4 NOVEMBER 2026 · 17:00 GST · 5 MIN READ · DUBAI CAREER NOTES

Start with the boring AI: how operators should use AI on the work they already hate

By Prudhvi Polavarapu, supply chain and e-commerce operations leader in Dubai

Most operators I meet want AI to do something dramatic, like run the warehouse by itself. The better first move is smaller: point it at the work your team already hates, the reading, sorting and summarising that eats the first two hours of every shift.

The 7am scene

Picture a last-mile control room at 7am. Overnight, 180 deliveries failed. Each one has a driver note: "customer not answering", "gate closed", "wrong floor", "asked to come tomorrow". A supervisor reads them one by one, decides what each means, and re-queues the order. It takes two hours, and the people who most need the answer, the customers, are still waiting. I have stood in that room many times, including during peaks above 5,000 orders a day. The reading was never the skill. The deciding was.

That split is exactly where AI is genuinely useful today. A language model is good at reading messy text and sorting it into a few clear buckets. It is not good at knowing your customer, your driver or your cut-off. So let it read, and keep humans for deciding.

The Tuesday Test

I use one rule to choose a first AI use case, and I call it the Tuesday Test. A task passes if it is Text-heavy, Repeated every working day, and Checkable in under a minute by a person who knows the work. Text, repeat, check. If a task happens on a random Tuesday, involves lots of words, and a supervisor can verify the output quickly, it is a good candidate. If a mistake would be invisible or expensive before anyone could catch it, it fails the test, however impressive the demo looks.

Let AI do the reading. Keep the deciding.

Five jobs that pass

Failed delivery triage. Feed the driver notes in, get back groups such as "reattempt today", "needs customer call", "address problem", "return to warehouse". Your supervisor reviews the groups, not 180 lines.

Shift handover summaries. Paste the day's log and ask for the five things the next shift must know. Handover quality is one of the quietest causes of repeat mistakes in any operation.

Supplier and carrier emails. Draft a polite, clear chaser with the PO number, the dates and the one question that needs an answer. You still send it, and you still own it.

Customer reply drafts. Common questions such as "where is my order" get a first draft in your brand tone, which an agent edits in seconds instead of typing from zero.

Product listing clean-up. Standardise titles, attributes and size formats across a catalogue, then let a person spot-check a sample before anything goes live.

What to keep away from it, at first

Do not start with anything that moves money or stock on its own. Refund decisions, inventory adjustments and customs declarations need a named person accountable, and the official rules in those areas should always come from the authorities, such as Dubai Customs, not from a chatbot's memory. Also be careful with data: do not paste customer phone numbers, addresses or confidential supplier prices into a public tool. Check your company's policy first, and use approved tools where your employer provides them.

The UAE is serious about this direction. The UAE National Strategy for Artificial Intelligence 2031, published on the official u.ae portal, names transport and logistics among its priority sectors, and local port operators have spoken publicly about using AI to support their people. For an operator in this country, learning these basics is not a hobby. It is part of staying useful in a market that is moving this way.

How I would run the pilot

Pick one task. Run it in parallel for two weeks: the person does it as usual, the AI does it as well, and you compare. Count how often the AI's grouping matched the human's. When the match is high and the misses are harmless, let the person start from the AI draft and review. Measure minutes saved per shift, and write the number down. In my own operations, trust in any new tool, from WMS to route software, was always earned by the same thing: a visible number, checked by someone who knew the work.

You do not need to become an engineer to use AI well. You need to stay the person who knows what good looks like. Start boring, check everything, and let the minutes you win go back to the work only you can do.