Adam Rotmil
McLeod Software · LLM for logistics

Respond

Every reply has somewhere to go.

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

Project summary

Respond helps specialists guide long-distance carriers to their next pick-up 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.

The challenge

Operators had to query data tables to answer each driver, while more messages kept coming in.

The business case

Every 30 minutes a fleet is delayed translates to $1.37 million in lost revenue.

What changed

User research and time-on-task baselines showed the difference: replies that used to take hours were drafted and waiting, ready for the operator to review and send.

In the conversation

The draft arrives inside the thread.

The LLM pre-drafts each response using real-time data. Operators review it, make edits if needed, and accept it, which sends it as their proxy.

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

Human-in-the-loop isn’t a compromise. It’s a feature.

I put an operator in the loop instead of full autopilot: for observability, for steering, and for the personal touch, since operators message drivers directly.

Accepting a draft sends it as their proxy. Operators stay in the center, able to make edits and add their human insight.

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

    Customers who use the current product responded strongly to the drafted workflow.

  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 educated the team about research methods.

Outcome

McLeod released it in 2025 as MPact.RespondAI, the company’s first AI-driven application. The company president asked me to look at their other software and find opportunities.

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