AI engineering · Logistics and supply chain · India
Twenty five operations leaders, and two things each of them could run on Monday.
Om Logistics runs first mile, middle mile and last mile operations across India. Their operations leaders had decades of process knowledge between them and a browser tab open on ChatGPT, used for the occasional email. We ran two days with them, twice, and they left having built working outputs from their own operations data.
At a glance
- Status
- Delivered, in two batches
- What we ran
- A two day AI literacy programme, twice
- Who attended
- Senior operations leadership and team leads
- Tools used
- ChatGPT, Claude and Claude Cowork
What changed
What changed
Before
ChatGPT open in a browser tab, used for the odd email draft.
Now
Two reusable skills built from their own operations data, run the next morning.
And a clear line between what AI prepares and what a person decides.
- Operations leaders, in two batches
- 25
- Functions represented, first mile through MIS
- 8
- Working outputs built by each participant
- 2
- Days between batches, so the first could report back
- 15
The ask
Why they came to us
Om Logistics moves freight across India through first mile, middle mile and last mile operations. Behind them sit fleet management, vehicle hiring, key account management and MIS. The people who run those functions have been doing it for a long time. They know where the work stalls better than any consultant will.
AI had arrived in the building the way it arrives everywhere: through a browser tab someone opened themselves. It was being used for the odd email draft. Nobody had shown this group what a model is actually good at, where it breaks, or what they should not paste into it.
What we built
How the two days ran
Their operations, not generic examples
Every illustration came from Om Logistics: the morning MIS dashboard, the Saturday review, the escalation emails, the SOP rewrites. What a model does well and where it fails were both shown against work they had done that week.
Played, not lectured
Myth or fact card decks on LLMs and on agents, sorted in teams before the walkthrough and scored after it. A control tower assembly exercise mapped the parts of an agent onto a structure they already understood.
Built on day two
A pickup exception escalation from a failed pickup report, walked through together. Then a client review deck from six months of performance data, on their own. Both saved as skills they could run again.
The method
How we worked
01
Sanitised their data first
We prepared the exercise data with their MIS head, so people worked on real operations records without real client names leaving the room.
02
Managed accounts, not personal ones
Tool access was set up on team plans rather than personal subscriptions, so usage stayed visible and retention was set by Om Logistics.
03
Twelve or thirteen at a time
Small enough that every participant was seen and helped. Twenty five people meant two batches rather than one large room.
04
Fifteen days between batches
The first group went back to work and tried it. What they found came into the second batch as real reflection rather than theory.
What we left out
Where we stopped, deliberately
The programme ends by naming what the tools in the room cannot do. No connection to live data, so everything is copied and pasted. No memory between sessions. No standard output, because every person writes their prompt slightly differently. No workflow automation, and no audit trail of what the AI was asked or answered.
That is not a sales turn at the end of a workshop. It is the honest boundary of a chat tool. It decides which jobs are worth building properly, and which should stay exactly where they are.
Beyond this client
The job transfers
The same problem turns up well beyond this one client. These are the teams it fits.
Anyone whose experienced operators are the ones being asked to adopt AI
- Logistics and supply chain
- Manufacturing operations
- Field services
- Any function run by long tenured staff
Next case study
Vehicle logistics group · AI engineering
Five document jobs, one way of working
Facing something similar?