How can companies deliver actual productivity gains from AI?

Several recent reports (such as these from MIT’s Nanda Lab and BCG) have made claims that the vast majority of companies have delivered little to no measurable business value from their investments in AI. These reports are, in my view, wide of the mark. They may have a point about the state of play today, but are misleading if read as a verdict on AI’s potential.

When a general purpose technology arrives, productivity rarely jumps overnight. The internet did not transform business in year one. It took time for firms to redesign processes, build infrastructure and rethink how work got done. AI is following a similar pattern.

We are still early in the adoption curve; the models are improving quickly, and most organisations are not yet changing how they operate in response.

The early stumbling blocks

The first mistake is using AI as a bolt-on rather than as part of the system. Many teams are using tools such as ChatGPT to draft emails, summarise papers or tidy up presentations, saving up to five or ten minutes at a time. It’s helpful, but it is incremental. It doesn’t redesign the workflow.

If your process still involves copying content into an LLM or chatbot, pasting the result back into a document and manually reviewing every step, you have not changed the underlying system. You have sped up one task.

The second mistake is waiting for the dust to settle. The pace of change can make it tempting to pause investment and hope the market stabilises. However, the firms that benefit most will be those that build now and improve as the models improve.

Where do the real productivity gains sit?

The biggest gains come when AI is integrated directly into core processes. That means stepping back and asking some yourself questions:

• What work is repetitive, rule-based and time-consuming?
• Where are we manually moving information between systems?
• Which tasks require judgement, and which are pattern recognition in disguise?

In many businesses, we spend large amounts of time triaging emails, reviewing documents, processing forms or preparing standard outputs from structured data. These are prime candidates for automation. The shift is from “AI helps me do my job faster” to “AI does this part of the job, and I oversee it”.

In practice, that often requires building bespoke tools that wrap around AI models and plug into internal databases and systems. Once that infrastructure is in place, model improvements deliver ongoing gains without redesigning the workflow each time.

To build or buy?

Not every firm can justify a large in-house AI team. Domain knowledge is key and off-the-shelf tools may solve 60–70% of a problem, which might be enough for smaller organisations.

But generic products rarely capture the full value available in a specialist business. The more complex and bespoke your processes, the stronger the case for internal capability or close partnership with technical experts who understand your domain.

The important question is not “are we using AI?”, it’s “where in our value chain does AI create defensible advantage?”.

How can AI drive innovation and profitability?

Efficiency is only one side of the equation. AI also creates opportunities to enhance client experience and develop new services. At LCP, we developed LCP Transpose [add link] to convert written reports into customisable videos and podcasts in real time. It allows clients to choose format, language and level of technical detail. That was not possible two years ago.

The underlying report is still written by experts. AI makes it more accessible and scalable. In doing so, it changes how clients engage with our insight.

Many organisations will find similar opportunities if they look closely enough. The constraint is usually imagination rather than technology.

What comes next

Today, most firms are focused on automating tasks. The next frontier is automating elements of decision-making. That raises harder questions. Where is it appropriate to delegate judgement to a model? What governance is required? How do you balance speed with accountability?

These are key strategic choices that will shape competitiveness over the next five to ten years.

Three practical steps to takeaway

If you are starting or reassessing your approach, I would suggest:

1. Map your workflows end to end. Identify repetitive, high-volume tasks before choosing tools.
2. Integrate, don’t experiment in isolation. Aim for system-level change rather than standalone use cases.
3. Invest with a long-term view. Model performance will improve. If you build the right infrastructure now, you benefit automatically.

AI is more than a short-term productivity switch; it is a structural shift in how knowledge work is done. The firms that see measurable gains will be those prepared to rethink their processes, not just their software.