Every business has a wild: systems built decades ago that still work, that nobody has opened since, and whose last expert retired in 2011. They weren’t abandoned because they don’t matter. Attention moved elsewhere and never came back. That is where the value of AI is hiding.
When businesses say they want value from AI, they mean the same few places: customer service, the contact centre, the app. These have had 20 years of investment, so what is left is refinement, and refinement brings incremental returns.
The places nobody has been are different. No modern tooling, no investment, some still running on fax machines. Work that was once too tedious to attempt can now be done quickly and reliably. There, AI isn’t an improvement. It is 20 years of evolution in a couple of months.
We admire those who leave the map, but then we stay safely inside the borders. Exploring looks romantic from the outside. From the inside, it means finding out expensively what is over the next ridge. We have made those trips with our clients, and these are our field notes.
1. Choose your horizon
“Deploy AI” is not a decision. Left alone, organisations pick the visible place, because that is where the complaints are, and complaints get funded. Ask instead: where do we hire people just to keep up? Where is someone retyping one document into another screen? Where is the rule written down but still applied by hand? And if the one person who runs this left tomorrow, would anyone know how it works? Don’t ask where AI would look impressive. Ask where nobody is looking.
2. Send a scout
Nobody has cracked the business case for AI, so buyers are asked to approve real spend against a benefit they have to imagine. Galileo faced the same problem in 1609. He didn’t lecture the Doge of Venice on optics. He pointed his telescope at the harbour and showed him ships approaching long before anyone on the ground could see them. He showed Venice its own world through the instrument.
We did the same with one overlooked process. A validation that took an hour now takes a minute. Across roughly 100 000 validations a year, that is close to 100 000 hours returned. Stop trying to prove the business case. Show it.
3. Pack for the terrain
The instinct is to point the biggest, most general model at a narrow, repetitive, auditable job. It hallucinates, drifts and costs a fortune, and when the regulator asks why a decision was made, nobody can answer.
Let AI do only what it is good at. The model reads the messy scanned document and finds the number; deterministic code decides whether the numbers match, so the answer is identical every time. Use small, purpose-built models: a frontier model doing document extraction is Einstein hired as a typist, bored enough to start improvising. And own your infrastructure. We learnt this when a model we depended on was retired mid-project, on someone else’s schedule. Be careful what you outsource.
4. Update your map
Most organisations change the work and leave the measurement alone, so people keep working to old targets and the gains leak away. Map every place you have put AI on one page. The empty space isn’t where nothing is happening. It is where nothing is being measured. Your key performance indicators (KPIs) are the waypoints. Move them.
5. Choose your party
Nobody good goes into the unknown alone. The owner of the business problem must drive the outcomes from the start, and the people who will use the system must help build it. Then you never have to sell it to them. Measure cycle time, not headcount replaced. Widen the job, not just the volume. And share the gains.
AI can rehumanise work
Our measure of success has never been how many people we could remove. What surprised me was that the people this work was supposed to replace asked for more of it. A back-office pile never ends, and clearing it faster makes for a better day at work.
Ubuntu is the operating principle: we can do more with people than instead of them. At one banking client, monthly workload fell from 223 hours, including 47 hours of overtime, to 136 hours with none, while productivity rose from 3.3 to 5 times and revenue held. The bank got the cycle time. The people got their heads back.
Go and find yours
Choose your horizon. Send a scout. Pack for the terrain. Update your map. Choose your party. Every business has a wild. Go and find yours.
- The author, Dr Andy Mabaso, is group CTO at Altron
- This promoted content was paid for by the party concerned
