Adam Rotmil
McLeod Software · LLM for logistics

Respond

AI-drafted replies for the operators who guide truck drivers.

Lead product designer · Product design & discovery · Web / desktopInside the workflow

The problem

Respond helps specialists guide long-distance carriers to their next pickup and drop-off locations. Operators begin with responses pre-drafted from live data, review them, and add the human contact that drivers value for the long haul.

My role

Lead product designer: product design and discovery, from concept through customer testing to development handoff.

Outcome

Time-on-task baselines showed replies that used to take hours drafted and waiting for review. McLeod released it in 2025 as MPact.RespondAI, the company’s first AI-driven application.

The business case

Lost revenue per 30-minute fleet delay

In the conversation

Call center operators start their day with high quality responses already drafted, grounded in real context.

The LLM pre-drafts each response using real-time data. Operators review it, edit if needed, and accept it, which sends it under their name. Testing and co-creation with operators have proven that the responses are as good as, and often better than, if the user had manually looked up the contextual information themselves.

A rebuilt messaging screen. A driver asks for the details of their next load. Respond drafts the answer inside the thread. The operator edits the draft, adds a personal note about a place to eat on their break, and sends it. The driver thanks them.

Decision

I designed this experience to keep the operator in the loop, because they are ultimately responsible for the messages sent by their agents. Also, human operators are freed up to provide a conversational touch to the drivers during long hauls.

They can see what the model drafts, steer it, and keep the personal touch drivers expect, since operators message drivers directly.

The same pattern,
in email.

The reply is drafted for the operator, who can send it, regenerate it or discard it.

Templates

Templates that steer the model.

Operators create reusable templates with variables (shipper, date, route, weather) that inform the prompts behind each response, for consistency and accuracy.

What we tested

Several rounds with active customers, on click-through prototypes.

Active customers · Conversation guide · Hypothesis board

  1. 01

    Overwhelmingly positive

    Operators confirmed that the draft messages were accurate, helpful, and actionable. The design satisfied one of the most commonly requested enhancements: “just tell me the next best action I should take.”

  2. 02

    Operators stay in control

    Edits and ratings give a runway to improve the quality and accuracy of responses.

  3. 03

    Research, shared

    I wrote the conversation guide, set up a hypothesis board and taught the team research methods.

Outcome

McLeod released it in 2025 as MPact.RespondAI, the company’s first AI-driven application. I stayed deployed with the client to work on additional software suites.

Credits

Role
Lead product designer: discovery, interaction design, prototyping, design system, user research
Worked with
Van Carlisle (product manager), Sujit Kunwor (data scientist), engineering
Client
McLeod Software
Surfaces
Messaging, email, templates, reporting, command center
Methods
Click-through prototype tests, conversation guide, hypothesis board
Delivery
Concept through customer testing to development handoff
Release
McLeod announced MPact.RespondAI on April 9, 2025, available to customers from June 1, 2025.
NextMy WellnessBuilding a research platform now used in NIH-funded cancer trials