IntercomDiscovery6 min readTwo products pretending to be one conversation
A chat bubble for the customer, a queue for the team.
- The surface
- The messenger a customer opens on a website, and the shared inbox where the team picks the conversation up.
- What the user wants
- As a customer I want an answer without opening a ticket. As support I want to handle fifty of these without losing track.
The launcher
A persistent bubble sits in a screen corner across the product and the marketing site, styled to the host brand, and opens a panel over the page rather than navigating away.
Help without leaving the task
Sending a stuck user to a support site costs them their place and often the whole session. Overlaying the conversation keeps the problem visible behind it, which makes the user's description of the problem better. A fixed corner position also means they never have to look for it.
Deflection before the human
Typing a question surfaces help centre articles and, more recently, an AI agent answer before a human is offered, with an explicit route to a person.
Tiered response
Most inbound questions have been asked before and answered well somewhere the customer could not find. Answering them at the point of asking is faster for the customer and cheaper for the business. The escape hatch to a human is what keeps this from reading as a wall.
Async by default
The conversation persists if the customer closes the tab, and replies can reach them by email or by a notification when they return.
Remove the waiting requirement
Live chat that dies when the window closes forces the customer to sit and wait, which caps how many conversations one agent can hold. Making the thread durable lets both sides treat it as messaging rather than a call. That single choice is what changes the economics of a support team.
The shared inbox
Conversations land in a team queue with assignment rules, and an agent can assign, snooze or close, with unassigned work visible to everyone.
Make the queue the system of record
Support scales on whether a conversation can be picked up by someone other than the person who started it. A visible shared queue turns individual memory into team state. Snooze is the honest acknowledgement that most tickets wait on something outside the agent's control.
Context alongside the reply
The agent view shows the customer's attributes, plan, past conversations and recorded product events next to the message thread.
Put the evidence next to the decision
The quality of a support answer depends almost entirely on knowing who is asking and what they just did. Pulling that into the same pane removes both the lookup and the temptation to skip it. It also turns each conversation into a small discovery interview the agent can conduct competently.
Support as a research feed
Conversations are taggable and searchable, so recurring problems can be counted rather than remembered anecdotally.
Instrument the qualitative channel
Support is the highest volume source of unsolicited user problems most companies have, and it is usually wasted because nobody structures it. Tagging converts a stream of complaints into something a product team can prioritise from. The risk is that tags reflect what support noticed rather than what users meant.
What reaches a person
What not to copy
- A proactive chat bubble that opens itself is an interruption dressed as help, and the version that pops up on a pricing page is a sales mechanic pretending to be support.
- The deflection layer gets tuned against cost, not resolution. When the article suggestions and the AI agent make the route to a human hard to find, the product is optimising a number that is not the customer's.
- Pricing has been repeatedly restructured around seats, contacts and resolutions, and customers have found the same usage priced very differently across changes. Forecasting the bill is a project.
- The customer sees a chat bubble and infers a chat response time. The async model that makes support scalable delivers a slower answer than the interface implies, and the gap is where the anger comes from.
The takeaway
Tag the support queue before you commission the research study you were going to run instead.
Finished the teardown? Bank it and the day counts toward your run.
Where the principles come from
- The Mom Test, Rob Fitzpatrick
- Continuous Discovery Habits, Teresa Torres
- 10 Usability Heuristics for User Interface Design (help and documentation), Jakob Nielsen, NN/g
Written from public behaviour of the product, not from inside it. Interfaces change often, so treat the flow described here as of the time of writing and check the live product before quoting it.