New engine, same factory

01The instinct to automate

When a process is slow and full of exceptions, the instinct is to put a platform on it. Today that platform is usually AI.

Automating a complex process makes every part of the complexity faster, and more expensive to run, govern and change. The exceptions still happen, now inside a system that bills for each one. The workarounds still exist, now coded in. The business gets a faster version of the process it already had.

02The factories that bolted on a motor

In the late nineteenth century, a factory ran off one large steam engine. Every machine was belted to a line shaft driven by that engine, so the machines were packed close together and stacked across several floors around the power source.

Electricity arrived, and many owners did the obvious thing. They replaced the steam engine with large electric motors and kept the shafts, belts and multi-storey buildings.

The economist Paul David traced what happened next. Electrification's payoff in manufacturing productivity came in the 1920s, about four decades after the first central power station opened. In the early 1920s, only slightly more than half of factory mechanical drive had been electrified.

The gains came when factories were redesigned around the new technology. Unit drive put a motor on each machine. Without overhead shafting, factories could be single storey and laid out by how materials flowed, instead of by distance from the power source. Productivity growth in manufacturing accelerated sharply through the 1920s.

Most organisations adopting AI today are where those owners were: a new engine in the same factory.

03The dip before the payoff

Economists Erik Brynjolfsson, Daniel Rock and Chad Syverson describe a productivity J-curve for general purpose technologies. Measured productivity dips first, because organisations are investing in new processes, skills and ways of working that don't show up as output yet. The payoff comes when those investments mature.

Inside a business, the dip has a familiar shape. People are doing two jobs at once: the work, and learning the new way of doing it. The board asks where the return is, often at the bottom of the curve. What turns the dip into the payoff is the redesign around the technology, more than the technology itself.

04What automating complexity costs today

• Cost. Usage-based bills grow with every exception and workaround the process still contains.
• Compliance. Every new tool is another place data goes, and another thing to govern and audit.
• Complexity. Another layer on top of processes nobody simplified, which makes the next change harder again.

A useful test: if your automation needs its own exception team, the mess was automated along with the work.

05The order that pays back

1. 1.Simplify. Take out the products, variants, routes and approvals the business carries without being paid for them.
2. 2.Redesign the work. Sort every step into Process (rules-based), Judgement (a person decides) and Control (a named person is accountable). Which approvals are real controls? shows the sort on a month-end close.
3. 3.Fix the data the decisions depend on. Neither a person nor a model makes a sound decision on data nobody trusts.
4. 4.Then automate the Process layer, support the Judgement layer, and concentrate Control where the risk is.

Our team has led this work for a leading global insurer: $213M in benefits delivered, and claims cycle times down 30 to 40%, with an end-to-end digital claims journey delivered globally.