Anyone can buy AI. Few companies redesign the work.

You might expect the first returns from AI to come from something advanced. At Hindustan Power, they came from coal logistics: checking seals, processing documents, tracking movements and recording tests. The company says this saves more than 1,000 hours of manual work and close to 50,000 pages of paperwork every month.

That is not glamorous work. It is the work that piles up.

In a Gulf business, the same pile looks different. Delivery notes photographed and sent on WhatsApp. Purchase orders typed again into a spreadsheet. A branch manager sending numbers every evening so someone can build a report in the morning. Each one is a person doing the computer's job.

That is where we start too. AI that reads an incoming order and opens the case is less exciting than a chatbot. It also saves hours from the first week.

Puri describes a habit worth copying. When his teams design a new solar or storage project, they decide what data it will produce, and who will act on it, at the same time as the engineering design. Not after the plant is running.

Most businesses do the opposite. They automate a process, then try to pull reports out of it later, and find the information they need was never captured.

When we map a workflow, every step gets a named owner and a clear record: who did it, when, and what they decided. Reports then come out of the work itself. Nobody has to compile them.

Puri has little patience for expensive systems that produce detailed dashboards and change nothing on the ground. In his words, the value is in systems that “recommend the next action and, over time, take it.”

A dashboard tells you a delivery is late. A working system notices the delivery is at risk, tells the right person, and suggests what to do, before the client calls.

This is why we design the action first and the screen second. The AI does the first pass. A named person approves. The record shows both.

His advice to other technology leaders is plain: start with an operating problem and a measurable target, and give every pilot a fixed window to prove itself before it is scaled up or closed.

A pilot without an end date is just a slower way of failing. It keeps a team busy, keeps a vendor paid, and never answers the only question that matters: did it work?

Our Quick Wins follow the same rule. One narrow task. One number agreed before we start, such as hours saved or days of turnaround. A short window to show it on real data. Then you decide.

Hindustan Power is placing AI engineers inside every function, from finance and procurement to the plant floor and the mines, working alongside the teams that run them. The reasoning: a model only creates value once it is built into how the work is actually done.

You cannot learn how a team works from a requirements document. You learn it by sitting with the people who do the job, watching where the work waits, and seeing which WhatsApp group the real decisions happen in.

That is why we map the work on site, and why we stay after go-live. The first version of any workflow is a good guess. The tenth version is the one that fits.

You do not need a power company's budget to apply any of this. In a business of 20 or 30 people, the same approach is smaller and faster:

It works at this scale too. At one Dubai dental lab with 27 technicians, median turnaround fell 21%, from 8.0 to 6.3 days, after every case moved onto one workflow. At Al Jabi Motors, four branches now report on the same day instead of twice a week, saving about 20 hours of reporting a week.

None of that came from a better AI model. It came from redesigning the work, then letting the system do the first pass.