Gen AI · Contact center · 2023–2024

Turning conversations into context, in seconds.

A human-in-the-loop AI summary, so advocates get the context without rereading the transcript.

✦The first Gen AI feature in the Sam's Club contact center
ContactCenter AI chat summary showing the caller's concern, the resolution and an Add summary to the incident button
Company
Sam's Club (Walmart)
Domain
Enterprise · Gen AI
Role
Lead Product Designer, Gen AI
Part of
Advocate Unified Workspace
  • 8,110+AI summaries generated
  • ~70%accepted without edits
  • −7saverage handling time
  • ~$460Kestimated annual savings
01

The problem

Advocates were documenting the conversation while trying to have it. Watching them work, they scanned every chat for the same three answers:

  1. 1

    What happened?

    What is the member contacting us about?

  2. 2

    What's been done?

    What troubleshooting, resolution or promises already happened?

  3. 3

    What's next?

    Is anything unresolved or needs follow-up?

02

Design iterations

Shorter wasn't the same as scannable. Three versions, each reorganizing the summary around how advocates decide.

  1. Compress
  2. Structure
  3. Prioritize
V1 · Paragraph
Chat summary

The caller bought a microwave online that arrived damaged and wanted to know if they could return it at the store. The agent confirmed they could, but that the online return already started would need to be cancelled. The agent cancelled it and told the caller to bring the order and microwave to the store with the receipt.

Shorter, but still prose to scan.
V2 · Structured
Chat summary
Caller's concerns

Microwave ordered online arrived damaged. Can it be returned in store?

Resolution provided

Online return cancelled. Bring the order and microwave to the store with the receipt.

Predictable sections to jump to.
V3 · Signals
Chat summary
ExchangeDamaged product
Caller's concerns

Damaged microwave from an online order. Wants an in-store return.

Resolution provided

Online return cancelled. Return in store with receipt.

Add summary to the incident
The signal before the detail.

Recreated to show the evolution. Example conversation, not real member data.

03

Designing trust into AI

AI writes the summary. The advocate owns the record. Nothing enters the member's history without a person.

  1. 01

    Generate

    AI drafts the summary from the conversation.

  2. 02

    Review

    The advocate sees it first, with a clear note that it's AI-generated.

  3. 03

    Edit

    Fix anything missing or wrong in place, no regenerating.

  4. 04

    Confirm

    Only “Add to incident” saves it to the record.

AI proposes → Human verifies → System records

Chat summary card with topic tags, Caller's concerns, Resolution provided, Cancel and Add to incident buttons, and an AI disclaimer
Review and edit
The confirmed summary saved into the incident's call or chat history
Confirmed, then recorded
04

When AI is wrong

Failure wasn't an edge case. It was part of the interaction model.

Incomplete
The advocate adds what's missing
Incorrect
Corrected before it's ever saved
Unhelpful
Cancel, and the original conversation is still there
AI-generated
Always labeled as such
Permanent record
Requires an explicit human confirmation

Trust didn't come from making AI look infallible. It came from making it correctable.

05

Impact

8,110+AI summaries generated
~70%saved with no edits
~30%refined by advocates before saving
~$460Kestimated annualized labor savings

The ~30% isn't a failure rate. It's the editable workflow doing its job. Closing tickets with AI also took 7 seconds off average handling time.

“Another win for our contact center advocates and members. I love the path we are on with new self-serve and gen AI capabilities.”

Contact center leadership, at go-live
Behind the designThe problem in depth & what's next

Two kinds of load

During a conversation

  • Capturing notes by hand while listening, troubleshooting and navigating the member's issue

Between conversations

  • Scanning long chat histories to piece together what happened, what was tried, and what comes next

How might we give advocates enough context to act, without asking them to read everything that came before?

What's next

From summarizing the past to helping with what happens next.

Smarter structure
Summaries that adapt to each type of interaction
Quality signals
Learn why advocates edit, and feed it back into the experience and model
Cross-channel context
Continuity across chat, calls and past incidents
Next best action
Go beyond what happened to what to do next
Next case study

The workspace behind the AI.

Enterprise Contact Center: one workspace instead of a maze of tools.

Contact Center

Questions, or want the full walkthrough?

Some of this work is confidential, so I share the details one-on-one. Happy to walk you through it.

Quick read · about a minute

Gen AI Summaries, in a minute.

Lead Product Designer, Gen AI · Sam's Club (Walmart) · 2023–2024

ProblemAdvocates took notes during conversations and reread long chats to catch up.

SolutionAI drafts a structured summary; the advocate reviews, edits and confirms it before it's saved.

Jump to a chapter

  1. 01The problem
  2. 02Iterations
  3. 03Designing trust
  4. 04When AI is wrong
  5. 05Impact