Sales, operations and M&R all travel to depots, suppliers and customers, and almost none of what they learned reached the people who could act on it. I designed a closed loop that turned scattered trip notes into one standard report and a searchable memory, and it was approved by the C-suite for global rollout.
AI agentsProcess designChange management
Company, partner and people names are left out and details are generalized to protect confidential information.
C-suite
approved for global adoption after a CEO demo
3 regions
US live, Latin America next, Europe after
4 teams
feed every trip: operations, supply, commercial, relationships
<30 min
for report, follow-ups and memory save, down from 2+ hours
01 · The problem
Every visit started from zero
At a global container trading and leasing company, three teams travel constantly: sales to customers and conferences, maintenance & repair (M&R) on depot review trips, and operations to meet suppliers. Each person reported in their own format, or not at all, so no team had a consistent picture of any depot. Open issues went unfollowed, and the next visit started over.
And every team looks at the same yard differently. Operations checks supplier strength and market structure, M&R checks repair capability and process improvements, and sales meets customers to find new ways to support them. Their reports need different things, and need to be saved and found differently. One generic template served none of them.
Before the trip: whoever traveled decided alone what to look at. Other teams were never asked.
During: notes lived in a notebook, a phone or an email draft.
After: 2+ hours per trip to write a report and follow-ups, often days later, in that person's own format.
Next trip: no shared history, so idle units and unrepaired damage were forgotten.
02 · What I built
A loop, not a report template
The agent is the visible part, but the value is in the loop: each visit feeds the next.
1 · Before: Prep sheet4 teams say what they need from this trip
2 · During: The visitRaw notes, any format, anywhere
3 · After: AI agentTeam email, report, and actions for each team
4 · Memory: Shared librarySaved by depot; open issues carry forward
↺ The next person to visit starts with what's still open
The visit loop · 4 stages
03 · The real design problem
One observation, every team
The hard part wasn't the AI. It was understanding what each team actually needs from a visit, so the agent can pull one line from anyone's visit notes and turn it into a different next step for every department. Reports are saved and tagged by team, so the whole company sees the same evidence instead of it landing only with sales.
One line from a visit report
“Low inventory across the entire yard. No one had 40' standards.”
Sales
Demand is outrunning supply, so it may be time to raise pricing on what's left.
Operations
Find a way to source 40' standards and supply the demand before a competitor does.
M&R
Get long-idle units repaired and out while inventory is low, and keep a closer eye on repairs moving through the yard.
Reframing it from grading reps to surfacing missed opportunities turned the people being measured into its biggest supporters.
The turning point
04 · Who built what
My part, and the team's
Requirements & design
Me: interviewed reps, operations and M&R; mapped what each team needs from a visit; defined the report standard and the loop.
Team: operations lead pushed coaching over grading; a senior rep added the trip mission.
First versions
Me: built the prep sheet, agent instructions, shared library, and save and recall flows.
Platform & access
Me: tested and rewrote instructions through each platform change.
IT: owns platform access and the security review.
Rollout
Me: demoed to the CEO and C-suite; supported the pilot at a national conference; extended it to M&R and operations trips.
Manager: sponsored the MVP. C-suite: approved global adoption.
05 · Workarounds
Building inside a traditional company
The team couldn't access the full agent platform
Rebuilt the agent in the simpler, shareable version so the pilot ran on time, then moved it back once access opened up.
Reports lived on personal drives
Moved storage to a shared library with columns for depot, date, rep, flags and themes.
Saving failed on multi-select fields and file paths
Rebuilt the save flow to convert names into the right format, and replaced a fragile file lookup with a library search.
The first output was too long for managers
Rewrote it depot-first, in the email body, as a skimmable memo with tasks grouped at the end.
06 · What I'd do differently
Lessons I'm carrying forward
Start with coaching, not grading.The first version scored people's notes. It would have felt like a performance review.
Settle platform access with IT before building.Moving between platforms cost weeks of rework.
Test the output with its readers earlier.The first reports were built for the writer, not the managers reading them.
07 · What's next
From visits to a living relationship record
A visit every six months is a snapshot. Relationships grow, or slip, in between. The next phase connects the ongoing conversation to the same memory: team and management updates, calls and email threads, changes in how a partner communicates and works with us. Each depot and partner gets a continuous record instead of a gap between trips.
Today
Each visit adds a report to the depot's history, and open issues carry forward to the next trip.
Next
Everyday touchpoints feed the record too, so anyone can see how a relationship is trending before they pick up the phone or book the flight.
