Every handoff has a cost

Most organisations measure the work people do. They are much worse at measuring the work created by moving work between people, systems and stages.

A customer sends information by email. Someone copies it into a system. Another person checks it. Something is missing, so they go back to the customer. A spreadsheet gets updated. A colleague asks where the latest version is. Someone reconstructs the context. A manager reviews it. The information moves again.

None of these actions looks particularly expensive. That is the problem.

The cost appears in fragments: five minutes finding something, ten minutes checking whether it is current, another email asking for missing information, a meeting to restore context, a manual reconciliation because two systems disagree.

Individually, they look like admin. Together, they can determine the capacity of an organisation. I heard a useful phrase for this at Accountex North: admin interest.

The idea is simple. Every unnecessary handoff creates a small cost. Leave enough of those costs inside a workflow and they compound. It is a useful way to think about knowledge work because much of the friction inside modern organisations does not come from the difficult work itself; it comes from everything required to make the difficult work possible.

The work around the work

We usually describe a process through its meaningful stages: research, draft, review, approve, publish. Or capture, review, submit, advise. The diagram looks clean because the diagram describes what is supposed to happen.

The lived workflow contains something else:

Find → copy → explain → chase → check → reconcile → reformat → resend → remember → ask again.

This is the work around the work, and it is often largely invisible.

That invisibility matters. If a task takes an hour, the organisation can see the hour. If the same task requires six people to repeatedly spend five or ten minutes restoring context around it, those costs disappear into everybody’s day. Nobody owns them, so nobody removes them. The organisation simply feels busy.

A handoff transfers more than a task

The usual assumption is that a handoff transfers work. In knowledge work, it also has to transfer context. Imagine a useful piece of information moving through an organisation:

Customer → interview → transcript → writer → document → reviewer → CMS → campaign → search → future content

At every boundary, something has to survive:

• what the information means
• where it came from
• whether it is trustworthy
• what has already been decided
• what still needs checking
• why it matters
• where it could be reused

If the system cannot carry that context, a person has to recreate it. That is where the interest starts accumulating.

A reviewer asks a question already answered in the interview. A writer searches for a source someone else already found. A campaign team commissions research that exists elsewhere. Six months later, somebody interviews another customer and asks exactly the same weak question because nobody retained what was learned the first time.

The information exists. The organisational memory does not.

Why AI can make this better — or worse

This is one reason I am increasingly unconvinced by AI strategies built around isolated tasks: generate the email faster, summarise the meeting faster, draft the article faster, produce the presentation faster. Useful, certainly. But making one stage faster does not necessarily make the system faster.

If the generated output still needs to be copied somewhere else, explained to another person, checked against another source, reformatted for another system and rediscovered three months later, the organisation has accelerated a task without removing much administrative interest.

There is an even stranger possibility: AI makes generating work so cheap that organisations create more things to hand off — more drafts, more summaries, more documents, more analysis, more messages.

The production cost falls while the coordination cost rises.

The better opportunity is therefore not simply AI that does work. It is AI embedded deeply enough in the workflow to remove the need for some handoffs altogether.

Capture information at source. Preserve its provenance. Categorise it automatically. Surface exceptions instead of making humans inspect everything. Carry decisions forward. Make previous evidence retrievable when the next piece of work begins.

The objective is not to remove professional judgement. It is to stop wasting professional judgement on reconstructing the administrative context required to exercise it.

This changes how I think about editorial systems

The same problem exists inside content organisations. We often focus on the visible output — the article, video, campaign, report or social post — but a large amount of editorial capacity disappears before anybody writes anything.

Someone has already interviewed the expert, but the transcript is inaccessible. Research exists, but nobody knows where it lives. A useful customer quote sits inside an old PDF. A campaign generated an interesting narrative, but the product team that could use it never encountered it. A product team is solving a problem that perfectly completes an editorial argument, but the content team does not know the product exists.

That last example became particularly obvious to me at Accountex. I had been working around a narrative about the hidden hours lost to invisible administrative work. At the event I encountered a product story about embedded accounting that attacked almost exactly that problem from another direction: cleaner data at source, fewer manual interventions, accounting by exception and AI inside the workflow.

The connection was stronger than either story by itself, but it did not arrive through a content brief. It appeared because the pieces of organisational knowledge happened to meet. That is an editorial intelligence problem.

The joins are where the value is

This is why I increasingly think the important unit of editorial intelligence is not the content asset. It is the join: research joined to a customer observation, a customer observation joined to a product capability, a product capability joined to a changing search question, an old interview joined to a new narrative, a field observation joined to an existing evidence bank.

The more effectively those joins happen, the less knowledge has to be rediscovered or recreated, and the organisation gets more value from work it has already paid for. This is also why editorial memory matters.

A good memory layer does not merely archive finished content. It reduces the cost of the next piece of work. The next writer starts with the evidence already found; the next interview starts with the questions we wish we had asked last time; the next campaign knows which claims already have proof; the next AI system retrieves decisions rather than inventing them again. In other words, memory reduces administrative interest.

Measure the distance between useful work

There is a temptation with AI to ask how much faster an individual can perform a task. I think there is a better question:

How much unnecessary work sits between the useful things people are trying to do?

Look at how many times information changes systems, how often context has to be explained again, how much work is spent chasing inputs rather than using them, and how frequently decisions are rediscovered. Look at how many people inspect routine work because the system cannot reliably identify the exceptions, and how much existing knowledge gets recreated because it cannot be found.

Those are not glamorous AI questions, but they may be where a large part of the economic value sits.

The goal should not be to make every person generate more. It should be to make the organisation lose less between one useful piece of work and the next.

Every handoff has a cost. The best systems make fewer of them necessary.