What Companies Getting Value From AI Do Differently

Almost every business you talk to is now using AI in some form. Somebody is drafting with it, somebody else is summarising calls, and there is usually a subscription nobody can quite account for.

Ask what it produced last quarter and the room goes quiet.

That gap between adoption and result is the defining feature of where we are. It is not that the technology does not work. It is that most deployments were never set up in a way that could produce anything measurable, and the reasons are consistent enough to be worth naming.

The capability is not the constraint

Stanford’s AI Index, published in April 2024, reviewed the research on assisted work and found that AI both speeds up task completion and improves the quality of the output. The finding that matters most for a business of your size is where the gains land: they are largest among people who were previously less skilled at the task.

That is not a marginal detail. It means the effect is not simply making your best person faster. It raises the floor, which is the actual constraint in a company where one experienced individual is the bottleneck on everything and where hiring their equal takes eighteen months.

So the capability is established. What separates businesses producing results from those producing subscriptions is entirely in how it was applied.

Failure one: it was scoped to impress

The most common mistake is choosing the use case that demonstrates well rather than the one that costs money every week.

Impressive use cases tend to be rare events. Something that happens twice a quarter, involves several departments, and makes a good slide. The system gets built, it works in the demonstration, and then it runs four times a year. Nobody develops a habit around it, nobody notices when it drifts out of date, and within a year it is switched off without ceremony.

The businesses getting value chose something boring that happens forty times a week. Lead responses. Document chasing. Report assembly. Routing. The payback is measured in weeks rather than years, and because it runs daily, problems surface immediately instead of at the next quarterly review.

Failure two: it was never connected

The second pattern is a tool that produces good output which then has to be moved somewhere by a person.

A system that drafts an excellent reply which somebody must copy into the CRM has not removed work. It has relocated it and added a step. The same is true of a summary that lands in a document nobody reads, or an extraction that produces a spreadsheet requiring manual import.

This is why integration is not a technical detail at the end of the project. It is where the value is created. A modest capability wired directly into the system where the work happens beats an impressive one that requires a human courier, every time.

Failure three: nobody owned it afterwards

Processes change. A supplier alters a file format, a stage gets renamed, the business starts handling a job type that did not exist when the system was built.

If nobody is responsible for noticing that the system is now doing the wrong thing, it degrades quietly. It does not announce this. It carries on producing output that is subtly less correct until somebody loses confidence and stops using it.

The businesses that get sustained value have a name against each system. Not a department, a person, with fifteen minutes a month. That is the entire governance requirement at this size and almost nobody does it.

What the successful ones have in common

Across the businesses actually producing something, the pattern is unglamorous and repeatable.

• One process, chosen because it runs every day. Not a programme. Not a strategy. One thing.

• The process was written down first. Every step, every exception, every point where work waits. Half the value arrives here, before anything is automated, because it exposes steps nobody can justify.

• A number agreed in advance. What we are measuring and what it is today. Without a baseline you cannot prove an improvement, which is how a great many projects end up described as producing no measurable return.

• Built on the systems they already run. No migration, no retraining, no parallel tool for people to forget about.

• A failure path built before the happy path. Retry, then escalate to a named person with the reason attached. Systems that fail silently get abandoned along with everything else you were planning.

What the ambitious ones did instead

They bought capability and waited for a use case to appear.

This is the most expensive version, because it produces genuine activity. People are experimenting, there are demonstrations, everybody can describe something interesting. What there is not is a process that runs differently to how it ran last year, and that is the only thing that shows up in the accounts.

The uncomfortable truth is that ambition is negatively correlated with results here. A business that automated one document chase and measured it is further ahead than one that ran a transformation programme, because the first one now knows how to do the second one.

How to tell which you are

Four questions, answerable this afternoon.

Name one process that runs differently than it did a year ago. If you cannot, adoption has not become change.

What number did it move, and what was that number before? If nobody established the before, nothing can be proven now.

Who owns it? If the answer is a department or a tool vendor, it has an expiry date.

What happens when it breaks? If nobody knows, it has broken already and you have not found out.

Where to start if the answers were uncomfortable

Pick the task somebody does every single day, that follows the same shape each time, and that nobody enjoys.

Write down how it actually works, not how it is supposed to. Agree what you will measure and record today’s figure. Build it into the systems you already use. Put a name against it. Then leave it alone for a month and look at the number.

That is a considerably less exciting plan than most of what is being sold, and it is roughly the difference between the businesses producing results and everybody else.

Sources

• Stanford Institute for Human-Centered Artificial Intelligence, Artificial Intelligence Index Report 2024, 15 April 2024.

AI Optimize scopes to a job somebody already does every day, writes the process down first, agrees the measure before the build, and puts a name against it afterwards. That work sits under Workflow Automation.