The DiagnosisNo. 05

The challenge of integrating AI: from individual productivity to EBITDA impact

Buying licences can speed up tasks. Turning the time saved into greater capacity, higher margins or growth takes more.

AI-generated narration of this issue.

Many companies are approaching AI adoption and integration in the same way: they buy ChatGPT or Claude licences and wait for the impact to follow. Personal productivity may improve without a corresponding improvement in business performance.

A small glass prism casts a broad fan of light across concrete.

McKinsey found that 80% of respondents report improvements in their individual productivity. Yet only 37% see a positive impact on EBIT, and just 6% qualify as AI high performers.

The difference is not simply how much people use the tool. It is which part of the operation has changed as a result.

01

Two different kinds of progress

Individual use helps someone complete their tasks better and faster.

Impact at team or company level emerges when those improvements change a shared process and translate into greater capacity or better results for the organisation as a whole.

KPMG reveals a similar tension: 64% of US CEOs believe they are keeping pace with AI developments or staying ahead, but only 9% describe the impact of AI agents as transformational.

02

Where the transition tends to stall

Someone learns to use AI and starts producing work faster. Without a clear purpose or shared criteria, analyses, proposals and presentations pile up unused. Activity increases, but so does the clutter.

Even when technology saves time, a task completed in minutes can still be held up waiting for approval. Unless processes are redesigned and people agree how to use the capacity freed up, the improvement remains isolated: people may simply fill the time saved with more work without leading to higher sales, greater capacity or faster decisions.

If the aim is to transform the business, individual initiatives are not enough.

03

Three levels of integration

We can think of AI integration as a pyramid: the individual, the team and the company. At each level, the aim is to make AI central to how work gets done, organising information, tools and processes so that it can work with them. This moves us towards an AI-native way of working. Technology handles repetitive work; people set objectives, supervise and make decisions.

Fig. 1Towards an AI-native operation

Company Connectedoperations Team Shared processes Individual Individual tasks Widerreach

At the base is individual work. Someone learns to use AI, gives it the context it needs and completes tasks more effectively. It is a valuable first step. But if that knowledge remains confined to their own account and working methods, the benefit remains personal.

The second level is the team. Moving up means organising shared work: bringing information together, agreeing standards and defining responsibilities and processes. AI can draw on that context and carry out tasks within a common process. One person’s knowledge starts to become a capability the whole team can use.

The third level is the company as a whole. AI, software and automation connect functions and processes, with shared information and defined controls. People set objectives, supervise, handle exceptions and bring judgement and tact to their relationships with colleagues, customers and suppliers. In future editions, we will look at how this is working in practice.

The challenge for leaders moving up this pyramid is to reorganise work and help people understand and take part in the transformation. That means deciding which process to change first, what information it needs, who is accountable for the result and how to extend what works.

At Fluency Labs, we believe this transformation rests on three pillars: understanding the technology, including its capabilities and limits; identifying and defining KPIs linked to EBITDA, to direct effort towards business outcomes; and active involvement from the team, to redesign processes and make the changes part of everyday work.

04

From use to measurable results

At any of these levels, improvement needs a cycle: identify an opportunity, redesign the work, make use of the capacity freed up, measure the result and start again.

For an individual, this might mean preparing an analysis. For a team, the month-end close. For a company, the entire process from making a sale to collecting payment. The scope and coordination change; the discipline of improvement remains the same.

Progress is demonstrated when the new way of working delivers a result: fewer delays, better quality, more capacity or a stronger margin.

Fig. 2Measurable business impact

  1. 01 Identify

    Process and baseline.

  2. 02 Automate

    Redesign before accelerating.

  3. 03 Use the freed capacity

    Decide its next use.

  4. 04 Measure and adjust

    Compare with the baseline.

  5. 05 Start again

    Apply what you have learnt.

A cycle that learns.
  1. Identify. Choose a process with significant volume or friction and establish its baseline: time, cost, quality, errors or capacity.
  2. Redesign and improve efficiency. Remove unnecessary steps, organise information and automate repetitive tasks. The aim is not to speed up a poorly designed process.
  3. Reallocate capacity. Decide in advance how the time saved will be used: handling more cases, improving service, selling, analysing or reducing delays.
  4. Measure and learn. Compare the result with the baseline. Review exceptions, errors and new bottlenecks.
  5. Update and repeat. Turn what you have learnt into better rules, information sources and ways of working. Then apply what works more widely.

The centre of the cycle is not adoption. It is measurable business impact.

In the same study, McKinsey highlights a telling difference: nearly three-quarters of AI high performers substantially redesign their workflows, compared with roughly a quarter of everyone else. They do more than add AI to existing work. They change how the work gets done. That is the key point.

05

A simple test

Before approving the next licence, training programme or initiative, complete these four lines:

  • The process we want to change:
  • The capacity we expect to free up:
  • How we will use that capacity:
  • The metric that should improve:

If all we can say is how many people will use the tool, we have an adoption plan without a strategy. Not a plan for impact.

The question is not simply how many people use AI. It is which process changed, what capacity we freed up and what business result it produced.

Thank you for reading. Until next time.

Written by

Gonzalo GilgesHead of Practical AI

This issue

AI Firsthand No. 05, The Diagnosis, by Gonzalo Gilges.

Cite as: Gonzalo Gilges, “The challenge of integrating AI: from individual productivity to EBITDA impact”, AI Firsthand No. 05, Fluency Labs.

AI Firsthand

A weekly look at what worked, what did not and the conditions behind the results.

Get the next issue by email.

Subscribe to AI Firsthand