How to decide where AI belongs in your product, and what to build first

To decide where AI belongs in your product, start with how your users already use AI and where it's letting them down, map the full customer journey (including the parts that happen outside your product), pick the few moments where you can add the most, and put the AI in the channels your users already use instead of launching something new. That's the process we followed at Pangea, a consumer social travel platform, and it's why the first AI feature we shipped is an email.

1. Ask where their current AI falls short. Your users are probably already using ChatGPT or Claude for the job. By March 2026, 56% of U.S. travelers used AI to plan trips, and over 70% of those under 35. So we asked what wasn't working. Three answers came back: memory that resurfaces the wrong things, generic "must visit" lists instead of recommendations from people they trust, and gotchas nobody knew to research (a city with no Uber, a yellow fever card required because of where you flew in from). That last one became our design principle: the most valuable travel help is usually the answer to a question you didn't know to ask.

2. Map the whole journey, not just your part of it. We mapped every stage of a trip: discovering places, deciding where to go, booking, preparing, the trip itself, and sharing recommendations afterward. Then we marked which moments happen inside Pangea and which happen in browser tabs and group chats we never see.

3. Pick the moments where you can add the most, then sequence by difficulty. Three stood out: Discover (destinations that fit your style but aren't on your radar), Prepare (getting past top-10 lists), and On-trip (a personal assistant for pivots and changing plans). Discover could have the biggest impact, but it needs to run through every surface of the app. So we built the infrastructure for it and shipped Prepare and On-trip first, since they help everyone with a booked trip right away.

4. Put it where users already are. Prepare is an email sequence, not an in-app feature. 71% of Pangea users open one of our emails every month, and email is easy to pull up mid-trip. There's also nothing new to learn: adding a trip, something users already do, triggers everything. A month out you get neighborhood picks, a week out money and connectivity, and the night before a sixty-second checklist.

5. Use the data only you have. A general model can write a decent Kyoto guide. It can't know that a friend of yours will be there the same week, or that three people in your network vouch for the same tiny counter in Nishiki. Build around that data. It's the part a competitor, or ChatGPT, can't copy.

For a GTM team the questions are the same: where is your buyer already using AI, what does it get wrong for them, and which channel do they already check every day?