AI has made the work faster. So why is the business still slow?

You're reading The AI Directive — practical AI intelligence for
business leaders. Subscribe
here if someone forwarded this.

Imagine a team that uses AI to transform its monthly performance
reporting.

What previously took three days now takes three hours.

The data is brought together. The first commentary is drafted.
Variances are highlighted. The pack is accurate, timely and cheaper to
produce.

On its own, that is useful—but strategically modest.

The reporting process was never valuable because it produced a pack.
Its value was supposed to be better decisions and faster
intervention.

Yet the business still manages performance in the same way. It
reviews the same lagging measures, through the same functional lenses,
with the same accountabilities and the same tolerance for problems
travelling through management layers before anyone acts.

AI has accelerated the production of information without changing the
management system that consumes it.

The task became dramatically faster.

The business did not.

That distinction is becoming one of the most important questions in
enterprise AI.

Last week, I asked whether your AI had actually added value. This is
the uncomfortable follow-on.

Even when an AI productivity gain is real, where does it go?

Because a business does not consume saved minutes. It benefits when
something meaningful changes: elapsed time, throughput, quality,
service, cost, working capital, revenue or risk.

If an employee produces a report five hours faster but the
organisation still interprets, challenges and acts on it in exactly the
same way, it has captured only a fraction of the gain.

The work is faster.

The flow is not.

It is worth being clear about this first.

AI genuinely can make bounded tasks faster.

It can prepare a first draft, compare documents, analyse a dataset,
summarise a meeting, find anomalies, structure a proposal or produce a
set of options in a fraction of the time previously required.

People using these tools often feel the difference directly. The
blank page disappears. The first pass arrives sooner. Mechanical work
reduces. More time becomes available for judgement and refinement.

Current evidence supports that experience.

Gallup reports that 65 percent of employees using AI say it improves
their productivity or efficiency. Yet only 12 percent strongly agree
that AI has transformed how work gets done in their organisation.

Atlassian found an even sharper contrast. Eighty-nine percent of
executives said AI had increased the speed of work, but only 6 percent
were confident they could point to specific organisation-wide ROI.

That is not evidence that AI does nothing.

It is evidence that individual productivity and enterprise
productivity are different things.

And the gap between them is where many AI benefits are now
disappearing.

For years, productivity in knowledge work has often been discussed
through proxies.

Documents produced.

Emails answered.

Meetings attended.

Reports completed.

Hours saved.

AI makes many of those activities faster. It can also make far more
of them possible.

That creates a danger.

If the organisation rewards visible activity, AI may simply create
more activity: more analysis, more slides, more options, more
communications and more material for somebody else to review.

The employee feels productive because they have produced more.

The recipient feels overloaded because they have received more.

The customer may notice no difference at all.

The right measure depends on the outcome.

Did the reporting cycle move from days to hours?

Did the customer receive a decision sooner?

Did the product reach market earlier?

Did the team process more work without adding cost?

Did quality improve without increasing review effort?

Did working capital, conversion, service or risk actually move?

If not, the saved time may be genuine. It just has not yet become a
business benefit.

This is easier to see in manufacturing.

If one machine doubles its speed, the factory does not automatically
double its output.

The next operation may lack capacity. Materials may not arrive
reliably. Quality problems may create rework. Operators may not have the
skills or authority to resolve exceptions. The planning system may
release work in the wrong sequence. Demand may not justify the
additional output.

At best, the faster machine creates no extra value.

At worst, it creates more work in progress, hides problems under
inventory and makes the whole system harder to manage.

No credible manufacturing leader would assess that investment by
looking only at the cycle time of one machine.

They would look at the system around it.

Yet that is exactly how many organisations are assessing AI. One task
becomes faster and the local improvement is presented as if it were the
enterprise result.

The same systems thinking should apply. The wider benefit depends on
people, process, data and technology working
together.

AI changes more than the time required to complete a task.

It can change who prepares the work, who checks it, who decides,
which skills matter and where accountability sits.

If those questions are left unresolved, the organisation often adds
AI work on top of the old work.

Employees produce the AI-assisted output, then repeat the manual
process for reassurance. Managers review material at a level of detail
that no longer makes sense. People save time but are given no clarity on
how to use it. Teams do not trust the result because nobody has agreed
what good looks like.

The people question is not simply whether users have been trained on
the tool.

It is whether roles, capability, incentives, confidence, capacity and
decision rights have changed with the work.

An employee saving five hours a week is not yet an organisational
benefit. The business has to decide whether that capacity will improve
service, increase throughput, reduce cost, strengthen quality or create
room for higher-value work.

If nobody makes that choice, diaries merely refill.

AI is frequently inserted into an existing activity while the wider
process remains untouched.

The work still crosses the same functions. It still requires the same
forms. It still follows rules created around older systems. It still
returns for clarification and rework. Decisions still sit at the same
organisational level, governed by the same assumptions about who may act
and when.

The technology has changed. The journey has not.

That is why a use case can look excellent in a demonstration and
disappointing in the operating results. The demonstration measures the
AI-enabled step. The business experiences the whole process from demand
to outcome.

Sometimes the correct answer is not to accelerate the step.

It is to remove it, combine it with another step, move the decision
closer to the work or stop producing an output that exists only because
the old process required it.

This is process redesign, not prompt improvement.

AI does not remove the data problem.

It can make the consequences arrive faster.

If product, customer, financial or operational data is incomplete,
delayed or defined differently across functions, AI can produce a
polished answer built on conflicting facts.

That creates checking and rework. It also weakens trust, which pushes
people back towards manual verification and parallel spreadsheets.

The important data questions are not only technical:

A faster analysis of poor or disputed data is not transformation. It
is a quicker route to the same argument.

The AI tool itself still matters.

It must be reliable, secure, affordable and appropriate to the task.
But enterprise value also depends on how it fits with the rest of the
technology landscape.

If employees have to copy information into a separate tool,
reconstruct context, download the output, reformat it and upload it
elsewhere, the business has automated a fragment and preserved the
friction.

The same applies when several teams buy overlapping AI products, each
with different data, controls and ways of working. Local enthusiasm can
create enterprise fragmentation.

The technology question is therefore broader than model quality.

Does the capability sit inside the real workflow? Does it connect to
trusted data? Does it reduce system movement and manual handling? Can it
be supported and measured at a sensible cost?

If not, the AI may remain an impressive assistant at the edge of the
process rather than a meaningful change to how the business
operates.

People, process, data and technology are often presented as four
boxes on a transformation slide.

They are not four independent workstreams to complete in
parallel.

They are one operating system.

Change the technology and you alter the process. Change the process
and roles or decisions may move. Change the data and some controls
become unnecessary while others become more important. Change the
measures and behaviour follows.

The value appears—or disappears—in those connections.

After 30 years around technology and business transformation, this
pattern is very familiar.

A new system performs broadly as promised.

The implementation team can demonstrate that a task is faster, a
screen is better or information is easier to access.

But the expected business benefit arrives slowly, partially or not at
all because the surrounding people, process, data and technology remain
out of alignment.

The technology gets blamed.

Sometimes that is deserved.

Often it is not.

The real mistake was treating a new capability as if it were the
transformation itself.

AI makes that mistake easier because the local gains can be so
immediate and impressive. A task that took hours can genuinely take
minutes. That creates a compelling before-and-after story.

But the organisation does not operate as a collection of independent
before-and-after stories.

It operates as a system.

And systems are governed by how their component parts work together,
not by the performance of one new tool.

There is a more constructive way to read this.

When AI makes one activity dramatically faster, it exposes what the
wider system can no longer absorb.

Perhaps preparation was never the real constraint.

Perhaps the process contains controls designed for a risk that no
longer exists.

Perhaps people are still measured for producing an output that AI can
now create almost instantly.

Perhaps the data definitions are inconsistent across functions.

Perhaps the tool sits outside the real workflow and creates another
manual handoff.

Perhaps every exception still travels to a manager because decision
rights were never revisited.

That visibility is valuable.

The wrong response is to declare the AI pilot successful because its
task completed faster.

The equally wrong response is to declare it a failure because the
enterprise metric has not yet moved.

The useful response is to review the whole system.

What outcome are we trying to improve?

How should the work now operate?

What must change for people, process, data and technology to support
that outcome together?

For one important process, I would run a very simple exercise.

Choose an outcome that matters: resolving a customer complaint,
launching a product, approving an investment, recruiting someone,
closing the month or responding to a sales opportunity.

Draw how it currently works from demand to delivery.

Then review it through five connected lenses:

Then overlay the AI use case.

Which part became faster?

What else has to change around it?

Has a role become more valuable, less valuable or simply
different?

Can a step or report disappear entirely?

Does the data support the new way of working?

Is the capability embedded in the workflow or bolted onto its
edge?

Did the customer, employee or P&L experience the gain?

That is a transformation review. The AI task is only one part of
it.

Pick one process where AI has made work faster.

Review the whole outcome through people, process, data and
technology.

If only the tool has changed, you have not completed the
transformation.

You have installed the first component of it.

AI can accelerate one component of the business.

The benefit arrives only when people, process, data and technology
change around it.

And once they do, the next constraint often moves towards the people
who review, approve and decide.

That is where this arc goes next.

See you Tuesday.

— Toby

@TobySmithFirth on
YouTube

Subscribe to The AI Directive here: https://theaidirective.co.uk/subscribe?utm_source=beehiiv&utm_medium=email&utm_campaign=issue_018

These are tools I use personally. Affiliate links marked - I earn a small commission if you sign up, at no extra cost to you.

Some links in this issue are affiliate links. I only recommend tools I actually use.