Litespace Coffee Chat cover

Litespace Coffee Chat

AI powered coffee chat for intentional, pressure free pairings across hybrid teams.

Project Type

Professional
Shipped
Luna, live prototype

Timeline

2 weeks · Sep 2023

Team

1 Engineer
Product Designer (Me)

Tools

Figma

What I did

  • Led end-to-end design of an AI powered coffee chat feature.
  • Owned the pairing model, from the matching signals to how a match is presented.
  • Designed the full scheduling flow, including the decline and reschedule paths.
  • Partnered with engineering on the MVP.

Bridging the gap between structured meetings and casual interactions

Coffee Chat is the 0-to-1 feature of Litespace, an AI powered enterprise platform for engagement in hybrid and remote workplaces. It pairs two people who would not otherwise meet.

63%

Users opted into Coffee Chat within the first month, signaling strong early adoption

200+

Chats generated in the first month through auto-matched pairings

Remote teams lack casual social interactions

I identified key challenges through observations, customer feedback, and teammate interviews.

Nobody wants to go first

Nothing tells you why you were paired, so accepting the invite and starting the conversation both carry social pressure.

A good pairing has to be inferred

There is no rich profile to match on. The only signals available are lightweight: profiles, onboarding inputs, and shared Slack channels.

Scheduling has two sides

Both people have to land on one time, and existing tools push that into direct messages, where message overload and drop-off follow.

Why existing tools fall short

Category Tool Pros Cons
Slack-Integrated Matching Bots
Donut (Slack bot)Donut
Seamless Slack integration
Fully automated matching
Scales for teams
Feels robotic
Repetitive if not personalized
Watercooler TriviaWatercooler Trivia
Fun and engaging
Low effort to run
Shallow connection
May become distracting over time
Virtual Spatial Platforms
GatherGather
Spatial, playful interaction
High sense of presence
Requires onboarding
Can feel game-like and distracting
Co-working Tools
FocusmateFocusmate
Promotes deep work
Accountability-based
Lacks social/casual interaction
Meant for individual focus

Existing tools approach casual interaction from different angles, but few clearly communicate the reason behind a specific pairing or reduce the social pressure of starting the conversation. This insight shaped Coffee Chat's direction: combining intentional pairing with a low pressure experience, without requiring direct message coordination.

Designing intentional connections with minimal social pressure

The rebrand from a green identity to a blue one was still in progress, and we had two weeks to design and ship. That scoped the MVP to one meaningful chat per week, and anything that added friction had to go.

Explain the match, and people accept it

Users were more likely to accept a pairing when they understood why they were matched. I championed using GPT-3 to infer a connection from those lightweight signals and explain each match.

AI-generated pairing card showing match reasoning

AI-generated intro

A warm profile summary grounded in each person's background, so users quickly understand who they're meeting beyond a static bio.

Explainable match

Transparent reasons for why the pairing was made.

Remove the messaging, remove the drop-off

Existing Coffee Chat Tool User Flow

  • ✕ Manual scheduling
  • ✕ Message overload
  • ✕ High drop-off
User A receives chat pairing DM notification sent Send DM or email to schedule Multiple back-and-forths Meeting confirmed Calendar link created

Our Coffee Chat Flow

  • ✓ Automated pairing
  • ✓ Clear fallback path (reschedule or skip)
  • ✓ No direct message coordination needed
User A receives chat pairing Accept chat (default) Reschedule chat Select user B's proposed time Suggest new times to user B Waits for user B's confirmation Decline / No response Next pairing in cycle Meeting scheduled

Simplifying Coffee Chat's interface

Before: original multi-functional widget
Before: One multi-functional widget. High cognitive load in user testing.
After: split into two simpler widgets
After: Two widgets, for availability and reasons. Won the internal A/B test.

Optimizing meeting format selection

I explored letting users choose between virtual and in person meetings. In person meetings were rarely chosen in our remote and hybrid setting.

Before: meeting format selection
Before: Users picked virtual or in person. A/B testing showed delays and confusion.
After: simplified scheduling without format selection
After: Format selection removed. Simpler, faster scheduling.

One pairing a week, scheduled for you

Each week you get a pairing built from your Google Calendar availability, with shared interests and suggested topics to start on. It arrives set to "Going" by default, so nobody has to be the one who says yes.

One pairing a week, scheduled for you

Handling chat declines without friction

If either person declines, or never responds, you both get a notification. No reason required. Asking for one would put the pressure back.

Handling chat declines without friction

Flexible scheduling for better availability

To reschedule, you pick from time slots the system recommends from your availability and theirs.

Reschedule with recommended time slots

No message thread to chase

If the other person proposes new times, you review them and pick one.

Review and select proposed times

When none of their times work

You send back your own availability for them to review.

Propose own availability

The identity was changing under the feature

With the identity mid-swap, there was nothing settled to design on. Building Coffee Chat on the system being retired meant shipping something stale on day one, and the one replacing it did not exist yet.

So I worked with another designer to define the new color tokens, typography, and components first.

New identity design system

When the barrier is social, the design work is subtraction

Every decision that made Coffee Chat work was a removal. There is no message thread to negotiate, the default answer is Going, and declining needs no reason. What looked like a matching problem was about deleting each moment where someone had to expose themselves socially.

Designing it mid-rebrand taught me the same lesson at the system level: build tokens and components, not finished screens, because those survive when the ground moves.

This time, the agent goes first

Coffee Chat solved the social barrier with structure. That was the right call in 2023, and it is not how I would build it now. So I rebuilt the flow as a personal exploration: Luna, a conversational agent that opens with a match already lined up and the reasons pinned beneath it.

Try it below: book the chat, move it, swap the match, or tell Luna what you would rather do. If it can't help, it says so.

Tell me what's on your horizon.
Hey, I'm Chang. Well, the AI version. Feel free to ask me anything.