What an operating model is, and where AI changes it

An operating model is how a business runs: who does the work, by which steps, on which systems, who decides, and how you know it works. MIT CISR describes it as the level of integration and standardisation of business processes needed to serve customers (Ross, December 2005).

I split it into five parts you can change one at a time. That frame is my method, not a standard. A target operating model is the same five parts written as you want them to be. Operating model design is the work of closing the gap between the two.

The CISR briefing asks two questions. How standard should the process be across the business? How joined up should the data be? The two answers give four models. The briefing’s company examples are large and from 2005, so use the questions, not the examples.

Pick a model on purpose. A business with several different service lines needs less standardisation. A business that repeats one job at every site needs more. Then check the target against the five parts.

Take one process that crosses people and systems. Order to cash is a good first choice. List each step with who does it, the system, what goes in and what comes out. Then add the hours a month and what an error costs.

Mark each step Rule, AI or Person, and write why. The template holds this sample, a blank map for your process and a today-and-target sheet.

The sample is a fictional business. Its hours and costs are invented to show the method, so replace them with your own records.

AI suits steps that read or draft free text: an order in an email, a reminder to a customer. A person checks the result before it is used. A fixed rule needs no AI. It needs a test. A decision that carries money or risk stays with a person who is accountable.

Small businesses seem to be behind. The Productivity Commission found that individuals and large firms are adopting AI heavily while small businesses ‘seem to be moving much slower’.

Hours and dollars sit in separate columns. Add them and you hide the answer. In the sample, the Rule steps hold most of the hours, and the Person steps hold most of the error cost.

Automating rules frees time. A person’s decision limits the damage. Time freed is not cash. It becomes cash only when the hours do paid work or avoid a hire.

Put the running cost beside every saving. In the sample, cash alone leaves A$140 a month before the one-off build. The case rests on the 48.5 hours, and on using them.

Write each part as it runs today and as you want it. Keep it to one line a part. The gap is the plan.

Technology is one part of five. A tool can fix the record, but it cannot name who decides or who reviews. The step that fixes the record is naming one owner for each record.

Change the cheap parts first. Decision rights and the process cost time, not software. Technology follows the map, so software takes only the steps you have shown to be rules. Governance starts in month one, because it tells you whether the rest worked.

Two Australian rules shape the technology step. ASIC says companies must keep financial records for at least 7 years, so keep a retired system’s history that long. Where the Privacy Act applies, APP 10 expects reasonable steps to keep personal information accurate, up-to-date and complete. Get advice where your case is unclear.

A board gets value from a changed process only when it sees the result. I rebuilt the month-end close at a manufacturer and at a retailer. It went from 21 days to 5 at the first and from 10 days to 5 at the second. In both, the board used the new reporting for capital allocation.

How long a change takes depends on the business, the systems and the team, so I give no fixed duration.

The first three steps need a pencil, not software. Start with one process.