Fleet managers run a company’s vehicles and drivers. They had all their data in the product, but getting an answer meant running reports and building spreadsheets by hand. I designed an AI feature that reads that data and tells them what changed and why it matters.
I led its design end to end, from what an insight is to how it reads at a glance, working with a product owner, four engineers and two data scientists.
What are the results of this work?
About 4,200 users activated it in the first two weeks.
That’s 14% of the 30,000 users it launched to in the US. They opened the feature and came back at least twice.
93% of beta users called it essential.
In a survey at the end of the six-month beta, 93% of the 300 people using it said the tool was essential to their work.
The situation
The data was all there, but the answers were many clicks and tasks away.
To get an answer, users had to pick from several reports and test settings until the numbers came out right. Then they downloaded CSVs and did the analysis in a spreadsheet themselves. When the result still didn’t answer their question, they got frustrated and stopped coming back to that part of the product.
The goal
Show users what changed in their operations, so they spend their time acting on it.
Fleet managers needed to find out what mattered without building the analysis themselves. Verizon Connect wanted its share of the AI market, with a feature built on its own fleet data.
Decision 01
An insight is a sentence a manager can act on.
The most important decision was the format of an insight. We had technology that could read and analyse a whole fleet’s data, so the question was what shape the answer should take.
Two things decided it: what we could build, and how our users work. A chat wasn’t possible yet because our platform wasn’t ready, and fleet managers have busy routines that need something they can act on fast.
Dashboard charts could work, but they still leave the analysis to the user. So we bet on ready-to-read analysis in plain language, something a manager can scan and get value from straight away.
“The things that you find I would never have seen probably 90% of the time, which was surprising. We haven’t been tracking how many times the truck stops or how long they idle.”
Decision 02
Insights are scannable first, so the story is clear in seconds.
A feed only works if every card makes sense at a glance. My assumption was that if people had to read every card in full, they would never use the feature. They needed to scan, with the key facts standing out.
I tested different versions of the card with customers and learned that not everything has to be on the card. It has to say what the insight is about and how much the number changed. The rest can wait until you open it.
“I was expecting a long list, but you’ve kept to the high points. And when you open the insight then there’s the detail, which is how it should be.”
Decision 03
Opening an insight explains the number, and keeps a timeline while the problem lasts.
One of the hardest parts was explaining a number in language our users could understand. That took many versions with engineering, language guidelines for the agent and rounds of testing with users.
An insight is also a story, and sometimes it has a timeline. A number that keeps rising started somewhere, and it may keep going.
So I designed a panel that links insights from the same timeline, with the same numbers and context. A manager can see where a problem began and whether it’s still growing.
Headlines under 8 words with one number and a natural time frame. Subtext under 20 words that names the cause, or the change between periods when there isn't one. No decimals for durations, and no emotive words.
The outcome
A tool users called essential to their work.
Operational Insights was one of the first AI analytics features in the fleet software market. It helped Verizon Connect compete on AI, and helped managers get their answers without building complex reports.
14% came back within two weeks.
Of the 30,000 fleet managers it launched to, 14% opened the feature and returned at least twice.
93% of beta users called it essential.
Across six months of testing, 93% of the 300 people using it described the tool as essential to their work.
“Insights save me at least 45 minutes per week by eliminating the need to run and manipulate reports.”