The Suggestion Nobody Takes

You are a product designer at Tallgrass, the field-service software company from the Practice challenge. This time you are on Tallgrass Dispatch, the web console used by dispatchers. 

Smart Assign suggests which technician should take each incoming job, based on skills, location, current route, parts on the truck, and the customer’s history. It shipped nine months ago. Here is how it appears today: 

The dispatch board is a grid: unassigned jobs down the left, technicians across the top as columns, the day laid out vertically. When a new job arrives, a small purple sparkle icon appears on the job card. Hovering the sparkle shows a tooltip: “Suggested: Marcus D.” Clicking the sparkle assigns the job to Marcus and the card moves into his column. There is no other Smart Assign surface anywhere in the product. 

The numbers, from nine months of use across 1,400 contractor accounts: 

Signal	Value
Jobs where a suggestion was shown	96%
Suggestions accepted (sparkle clicked)	12%
Suggestions where dispatcher assigned the same tech manually, without clicking	31%
Suggestions where dispatcher assigned a different tech	57%
When followed, jobs completed on first visit	84%
When overridden, jobs completed on first visit	79%
Dispatchers who have used Smart Assign at least once in the last 30 days	44%
Median time a dispatcher spends assigning one job	11 seconds

Two things from the field. Dispatch managers in interviews said versions of the same sentence: “I don’t know why it picked him.” And a Tallgrass support lead noted that the four largest accounts have all asked whether Smart Assign can be turned off entirely for their team. 

The suggestion is better than the dispatcher 5 points of the time — and the dispatcher takes it 12% of the time. Redesign the interaction so the dispatcher can decide whether to trust it. 

Constraints to consider 

• You cannot slow down the 11-second assignment. Dispatchers at a busy contractor place 60–200 jobs a day. Any design that adds a step to the common case will be turned off, and you have four large accounts already asking for the off switch. 
• The model produces a ranked list of technicians with scores, but no natural-language reason. If your design shows a “why,” you have to say what it is built from — you cannot invent an explanation the system does not have. 
• You cannot add a new data source. No new fields, no new integrations, no asking dispatchers to enter anything they don’t already enter. 
• Legal has told the team that any customer-visible surface produced or influenced by the model must carry a clear AI disclosure — but the dispatch board is internal, so you must decide whether disclosure applies here at all and say why. 
• One designer, and engineering capacity for one release. You get one intervention, not a program of them. 

AI Usage Guidance 

We expect you to use AI tools. We evaluate how you use them — not whether you use them. Evidence of iteration, redirection, and critical evaluation scores higher than a polished output with no process documentation. 

The single highest-signal indicator: your video answer to the mandatory AI question. If you cannot name a specific moment where you redirected AI output, evaluators will assume you did not. 

Mandatory AI question for your video: Walk me through one moment where you disagreed with, pushed back on, or redirected what the AI gave you — and what you did instead. Name the specific moment. Explain what the AI produced that didn’t meet the bar, what you did differently, and why. 

Communication clarity matters for this role. We assess structure, stakeholder awareness, and ability to simplify complex ideas — not presentation style or accent. 

A note on tools and fidelity. We are not scoring visual polish and we are not scoring your tool. A photograph of a pen-and-paper flow sketch, taken on your phone, scores exactly the same as a Figma file. What we score is whether the frames and annotations tell us what you decided and why. Do not spend your twenty minutes on finish.