-purpose assistants for writing and research work without one. As soon as AI is expected to analyze customer histories or contracts, however, the data estate determines the result. For the executive team this implies a fixed sequence: use cases come first, models and platforms last. This article sets out what a data strategy defines, how to assess your data estate in four steps, what building one costs and how the strategy becomes delivery.

## What is a data strategy?

A data strategy is the binding plan that sets out which data a company provides for which business decisions, who is accountable for it and under which rules it is used. It answers four questions. Which business decisions should improve through data? Which data is needed for them? Who is accountable for that data? Under which rules is it made available?

Where these answers are missing, a familiar pattern emerges. Each function maintains its own spreadsheets, while each AI application builds its own data connection. Before long there are several versions of the truth about the same customer or product, and every further application becomes more expensive by the cost of reconciling them. An interview study with 65 experienced data scientists and engineers names the lack of suitable data as one of five root causes of AI project failure.

A data strategy treats data as an operating asset. Like a machine or a contract, it has an owner and a maintenance rhythm. Each new AI application then builds on a governed stock of data and starts with a head start.

**A data strategy defines which data must be available for which decision and who is accountable for it. Every AI application inherits the maturity of that arrangement.**

The strategy also addresses permissibility: which data may be used for which purpose, for instance where it relates to individuals? This article provides an overview and does not constitute legal advice. Case-by-case assessment belongs with your legal department or your legal advisers. The data strategy ensures that this assessment takes place at a defined point and that its outcome is documented alongside the data.

## Data strategy and data governance have separate jobs

Data strategy, data governance and AI strategy are often conflated, yet they answer different questions. The AI strategy sets out which business objectives a company pursues with AI and in which order it implements applications. The data strategy derives from this which data those applications need and how the estate gets there. Data governance is the set of rules that makes the data strategy effective day to day: decision rights over data and the quality rules that apply to each data set. Without data governance, a data strategy remains a document. If you want to settle objectives and priorities on the AI side first, the article on [AI strategy for companies](/blog/ai-strategy-for-companies/) describes the approach.

## Four steps from use case to data requirement

A data strategy becomes effective when it is developed backwards from the use case. The approach follows four steps:

1. Define the use case and the target metric. A use case is only fully described once it is clear which decision or process changes and which metric shows success.
2. Derive the data requirement. The use case determines which information the application needs at the moment of decision and in which form.
3. Assess the data estate. Comparing requirement and estate reveals the gaps.
4. Close the gaps by value contribution. Data work follows the benefit of the use case it unlocks.

For the third step, an assessment against five criteria has proved its worth. The international standard on data quality describes fifteen quality characteristics. To decide whether a use case is ready to start, these five are sufficient:

- Availability: does the data sit in a system the application can access?
- Timeliness: does the data's currency match the pace of the decision?
- Completeness: are the fields required for the use case populated?
- Consistency: does the same term mean the same thing in every source?
- Permissibility: is use for this purpose clarified?

The result is a data map for each use case with a traffic-light status. Green means all five criteria are met and the prototype can start. Amber means a gap will be closed with limited data work within the initiative. Red means availability or permissibility is missing and the use case is deferred. The map shows the executive team which applications can start immediately and which require data work first. The same assessment forms the basis for a company's [AI readiness](/blog/ai-readiness-technology-data-organisation/) and provides the sequence for the roadmap.

**Data product and data mesh**
A data product is a data set with a named owner. Its meaning is documented, its quality is defined, and access is regulated. Applications draw on data products and no longer reach into individual tables. The term comes from the data mesh approach that Zhamak Dehghani described in 2019. The number of data products starts small: a company begins with the data sets its first use cases require.

## Data ownership belongs in the business function

The most frequent weakness of data strategies lies in responsibility. If data work is assigned to IT alone, the business decision is missing on what a customer status, a stock level or a contract end actually means. Only the function that owns the process can make that decision.

A division of roles across three levels has proved effective. The data owner is a manager in the business function. The owner decides on the meaning and use of the data, releases it to applications and is accountable to the executive team for its quality. The data steward works operationally, maintaining definitions, checking quality values in daily business and triggering corrections in the source systems. A small central data team sets standards and operates the shared platform. With fewer than three use cases, a specialist in the team takes on the steward role alongside other duties.

Two common routes to a data foundation can be compared with a third:

**Three routes to a data foundation for AI**
Criterion | Data project per AI application | Central data platform without business link | With Niehues Consulting
Starting point | The individual application | The technology | The decisions and processes that should improve
Data ownership | Temporary project team | IT | Data owner in the business, standards set centrally
Time to first application | Short | Long | Short, because the first use case defines the data scope
Reuse | Low | High, but without link to benefit | High, because data products grow along the use cases

## How much does a data strategy cost?

The cost of a data strategy consists of consulting days and internal working time, and external support is typically billed in the German market at around 1,000 to 2,500 euros per consultant day. Scope determines the total: a strategy for the first two or three use cases is a manageable undertaking, whereas a company-wide inventory of every data source is a large one. External days multiplied by the day rate give the consulting share, and internal days are valued at the internal daily cost rate.

**Worked example with assumed values**
- 20 external consultant days at 1,500 euros come to 30,000 euros.
- Three data owners with six days each plus eight days from IT come to 26 internal days.
- At an internal daily cost rate of 600 euros, this equals 15,600 euros.
- The data strategy for three use cases therefore costs around 45,600 euros.

The data work that follows usually outweighs the strategy itself, namely connecting and cleansing the data sets. These costs belong in the business case of each use case, because that is where a concrete benefit offsets them. A data strategy that sets out to cleanse every data set before the first benefit ties up budget without return.

## What does a data strategy template look like?

A workable template fits on a few pages and runs from benefit to data estate. Many companies use the framework of DAMA, the international professional association for data management, as a reference for roles and rules. For the strategy itself, the following checklist is sufficient:

1. Target picture: which business outcomes should change measurably through data and AI?
2. Use cases: which two or three applications start first, and with which target metric?
3. Data requirement: which information does each application need at the moment of decision?
4. Data map: how does each required data set score on the five criteria, with traffic-light status?
5. Data products and roles: who is data owner and who is data steward for each data set?
6. Standards: which definitions, quality thresholds and access rules are binding?
7. Permissibility: where is the assessment of use documented, and who signs it off?
8. Roadmap: in which order are the gaps closed, measured by the benefit of the applications?

An example from sales: for AI-supported prioritization of renewal conversations, the data map shows contract data in green and service history in amber because of inconsistent categories. The roadmap therefore starts with contract data and harmonizing the service categories.

## From strategy to delivery

A data strategy delivers its value once an application in operation draws on it. The strategy therefore belongs in the same hands as the implementation. [Niehues Consulting](/) combines AI management with delivery in operations and works according to the Niehues Delivery Model, with the phases Understand, Build and Roll Out.

In Understand, business analysis and requirements are produced, together with a business case for each use case. The first test question is: which business outcome should change measurably? The data requirement follows from the answer, and the sequence of applications is set by benefit and data maturity. In Build, prototypes follow, tested on whether they prove their benefit in the real process with real data. This is where the data gaps that stay hidden on paper become visible. In Roll Out, the solution moves into operations with internal or external teams, and the test question is who is accountable for operations from day one.

Founder Marin Niehues is a management consultant who works from the delivery perspective, as delivery manager, project manager or product owner. His projects range from VC-funded scale-ups to DAX 40 corporations in 20+ countries on six continents. The full path from the first use case to regular operations is described in the article [From pilot to scale](/blog/from-pilot-to-scale-ai-operations/).

The entry point is a free initial consultation of 30 minutes. It establishes which use case offers the best starting point and which data it requires. You can [request an appointment directly](/#kontakt).

**The sequence decides**
The executive team sets the use cases first. The data requirement follows, then the assessment of the estate and the allocation of ownership. Platform and model choice come as the last decisions. This produces data work with a clear link to benefit, and every application builds on data that is actively maintained.

**Sources**
- Ryseff, De Bruhl, Newberry: The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, RAND Corporation, 2024
- ISO/IEC 25012: Software engineering, Data quality model, 2008
- Bundesverband Deutscher Unternehmensberater: Honorare in der Unternehmensberatung, 2022

## Frequently asked questions

### What should a data strategy include?

A data strategy contains the prioritized use cases, the data requirement derived from them, the assessment of the existing estate and the responsibilities for each data product. It also sets standards for definitions, quality and access, together with a roadmap that closes the gaps.

### How long does it take to develop a data strategy?

For two or three use cases, a first draft with a data map is available after four to eight weeks. The strategy develops in parallel with the first use cases, and the data products grow with each application. A multi-year preparatory project before the first benefit is not required.

### Does a company need a data warehouse or data lake before starting with AI?

No. To start, the data for the first use case in a reliable form is sufficient. A central platform pays off once several applications use the same data. The decision follows from the use cases and the existing IT landscape.

### Who is responsible for the data strategy?

The executive team sets the objectives and use cases. The business functions take ownership of their data, and IT provides the platform. Delivery management holds the work together and ensures that the data work is ready in time for each use case.

### What is the difference between a data owner and a data steward?

The data owner decides, the data steward maintains. The owner is a manager in the business function who defines the meaning and use of the data and releases it to applications. The steward ensures in daily business that definitions are followed and quality defects in the source systems are fixed.

### What is data mesh?

Data mesh is an approach to data architecture and organization that places accountability for data in the business domains. Each domain provides its data as a product, while a shared platform and common rules hold the whole together. Companies with only a few use cases can adopt the principle of the data product without rebuilding their entire architecture.

### Do you need a data strategy for ChatGPT or Microsoft Copilot?

General use as a writing and research assistant does not require a data strategy, although it does require clear usage rules. As soon as an assistant accesses internal documents or business systems, the data strategy becomes a prerequisite. Access rights and the upkeep of the data then determine whether the answers are correct and whether confidential content stays protected.
