The hidden data investment behind AI

New evidence from UK businesses suggests that the productivity benefits associated with Artificial Intelligence may begin well before a firm adopts AI.

Artificial intelligence does not run on algorithms alone. AI systems must be trained on data, operate using data and, in many cases, generate new data as their main output. Businesses therefore need reliable information, suitable software and employees with the right skills before they can use AI effectively.

These foundations are often overlooked. Investment in data rarely appears as a separate item in company accounts, and official economic statistics capture it only partially. As a result, estimates of the economic benefits of AI may miss an important part of the story: the investments businesses make to become “data ready” before adopting AI.

In a new paper, co-authored with Larissa da Silva Marioni, Ana Rincon-Aznar, we examine this hidden component of the digital economy. Using a specialised survey of almost 2,000 UK businesses, the study investigates which firms invest in data, how they use it and whether stronger data capabilities are associated with higher labour productivity.

Its central finding is clear: simply collecting data is not enough. The largest productivity advantage is found among firms that use data for advanced, outward-looking and scientific activities, and that have employees with the skills needed to perform those tasks.

What does investing in data mean?

Data investment involves more than purchasing a database. It includes the software, organisational systems and employee training required to collect, organise, analyse and apply information. Our study draws on the UK Data Use and Productivity Longitudinal Survey, launched in 2022–23, funded by UK Department for Science, Innovation and Technology. The survey gathered information on businesses’ investments in software and database management, raw data and databases, and data-related employee training. It also asked how firms used their data and what skills their workers needed to perform data-related tasks.

Basic activities included collecting and handling information, for example, recording sales, processing payroll or maintaining customer contact details. More advanced activities involved analysing or sharing data. The study also classify data according to its operational purpose. Internal uses included accounting, human resources and stock management. External uses included analysing customer behaviour, conducting market research and monitoring economic or environmental conditions. Scientific uses covered activities such as research and development, product development, sensor data, geolocation and specialised technical analysis. Finally, we distinguish between routine and high-skilled data tasks. Routine tasks included data entry and simple calculations, while high-skilled activities involved advanced analysis, programming languages such as Python, complex simulations and specialised software used in sectors such as engineering, architecture and manufacturing.

Data use is widespread but not always productive

Data use is already common among UK businesses. Approximately 94% reported at least some basic use of data, while around half used data for external or analytical purposes. However, the productivity differences between firms depended heavily on how the data was used. Businesses limited to basic collection and internal record-keeping did not display a clear productivity advantage over comparable firms that did not undertake these activities. Collecting information may be necessary for running a business, but it does not automatically translate into greater efficiency or higher output. The picture changes when firms used data for more sophisticated purposes. Businesses applying data to external operations, such as understanding customers and markets, or to scientific and technical activities record significantly higher output per worker. The same pattern emerges for employee skills. Data activities were positively associated with productivity when they involved highly skilled employees. Routine activities requiring only basic or intermediate skills did not produce the same clear advantage.

Overall, the paper reports that labour productivity among firms undertaking advanced, external or high-skilled data activities was around 1.4 to 1.5 times the level observed among firms without these data investments. The key message is therefore not that “more data is always better.” Data creates economic value when businesses have the tools, skills and organisational capabilities to use it effectively.

The study also identifies some of the investments that help firms move from routine record-keeping to more productive uses of data. Investment in digitised information, particularly software and database-management systems, was strongly associated with more advanced data activities. Simply accumulating raw data was less consistently related to sophisticated use. This suggests that the infrastructure used to organise and process information may matter more than the volume of information collected.

Employee training also played an important role. General data training was associated with a broad range of data activities. Specialist training in coding, advanced software and analytical methods was particularly closely connected with scientific applications, big-data methods and high-skilled tasks.

The study estimates that businesses investing in digitised information and data-related training were around 20% to 40% more likely to use data for external or scientific purposes, undertake advanced applications and perform skill-intensive data activities. These results reinforce a wider lesson from the history of technological change: new technologies rarely deliver their full benefits in isolation. Their value depends on complementary investment in skills, software and organisational practices.

What does this tell us about AI? Implications for businesses and policymakers

Before adopting AI, a firm generally needs to digitise its information, build reliable databases, establish appropriate data-governance procedures and employ people capable of analysing that information. Businesses that have already developed these capabilities are likely to be better placed to select, implement and benefit from AI tools. Some of the productivity gains commonly associated with AI may therefore originate further upstream. They may partly reflect data-related investments made before AI adoption becomes visible. This helps explain why purchasing an AI tool does not necessarily produce an immediate improvement in business performance. If a company’s data is incomplete, poorly organised or inaccessible, or if its employees lack the necessary skills, the technology may be difficult to integrate into everyday operations. AI readiness is therefore not simply a question of acquiring the latest software. It is the result of a longer process of building data infrastructure, human capital and organisational knowledge.

For business leaders, the findings suggest that an effective AI strategy should begin with an assessment of the company’s data foundations. Does the business have reliable and accessible information? Can data from different parts of the organisation be combined? Are appropriate governance arrangements in place? Do employees have the skills to interpret data and use it in operational and strategic decisions? Investment in these areas may be just as important as investment in AI itself.

The findings also matter for public policy. Government programmes that subsidise the purchase of AI tools may generate limited returns if businesses lack the underlying infrastructure and skills required to use them. Support for AI adoption may be more effective when accompanied by investment in digitalisation, data management and workforce training.

This issue may be especially important for smaller businesses. Building data infrastructure and recruiting specialist workers involve substantial upfront costs. Smaller firms may have less financial and managerial capacity to absorb these costs, potentially widening the productivity gap between them and larger companies.

There is also a regional dimension. Advanced data and AI skills in the UK are concentrated in areas such as London, Oxford and Cambridge. If firms elsewhere have less access to these capabilities, the diffusion of AI could reinforce existing regional inequalities.

Finally, official statistics need to capture data investment more accurately. Better measures of spending on data infrastructure, governance and skills would provide a more complete picture of business investment and help policymakers evaluate the true economic contribution of AI. The study provides detailed evidence about UK businesses at a particular point in time, but it does not yet track the same firms over many years. The findings therefore demonstrate a strong association between advanced data capabilities and productivity; they do not conclusively prove that data investment alone caused the observed productivity differences. More productive firms may have greater resources to invest in data, while other characteristics could influence both investment and performance. The researchers use econometric methods and a wide range of business characteristics to address these concerns, and the results remain similar under alternative specifications. Nevertheless, evidence collected over a longer period will be required to establish the direction of causality more firmly.

Future survey waves should make it possible to test whether firms with stronger data foundations are more likely to adopt AI and whether they achieve greater productivity benefits after doing so.

• Marioni, d S. L, Rincon-Aznar, A., Venturini, F. (2027)  Data Investments and Productivity: Business-level Evidence from the UK, NIESR N.