Intercom · AI product · Mid · 45 to 60 minutes

How should Intercom scope a support bot?

Intercom makes customer messaging software that businesses use to chat with prospects and support customers on their websites and apps. Answer this product management case and get it graded line by line against a weighted rubric.

product strategy
AI scoping
conversation UX
experimentation
trade-off judgment

The situation

You are a PM on Intercom's messenger experience. Leadership wants a recommendation on whether Intercom should invest in a bot inside the messenger, and if so, what the first version should actually do. The timing is tricky. Chatbots are getting a lot of attention in the market, so there is pressure to show an AI story. At the same time, many early bots are frustrating, overpromise on natural language, and make it harder for people to get help. Intercom customers mainly use the messenger to talk to human teams. Still, there are moments where a human is unavailable, such as after hours, or where the same small tasks repeat across many conversations. You need to decide what user problem is worth solving first, how much conversational behavior to attempt, and how to avoid damaging trust in live chat. Your recommendation will shape the next planning cycle, so it needs to cover product scope, user experience principles, and how success will be measured. The hard part is balancing ambition with a narrow, reliable experience that works with existing support workflows.

What to produce

A one to two page product recommendation memo.

Your answer must

Define the user problem the bot should solve first, and explain when a bot is helpful versus when a human should stay in control.
Recommend an MVP scope with up to three concrete use cases, plus explicit non-goals for the first release.
Describe the conversation design principles for the MVP, including how the bot should present itself, how copy should set expectations, and how users can reach a human or type freely.
Propose success metrics and guardrails that capture both value and harm, including a plan to learn from early conversations after launch.
Identify the biggest trust, usability, and operational risks, and give a mitigation for each.

Constraints

Do not assume human-level NLP or broad open-ended AI conversation ability.
The solution must fit inside Intercom's existing messenger and complement human support, not replace it.
Users must have a clear path to reach a human or leave a message, rather than being trapped in a rigid menu flow.
Keep the first release narrowly scoped to simple, repeatable tasks.

Format

Use brief headings such as problem, recommendation, scope, UX principles, metrics, and risks.

How Intercom actually played it

Written at the peak of chatbot hype, this post shares what Intercom learned building its own bot as expectations fell into disillusionment. It focuses on how bots fail when execution is weak and what product choices help them help users. An early, honest take on conversational product design.

Hype cycles punish weak execution, so set modest scope.
Bots should help users reach humans or answers quickly.
Learn from real conversations before expanding.

Built from What we learned building our bot by Kostya Gorskiy.