The Self-Improving Company

The learning loop in transformation is far too long.

You can spend months preparing a business case for a programme that will take years to deliver. You have to predict the benefits before you’ve had much opportunity to test the changes. Then, once the investment is approved, an enormous amount of attention goes into delivering what was promised.

Meanwhile, the question that justified the whole thing can become surprisingly difficult to answer: is this actually making the business work better?

By the time the answer becomes clear, a lot of money has been spent. People have learned new systems, reorganised their teams and worked around things that didn’t quite land. The next programme may already be underway.

I think we should expect something much better. A company should be able to understand how it works, identify the changes most likely to help, and see their effects quickly enough to adjust. It should be able to keep doing that as its goals, people and technology change.

That is what I mean by a self-improving company.

What is actually happening to the work?

As work is shared between humans, AI agents and software, understanding the business becomes a different challenge.

An organisation chart tells you who is responsible. A process map tells you how something is supposed to happen. A dashboard might tell you how often a new AI tool is being used. Those are useful views, but they don’t tell you enough about the work itself.

Who is doing it? What are the agents taking on? Where does human judgement matter? What has become easier, and what now needs checking? Where is effort going into serving customers, building products or solving problems—and where is it disappearing into coordination, administration and avoidable friction?

We’ve encountered this in our own business. We use AI heavily, but looking closely at our work exposed how much operational overhead remained. Keeping work organised, reconciling statuses, preparing handoffs, maintaining references. Agents getting stuck because they lacked access or needed a permission resolved. Humans stepping in to get things moving again.

What surprised me was the difference between how we thought we were working and what the evidence showed.

That matters because it changes what you choose to fix. If an agent keeps stopping because of inconsistent permissions, giving it a more capable model may do very little. If people spend their time reconciling conflicting records, automating another isolated task may leave the underlying problem untouched.

The more work we delegate, the more important it becomes to understand the whole arrangement.

Imagine a customer team using AI to prepare responses. Drafting becomes faster. But people may spend longer checking those responses, finding missing information or resolving exceptions. Alternatively, they may genuinely gain more time for useful conversations with customers.

Both are plausible. A measure of drafting speed won’t tell you which is happening.

An accountable leader needs to understand whether the team has become more effective—and what is preventing it from becoming more effective still.

Understanding should lead somewhere

The purpose of this understanding is to make better decisions about the business.

A leader trying to improve customer relationships needs a different view from one trying to reduce service costs or accelerate product development. The same activity can be essential in one context and unnecessary in another. Review can protect a customer from a serious mistake, or it can be a habit nobody has questioned.

So understanding work has to connect to intent: what are we trying to achieve, what is getting in the way, and what are we prepared to change?

It also has to include the people doing the work. They know why a workaround exists, where a system fails them and which apparently inefficient steps prevent bigger problems. They should be able to recognise the picture being presented and challenge what it misses.

The ambition is deep understanding of the operating system formed by humans, AI and software. That requires trust. People should be able to participate in improving their work without feeling that every observation will become an assessment of them individually.

Done well, this creates a much more useful conversation between leaders and teams. They can look at the same evidence, understand the problem and decide what to try.

Closing the transformation loop

Once you have that understanding, the approach to change can become much more direct.

Start with the outcome you want. Look at how the relevant work happens today. Identify an intervention with a credible reason behind it. Make the change, establish that it has actually reached the work, and examine what follows.

Did it release useful capacity? Did customer-facing time increase? Did it remove friction, or move it somewhere else? Has the team gained confidence in what it can delegate? Are people spending more time on the work that matters to the business?

Then use what you learn.

Expand the change if the evidence supports it. Adjust it if the result is mixed. Stop it if it is creating more problems than it solves. Investigate further when you still don’t understand what happened.

This is closed-loop transformation: the consequences of change feed back into the next decision.

It gives leaders a way to steer while the work is happening. They don’t have to wait for a programme to finish before discovering that an assumption was wrong. A disappointing intervention can still be useful if it exposes the real constraint early enough to do something about it.

There will still be major investments, architectural decisions and changes that take time to show results. But even a large programme contains assumptions that can be tested along the way. Its scale should make learning more important, not postpone it.

A capability of the company

The larger opportunity is to make this an enduring capability of the business.

Today, understanding how an organisation works can be a substantial piece of work in itself. It is often assembled for a particular decision or programme, then becomes harder to maintain as the business changes.

Imagine instead being able to keep that understanding close to the way the company actually operates.

A leader could see where a strategic priority was failing to translate into day-to-day work. A team could identify which parts of a workflow were ready for AI and which needed better information or clearer decisions first. After a change, they could see whether the expected improvement was appearing and where further attention was needed.

Each intervention would leave the company better informed about how to achieve its goals. Successful changes could be expanded with a clearer understanding of the conditions that made them work. Problems could be addressed before they became another expensive lesson at the end of a programme.

Leadership would still involve judgement, ambition and difficult choices. But those choices would be grounded in a much deeper understanding of the business, with a much shorter distance between deciding, acting and learning.

That is the company we want to make possible with Teho.

A company that understands what its humans and AI are doing, how their work connects to its goals, and what needs to change. A company that can see the effects of those changes and respond.

Improvement becomes something the business knows how to do continuously—not something it has to wait for the next transformation programme to deliver.