Redesign before you automate

Module 4 left you with one bottleneck, someone to lead it and a written prediction. From day 15 to day 90 you improve the work around it, with the tools and people you already have. That's the Improve move, and it has one firm boundary: it spends no new money and no engineering time. Configuring a tool you've already approved counts. Buying a new one doesn't.

The first lesson is about order. What you do to the bottleneck matters less than what you do first.

Automating a broken step makes the mistakes faster

The obvious move is to point AI at the bottleneck as it stands. It feels decisive, and it usually disappoints, for a reason Michael Hammer set out in the Harvard Business Review in 1990, long before anyone had a chat window. His article was called “Reengineering Work: Don't Automate, Obliterate”.

His best-known case was Ford. In the early 1980s its North American accounts payable department employed more than 500 people. Management planned to cut that by about 20% by tidying up the process and installing new computer systems. Then it looked at Mazda, whose accounts payable department had five people.

The difference wasn't faster clerks. Ford's clerks spent most of their time on mismatches between the purchase order, the delivery record and the supplier's invoice. Hammer's point was that Ford could have helped them investigate mismatches faster, but the better choice was to stop the mismatches happening. Ford stopped asking suppliers for invoices. Goods were checked against the order when they arrived, three items were matched instead of fourteen, and payment followed automatically.

Ford measured the result the way companies did then, in headcount. Your aim measures it differently: freed hours go where your aim page says, and the promises you made in module 1 hold. The lesson for you is the order. Automating the old process would have made the mismatches cheaper to chase. Redesigning it made most of them disappear.

Where the time actually goes

Hammer had a second example that explains why speed at a step so often changes nothing. One insurer estimated that while an application spent 22 days in process, someone was actually working on it for just 17 minutes. The rest was waiting: in in-trays, between departments, for a signature.

AI is very good at the 17 minutes. It drafts, summarises and checks in seconds. But if the work is mostly waiting, rework and handoffs, faster minutes barely move the total. The current evidence says the same thing in survey form.

Four questions, in this order

Redesign sounds grand. In practice it's four questions, asked of every task the bottleneck does, in this order.

1. Does it need doing at all? Approvals nobody reads, data typed in twice, chasing information that should have arrived with the request. Often the fix sits upstream: stop the cause, as Ford stopped the invoices.
2. Could it be simpler? Combine two steps into one, standardise the form or the template, or move a check earlier so problems surface before they reach the bottleneck.
3. Does it need the bottleneck? Formatting, status updates and chasing rarely need your scarcest people. Move them to a team with spare capacity, and agree it with that team first. Moving work is only an improvement if it lands somewhere with room for it.
4. Then: would an approved tool help with what's left? Now AI earns its place, on work that genuinely belongs at the bottleneck, with the person there checking what it produces.

The order matters because each question shrinks the work the next one acts on. Ask the fourth question first and you'll automate tasks the first question would have removed.

Try it below. List what the bottleneck spends a typical week on, then mark each task. If you named your bottleneck in module 4, it's shown at the top. If not, start from one of the examples.

Who does the redesign

The people who do the work redesign it, in their protected hours, with the bottleneck lead. They know where the time goes, and they'll only tell you if it's safe. That's the second promise from module 1 doing its job: no one is punished for improving their own work. When people believe it, the waste in their week becomes visible. When they don't, it stays hidden, and so does your improvement.