← work

ServiceNow · Agent workspace

Wrap-up

From a countdown to a workflow: how a few seconds after every call became a protected, structured and AI-assisted part of the agent's work, while keeping the agent in control.

role

Product designer, design owner

team

PM, UX research, engineering

timeline

2024 – 2026

status

Shipped across 4 releases

Wrap-up dialog over a live chat interaction, with AI-drafted code and notes shown close up
01

the problem

Agents had as little as 15 seconds after a call to record what happened, and the next call often arrived before they finished.

02

what I did

Started wrap-up alongside another designer, then took it over and owned it through four releases, into mid-call and AI-assisted wrap-up.

03

the result

A dedicated wrap-up surface, shipped in four releases, that became the place ServiceNow's AI writes its wrap-up drafts.

context

what research showed in 2024

After every call, agents record what happened: a wrap-up code and notes that supervisors, managers, admins and the next agent all rely on. Our 2024 research with voice agents found that the time allowed for this ranged from 15 seconds to 3 minutes.

When time ran out, agents turned off their availability to finish, finished later from memory, or made the next caller wait. Some took notes in Notepad during the call just to keep up. Rushed wrap-up meant less accurate records, more stress, and problems for whoever handled the customer next.

15s

the shortest time agents had to finish after-call work

3m

the longest, depending on the organization

4

audiences relying on the data: agents, supervisors, managers, admins

the question

How might we protect the moment after a call, so agents can record what happened accurately without losing their place in the work?

my role

I started wrap-up paired with another designer at my manager's request, then took it over and became its design owner, partnering with a PM, UX research and engineering. From there I carried it through segment-based wrap-up, Submit & Close, mid-call wrap-up and AI-assisted wrap-up, and worked with engineering on how time spent in wrap-up is tracked.

four releases

each one moved documentation closer to the moment it happens

  • Yokohama

    Protect the moment

    A modeless dialog with code, notes and timer. A default code submits if time runs out.

  • Zurich

    Fit the real work

    Segment-based wrap-up, so a transferred call gets a separate wrap-up from each agent.

  • Australia

    Make it lighter

    AI-drafted codes and notes that agents review, plus Submit & Close in one step.

  • Brazil

    Make it earlier

    Open wrap-up mid-call from the voice controls while the call stays live.

key decisions

and what each one traded off

decision 01

A dedicated surface that doesn't trap the agent

considered

Fields in the interaction form

Familiar, but on a form admins configure, often heavily, so wrap-up could scroll out of view.

considered

A modal

Has to be triggered to appear, and blocks the page while it's open, so agents can't look back at the conversation while writing notes.

chosen

A modeless dialog

Opens automatically when the call ends, and leaves the interaction underneath fully usable.

why

Wrap-up needed its own home at the moment a timer is counting down, but agents still had to look things up in the interaction. A modal would have locked that context away at exactly the wrong moment. Research also showed agents valued controls they could keep with them as they moved between tabs. A modeless dialog gave wrap-up a dedicated space without locking the workspace.

trade-off

We also considered letting agents minimize it. A minimized dialog would hide an expiring timer, so we cut minimize from the first release rather than risk agents losing track of time.

Three placements compared: wrap-up fields in the form, a blocking modal, and the chosen modeless dialog
decision 02

Let agents wrap up during the call, from the controls they're already using

Agents often knew the outcome before the call ended, but could only record it afterward. A later customer study made this concrete: agents had about 6 seconds after a call to choose from more than 200 codes, so they were taught to pick the code while still talking.

I kept wrap-up out of the page header and in the call controls, where the agent's attention already is. The harder question was where within them.

my first direction

The overflow menu

Wrap-up isn't a core call action like mute or hold, so I leaned toward keeping the main bar clear.

shipped

A top-level "Open Wrap-Up" control

A lower-priority quality-flag action moved into the overflow menu instead.

why I changed course

In reviews, it became clear that for teams with mandatory, timed wrap-up, this is something agents do on every call. Burying it would cost seconds they don't have. We separated optional and mandatory wrap-up scenarios, and reranked the controls around how often each is used.

trade-off

One more control in a crowded bar. Mid-call wrap-up is also voice-only for now; chat agents still wrap up when the conversation ends.

Voice controls with Open wrap-up in the overflow menu, and as a top-level control
decision 03

AI drafts. The agent decides.

When AI began drafting codes and notes from the transcript, the risk was agents accepting text they hadn't read, or losing work they'd already started. I designed around four rules:

01

Label it, until it's yours

AI content carries a sparkle. Once the agent edits it, the sparkle disappears, because it's now their content.

02

Suggest, don't decide

The recommended code sits at the top of the list under an "AI Recommendation" heading. Every other code is one scroll away.

03

Never overwrite the human

Anything the agent enters before the AI finishes is kept. After they edit, "Insert AI notes" lets them add the draft on purpose instead of replacing their work.

04

Fail visibly

If generation fails, a banner asks the agent to complete it manually. No silent blanks.

The details carried the same idea. I rewrote the tooltips ("Generate AI-assisted wrap-up content") so the UI never implied AI was required to finish, and styled AI labels as quiet information, not buttons.

Four AI wrap-up states: generating, draft ready with AI labels, agent edited, and AI unavailable fallback

the hard parts

a timed, cross-system flow behind a simple form

  • the timer runs out

    What happens if an agent submits with two seconds left, closes the page, or the contact-center timer drifts out of sync? A default code submits on expiry so no interaction goes undocumented.

  • going back

    Should agents reopen a wrap-up after it expires? We kept completed wrap-ups locked, consistent with other contact-center tools, so reporting stays trustworthy.

  • codes that change

    A code picked early could become invalid if the interaction changed afterward. We moved that case past the first release rather than ship inconsistent reporting data.

  • transfers

    Each agent on a transferred call gets their own wrap-up segment, so nobody's notes overwrite another's.

solution

the shipped experience

before

Before: only a wrap-up timer under the interaction title, with nowhere to record a code or notes

after

After: the first wrap-up dialog with timer, wrap-up code and notes
Active call with Open wrap-up in the voice controls, shown close up

01Mid-call wrap-up

Agents capture the code and notes while they listen, and the call stays active. Nothing has to be remembered for later.

AI-drafted wrap-up with the code list open under the AI Recommendation heading

02AI drafts

The code and a notes summary arrive pre-filled and clearly labeled. Agents fix what's wrong, then Submit & Close in one step.

impact

Wrap-up went from a countdown with no input to a dedicated workflow that shipped in four consecutive releases, giving agents, supervisors, managers and admins a consistent record of how every interaction ended.

The direction held up as more evidence came in. A later customer study showed agents choosing codes during the call, which mid-call wrap-up now supports. Later agent research found agents valued pre-filled wrap-up but wanted control over AI output, the principle the AI rules were built on. Time in wrap-up has since become part of how ServiceNow tracks handle, hold and after-call time for every interaction.

reflection

what worked

Designing for agency

Every AI decision came back to one question: does the agent still own the record? That made hard calls, like keeping their input over the AI's, easy to explain.

what I'd change

Test the AI experience directly

Validation for AI wrap-up came later, through broader agent research. I'd push for a focused study with agents before release, especially on how they review AI-drafted notes.