The MIT study The GenAI Divide shows that only a fraction of GenAI pilot programmes deliver real economic impact – the majority stagnate.

The MIT initiative's research reveals a clear divide: ‘About 5% of AI pilot programmes achieve rapid revenue acceleration; the vast majority stall, delivering little to no measurable impact on P&L.’ The greatest successes are not in marketing and sales, but in the back office. The results also show that purchased solutions and partnerships are more successful than in-house developments and that adoption is more successful when line managers drive the implementation. In this interview, the findings of the study are classified and linked to my own perspectives on modern data strategy, in particular on the role of governance and data & AI literacy as enablers.

Question:

The MIT study ‘The GenAI Divide’ shows that the greatest ROI for GenAI projects is achieved in the back office, not in sales or marketing. Why is that so surprising?

Answer:

Because many companies have primarily invested in sales and marketing in the hope of achieving additional growth through data-driven campaigns or new customer interactions. However, the study makes it clear that automation in the back office in particular brings greater efficiency and cost advantages. There, outsourcing can be reduced, processes streamlined and throughput times accelerated – effects that are immediately reflected in the P&L. This is surprising insofar as the buzzword ‘data-driven marketing’ has dominated for years, while GenAI is still struggling in this area. The reasons for this are quality risks, reputation issues and sensitivity in customer communications. In the back office, on the other hand, processes are more clearly structured at a higher level and are therefore better suited to delivering fast and measurable results.

Question:

But automation in the back office is not a new topic. Why haven't traditional approaches such as RPA or digitalisation already achieved this?

Answer:

It is true that automation is not new there. However, technologies such as RPA were heavily rule-based and could only work reliably in very stable, structured processes. In reality, back-office processes are standardised at the macro level – every finance or HR department is familiar with tasks such as ‘paying invoices’ or ‘managing contracts’ – but at the micro level, there are enormous differences. Different invoice formats, varying workflows depending on the country or department, and individually formulated contract clauses quickly pushed classic, rule-based solutions to their limits. This is exactly where GenAI comes in: it understands language and context, recognises patterns in unstructured documents or emails, and can handle exceptions more flexibly. This means that, for the first time, even those processes that were too complex and varied for RPA can now be automated.

Question:

That sounds contradictory: on the one hand, the processes are standardised, but on the other hand, they are highly varied. How does that fit together?

Answer:

It's not a contradiction, but a question of perspective. At the macro level, the processes are clearly defined and comparable – invoices must be checked, payments approved, contracts filed. At the micro level, however, i.e. in detail, there is enormous variance. Every invoice looks different, every contract contains individual clauses, every location has its own approval processes. This diversity was almost impossible to manage with traditional systems. GenAI closes this gap because it recognises semantic relationships and can work reliably even with unstructured data.

Question:

Why don't companies implement back-office automation more consistently despite this potential?

Answer:

There are several reasons for this. First, the back office is still primarily perceived as a cost centre and therefore receives less strategic attention than customer-facing areas. In addition, finance, HR and procurement departments are often weaker in organisational terms and have less budget and sponsorship at board level. Another reason is employee fear: many fear being replaced by automation and offer open or covert resistance. Technically, the situation is further complicated by the fact that processes have evolved over time, are poorly documented and are characterised by heterogeneous legacy IT. Finally, there is often a lack of clear efficiency metrics, making it difficult to substantiate business cases. Taken together, this shows that the challenge lies less in technology and more in organisation, culture and prioritisation.

Question:

How do these results compare to the classic data strategy approach?

Answer:

There are clear similarities, but also shifts in emphasis. Both the classic data strategy and the MIT study emphasise the importance of business value-oriented use case selection, deep integration into processes and the development of competencies. However, there are differences in the role of governance. Modern data strategy does not see governance as a stumbling block, but explicitly as an enabler: it creates standards, ensures data quality, defines responsibilities and thus lays the foundation for digitalisation and the successful use of AI. The MIT study, on the other hand, documents that many executives perceive governance as more of a blockage and use it as an explanation for why projects are not progressing more quickly. This does not mean that governance is unimportant in the eyes of the study, but merely that it is often misunderstood in practice. Without governance, any GenAI initiative remains piecemeal, but governance must be thought of as an enabler, not a stop sign.

Question:

Does that mean the MIT study underestimates the importance of governance?

Answer:

Not really. Rather, it shows that governance is often misclassified in companies. Instead of being seen as the foundation for standardisation and scaling, it is perceived as an obstacle. In reality, however, the problems lie in integration, resource alignment and adoption. Governance remains absolutely central – but it must be designed to bridge the gap between security and innovation.

Question:

How must governance be conceived in the context of AI so that it actually acts as an enabler?

Answer:

Governance becomes an enabler when several dimensions interlock. First, standardisation is needed: clear data models, processes and interfaces form the basis for any form of automation. Equally important is transparency about who is allowed – or even should – use which data for what purpose in order to make informed and data-inspired decisions. Governance therefore means not only preventing misuse, but also actively providing guidance and making data usable in terms of value creation and responsibility. Furthermore, it must not act as a brake on innovation, but should rather create a secure framework within which new applications can be developed responsibly. Finally, data and AI expertise plays a central role: managers and employees must not only be able to understand and use data, but also develop a realistic picture of artificial intelligence. This includes knowing what AI can actually do, where its limits lie, how bias and hallucinations arise, and how results should be critically questioned. Only when these elements work together does governance become the basis for responsible, scalable and effective use of AI.

Question:

If classic data strategy governance places such strong emphasis on this, how does a modern data strategy differ?

Answer:

A modern data strategy shifts the focus. While classic approaches have emerged strongly from a risk-oriented perspective – creating order, setting standards, ensuring compliance – today the focus is more on highlighting the value contribution. Modern strategies are more dynamic, more closely linked to specialist areas and integrate data and AI from the outset. They understand data as a product that is maintained, developed and used – with clear ownership and measurable benefits. Governance remains the foundation, but not as an end in itself, rather as an enabler so that data products and AI solutions can be scaled reliably and trustworthily. In addition, topics such as data and AI literacy are coming to the fore: companies must empower their managers and employees to understand the opportunities and limitations of AI in order to make responsible decisions.

Question:

What does this mean in concrete terms for companies that want to sharpen their data and AI strategy now?

Answer:

Specifically, this means: Companies should reorder their priorities and not just focus on shiny marketing or sales use cases, but specifically address back-office processes where quick and measurable successes can be achieved. They should also leverage partnerships and not develop everything themselves, as specialised providers are often more successful. It is equally crucial to involve the specialist departments more closely and build up expertise, rather than just anchoring AI in central labs. Governance must be designed in such a way that it enables standardisation and trust, but at the same time allows for innovation. All in all, a modern data strategy means balancing business value, governance, literacy and empowerment – this is the only way to not only try out GenAI in pilot projects, but also to achieve real economic impact.

Summary:

The MIT study confirms many principles that are also anchored in classic data strategies, but shifts the emphasis towards decentralisation, back-office focus, buy vs. build and organisational learning. Governance remains the connecting foundation – already understood as an enabler, but often misunderstood as a blockage in practice. A modern data strategy builds on precisely this foundation, but places greater emphasis on value contribution, data and AI literacy, product thinking and the integration of AI. Companies that combine these elements create the basis for a data- and AI-driven organisation that generates sustainable value.

More articles on related topics:

• Advisory | Impulse Talks | Trainings
• Data to Transition: The superiority of Digital AI-Driven Ecosystems
• Why do I rarely talk about AI and a lot about data-driven?
• Be a Data Punk
• Holistic Data Management Model 5.0