SKIP TO CONTENT
← ALL WORK
04 — ENHANCE CONTACTS WITH AI · SAAS · AI · 2025

AI that tidies the contact list without taking the decision away from the agent.

The contacts MVP for Tawk.to. It saves details out of the conversation as they come up, flags duplicates with the evidence behind the match, and puts a summary of the person on screen before the agent replies.

Role
Product designer — dedupe & segmentation flows, hi-fi, prototypes, testing
Client
Tawk.to · shipped
Timeline
2025
Platform
Web dashboard
Contacts with AI — Tawk.to
Auto
contact details saved from conversations, no re-typing
0
destructive merges: every AI action shows its evidence and can be undone
3
AI insight types: summary, segments, suggestions
Live
shipped as the contacts MVP in the product

Problem

Teams holding thousands of contacts collect several copies of the same person, and with no way to group them, the same customer gets treated as a stranger twice. Agents lost time to cleanup and made more mistakes doing it.

Every request came with the same worry attached: a cleanup that quietly loses something.

Research

Agents accept a suggestion when they can see what it matched on.

I ran sessions with agents who manage thousands of contacts. They took any suggestion that showed its reasoning and could be undone, and ignored the rest. What limited automation here was never the model’s accuracy.

1000s
of contacts per workspace in the teams studied
#1
fear: losing data in a merge
4
needs: safe dedupe, explainable AI, segments, in-the-flow

Process

  1. DISCOVER
    Agent interviews, contact-database audits, duplicate-pattern analysis
  2. DEFINE
    Principles: explainable, reversible, embedded in the existing flow
  3. DESIGN
    Flows for deduplication and segmentation; hi-fi mockups in the design system
  4. PROTOTYPE
    Interactive prototypes for merge and segment flows
  5. VALIDATE & SHIP
    Usability testing on trust and clarity; shipped into the live product

Decisions

01

Show the evidence for every AI action.

Duplicate and merge suggestions list the matching fields and the confidence before anything happens.

TRADE-OFF

Each suggestion needs more room on screen. Without the evidence, agents ignored suggestions altogether.

02

Reversible by default.

Merges and cleanups can be undone, and nothing disappears without a record.

TRADE-OFF

History and rollback cost engineering time. Nothing else would have made merging acceptable.

03

Insights sit on the contact and in the conversation.

Summaries, segments and suggestions appear where agents already work, rather than in a separate AI panel.

TRADE-OFF

There is less to put on a feature list. Agents used it, because it never interrupted them.

PrototypeTap the glowing hotspots to move through the app, like the real prototype.

Contacts with AI — Smart contact creation
Contacts with AI — Duplicate detection
Contacts with AI — Merge suggestions
Contacts with AI — AI contact summary
Contacts with AI — Segments and grouping
Contacts with AI — Product and lead suggestions
Smart contact creation
Details saved straight out of the conversation, so nobody re-types them.
01 / 06

Before → After

BEFORE

Duplicate, fragmented contacts, manual cleanup that risked losing data, and no way to group or personalise at scale.

AFTER

Details save themselves as the conversation happens, duplicates arrive with the evidence for the match, every merge can be undone, and the summary is already on the agent’s screen.

Outcome

  • Contact records stay cleaner, and agents open a conversation already knowing who they are talking to.
  • Visible reasoning and reversible actions made the automation safe enough to use.
  • Manual cleanup dropped without moving the decision away from the agent. Shipped into the live product.

Reflection

Merging duplicates only helps if the agent believes the match. The honest measure is how often merges are accepted against how often they are undone, and that is the first thing I would track next.