Veeva

Principal Product Designer, CRM

Veeva tasked me with making a more attractive version for viewing the report. Lipstick on a 14-year-old Frankenstein of a form wasn't going to fix anything, so I reframed the whole problem around respecting the user's workflow.

I changed it from a form reps fill out after the fact into a workflow that runs across the entire call and does most of the report's work before the rep ever reaches the form. Veeva was building four AI agents to ship on their own. My idea was to make them disappear into that workflow instead.

The Problem

Sales, Not Data Entry

The call report is the most dreaded form in pharma sales. Reps hate it. After a visit with an HCP they have to log what content they showed, how long they spent on each page, how it went over, what samples they left, what they handed out, plus their notes on the whole thing. It's a wall of fields, and it turns sales into a data entry job.

So they put it off, often letting a week of calls stack up to do all their reports on Friday, long after the details worth writing down have faded. The data suffers and the notes suffer, and the company plans its next quarter on both.

The Vision

Call Phases

From my time and interviews at Tact.ai, I knew a call isn't one moment of data entry. It's three phases, and the tool should treat it that way. There's the planning before the call, the call itself, and the update after. Veeva only really supported the last one, and not well.

There was another piece to it. Veeva had four AI agents in the works and each was going to ship as its own standalone feature. I looked at the same four and saw parts of one workflow the rep already lives in, not four tools a rep has to stop and go find.

I got the PM on board and started mapping the UX for each phase, with the goal that the report at the end could fill itself in from everything that happened before it, instead of starting from a blank form.

Phase One

Planning

Before a call, the Pre-Call Agent proposes a plan. It looks at the context that should shape a call but usually stays buried, the rep's past calls with that HCP, potential objections, a recent support ticket, an MSL inquiry, and then suggests what to focus on.

Through it all the rep stays in charge. Take the plan as is, adjust it with a quick prompt, or write their own. The AI sits in the workflow. It doesn't control it.

Additionaly the Content Agent suggests videos or decks or rich content that emphasizes and drives home the plan's message.

Phase Two

Execution

During the call, the rep is sitting across from an HCP, so the view drops the plan and centers on the content they're planning to show. It tracks what gets presented and how long each page holds attention, the response to the content, and more. This is standard.

What's new is the Media Agent. HCPs ask questions, pose objections, require evidence that the rep didn't anticipate. The Media Agent is available to find content deep inside presentations. Natural language searches into vectorized content allow fast and accurate access to the answers the rep needs.

Phase Three

Reporting

The call report form doesn't go away. Fourteen years of technical debt saw to that. What we can do is alleviate the burden of filling out all the content by taking what we know from the call plan, content shown, samples, signatures, all that gets carried over. What's left is the story of what actually happened, and that's where the Voice Agent enters.

The rep gives their notes in plain language, dictated or typed, ideally within ten minutes of walking out while it's still fresh, or later during the full update if that's how the day goes. The agent reads what they mean, not just what they said. "Follow up tomorrow with the efficacy data" turns into a task instead of yet another workflow the rep has to come back to. The Compliance Agent checks against the content and they're done. They're confirming what happened, not rebuilding it from memory.

A few minutes back on every call doesn't sound like much. Multiply it across a field team and a quarter and it's real time returned to reps, and cleaner data reaching the company a lot sooner.

The Takeaway

One Workflow

The point was never four agents. It was one workflow. Where Veeva saw four features to ship, I saw what the rep does, plan, make, and record a call. AI could live inside the workflow instead of off to the side. That's the same conviction I've carried through every regulated, mobile-first product I've worked on. Put the answer where the user already is. Favor structured, predictable output over an open-ended chat box. Let the software do the assembly so the person can do the judgment.

The report was never the problem. Treating the whole call as nothing but the report was.