Stop making bad work faster

Thirty-six years after Michael Hammer told companies to obliterate obsolete processes, we're about to automate them at a scale he couldn't have imagined.

In 1990, Michael Hammer wrote one of the great business article titles: Reengineering Work: Don't Automate, Obliterate. His argument was that companies were spending heavily on information technology to make old processes faster when they should have been questioning the processes themselves. He wrote that businesses were mechanising old ways of working rather than using new technology to redesign the work.

Thirty-six years later, we have much better technology. We appear to have preserved the mistake.

The meeting is easy enough to picture. Someone puts the current process on a screen. Twelve boxes. Three systems. Four approvals. Two spreadsheets. An email that, somewhere along the way, appears to have become infrastructure. The conversation turns to which parts can be automated.

I'd start by asking why any of it exists.

Nobody designed this mess

Most bad corporate processes weren't designed. They accumulated.

A control appeared after something went wrong. A spreadsheet filled the gap between two systems. An approval gave a manager visibility. A report was created for an executive committee. The manager moved on, the systems changed and the committee disappeared. The process remained, quietly achieving tenure.

Eventually someone is downloading information from one system, correcting it in Excel, uploading it into another, emailing somebody to tell them it is there, then updating a third system to record that the email was sent. Nobody would design this from scratch. The trouble is that nobody is designing it from scratch.

The new generation of automation tools can remove enormous amounts of manual work from arrangements like this. Good. A process can become faster without becoming necessary.

The numbers suggest we're starting at the wrong end

Australian directors have reason to care about this. The AICD's first-half 2026 Director Sentiment Index surveyed 828 company directors. Forty-three per cent nominated productivity as the number one short-term issue for the Federal Government, while 68 per cent said regulatory and compliance requirements were limiting productivity growth in their businesses. Almost two-thirds reported that AI tools had already delivered productivity benefits, and more than four in five expected implementation to increase over the following year.

The tools are improving productivity, investment is increasing, and productivity remains the problem.

Research into how companies are implementing the technology offers one explanation. McKinsey tested 25 organisational attributes against the EBIT impact companies reported from their use of generative systems. Of those 25, redesigning workflows had the largest effect. Only 21 per cent of respondents whose organisations used those systems said they had fundamentally redesigned at least some workflows.

Deloitte found much the same from another direction. In a survey of C-suite leaders at large companies, 59 per cent reported a technology-focused approach to investment, while 16 per cent said they had fully designed roles, processes and operating models to integrate the technology into work. The technology-focused group was 1.6 times more likely to report investments that were not exceeding expectations.

We have bought a new engine and left it attached to the old gearbox.

Six days saved. Zero dollars earned

Suppose a monthly management report takes two people three days to produce. New tooling reduces that to twenty minutes. The business case practically writes itself: nearly six person-days saved every month.

I'd want to know who reads the report, what decision it changes, and what happened the last time somebody acted differently because of it. Then I'd ask what would happen if we stopped producing it for three months.

If the answer is "nothing", we didn't save six days. We discovered that we had been wasting six days.

The same test works elsewhere. If almost every request passing through an approval process is approved, the opportunity is not a faster approval pack. It is the approval. If people spend their days moving information between two systems, teaching software to copy it faster may be less useful than fixing the systems.

And if five people are required to produce a number that already exists in an authoritative system, I'd be curious how the number acquired four chaperones.

Start with the reason

Before investing in the automation of an existing process, management should be able to answer a few ordinary questions.

What outcome does this process produce? Not what does it do. What changes in the real world because it exists? A customer receives something sooner. Money arrives. A risk is prevented. A regulatory obligation is satisfied. Someone can make a decision they couldn't make before.

Then trace the process backwards. Which steps are genuinely required? Find the actual regulation, contract, control or customer need behind them. "We've always done it this way" has enjoyed a long and successful career as a business requirement.

What happens if we remove a step? Do it somewhere controlled and observe what breaks. Some steps will prove their value quickly. Others may reveal that their principal function was supplying information to the next unnecessary step.

Then find the constraint. Where does the work actually wait? Making one activity ten times faster achieves little if the work then spends three days sitting in somebody's approval queue. Local efficiency can move the traffic jam twenty metres down the road.

Finally, follow the money. If the proposed improvement works, what changes? Cost should fall, throughput should rise, revenue should arrive earlier, errors should decline, exposure should reduce or future headcount should be avoided.

"People will have more time for higher-value work" is not a finished sentence. What work?

Faster people can produce a slower company

Producing work is getting cheap. More reports can be written. More analysis produced. More presentations assembled. More correspondence sent. More code generated. Each individual can look more productive while the organisation becomes busier consuming everybody else's productivity.

Microsoft's analysis of aggregated and anonymised Microsoft 365 activity gives some sense of the existing coordination load. Among the 20 per cent of users receiving the most digital interruptions, a meeting, email or chat arrived roughly every two minutes during an eight-hour workday. Among the 20 per cent with the highest meeting volumes, 60 per cent of meetings were unscheduled or ad hoc. These are deliberately selected high-volume cohorts, not representative figures for every office worker, but they show what the busy end of modern knowledge work already looks like.

McKinsey's most recent global survey puts the same gap in one place. Eighty per cent of respondents said AI had improved their own productivity. The share of organisations attributing any EBIT impact to AI use was 37 per cent, essentially unchanged from a year earlier. Individual output rose. The company's numbers did not move.

Now make producing things easier. Finance can generate more analysis for management to review. Marketing can produce more variants for someone to approve. Developers can create more code for someone to test. Management can produce more papers for someone to read.

Everyone gets faster. The queue moves downstairs.

That is why I wouldn't measure one person's saved hours and call the result productivity. Follow the work from beginning to end. Did the customer get something sooner? Did fewer people touch it? Did money move? Did risk fall? Did something disappear?

If nothing changed outside the department that bought the tool, I'd keep looking.

Delete is a legitimate transformation strategy

Once the work is understood, the technology decision becomes much less interesting.

Some work should disappear because its reason has disappeared. Some should be simplified because the outcome still matters but fifteen years of organisational sediment doesn't. Some is sound, repetitive work that machines should handle. Some should remain substantially human because judgment, accountability, trust or consequence makes that the sensible place for it.

The sequence matters more than the labels: delete what shouldn't exist, simplify what survives, automate what remains.

Otherwise we may spend the next five years teaching capable machines to perform corporate rituals nobody got around to cancelling.

One question for the next investment paper

A board doesn't need to redesign an operating process. Management does. But before approving another business case promising thousands of hours saved, I'd put one question on the table:

If we were designing this company today, would we create this process?

If the answer is yes, improve it. If the answer is partly, cut it back. If nobody knows, finding out is worth more than the proposal currently asking for funding.

Hammer's argument in 1990 was not anti-technology. New technology was too important to waste on preserving old processes. We now have machinery that can do far more of that work, far more quickly, and much more of it at once.

The first question isn't what it can do. It's what we can finally stop doing.

Sources

Michael Hammer, "Reengineering Work: Don't Automate, Obliterate", Harvard Business Review, July-August 1990 (vol. 68, no. 4; HBR reprint 90406). Hammer argued that companies were using information technology to speed up established processes rather than fundamentally redesigning work, and advocated organising around outcomes rather than inherited tasks.

Australian Institute of Company Directors, Director Sentiment Index 1H 2026, published 16 April 2026. Survey of 828 Australian company directors conducted by Roy Morgan between 20 February and 10 March 2026. Supports the figures on productivity (43 per cent), regulatory burden (68 per cent), AI tools already delivering productivity benefits (almost two-thirds) and expected increase in AI implementation (more than four in five).

McKinsey & Company, "The state of AI: How organizations are rewiring to capture value", 12 March 2025. Of 25 organisational attributes tested, the redesign of workflows had the biggest effect on reported EBIT impact from gen AI use; 21 per cent of respondents reporting gen AI use said their organisations had fundamentally redesigned at least some workflows. Fieldwork ran 16 to 31 July 2024, with 1,491 respondents in 101 nations.

Deloitte, "Work Design Essential to Realize AI Return on Investment", 27 October 2025. Survey of C-suite leaders at organisations with more than 5,000 employees. Reports 59 per cent taking a technology-focused approach to AI investment, 16 per cent having fully designed roles, processes and operating models to integrate AI into work, and the technology-focused group being 1.6 times more likely to report investments not exceeding expectations.

Microsoft WorkLab, "Breaking down the infinite workday", 17 June 2025. Based on aggregated and anonymised Microsoft 365 productivity signals ending 15 February 2025, excluding education and EU tenants. The two-minute interruption measure applies to the top 20 per cent of users by ping volume; the 60 per cent ad hoc meeting figure applies to the top 20 per cent by meeting volume.

McKinsey & Company, "The state of AI in 2026: On the road to ROI", 25 August 2026. Online survey in the field from 4 May to 8 June 2026, 1,719 respondents in 97 nations. Eighty per cent of respondents report AI has improved their individual productivity; 37 per cent attribute at least some EBIT impact to AI use, essentially unchanged year on year. Nearly three-quarters of AI high performers report fundamentally redesigning workflows, against one-quarter of other respondents.