Slides · AI · Kickoff to public release
AI activity generation in Slides
Generating activities in Slides with AI, built on Sidekicks (Miro's AI assistants) and the platform's wider AI capabilities. It was a new surface in Slides, so there was no playbook for what was feasible or what to ship first.
People kept the polls the AI made more often than polls they made by hand. It worked because it saved them the first draft, inside a deck they were already building, and they could still change it.
Why
Activities, such as polls, take time to build by hand. AI could build them for people, but this was a new kind of AI feature in Slides, and nobody had a template for what was feasible or what to ship first.
To get everyone aligned, I wrote down two versions: where the feature could end up in the long term, and a much smaller first release we could build straight away. That gave every team the same picture of the destination and of the first step.
My contributions
- Product manager from kickoff to public release.
- Two versions of the plan, a small first release and a fuller long-term vision, aligned stakeholders on what to build first.
- Worked through the real technical limits with engineering before we committed to a scope.
- Co-led customer research, so design followed the actual user journey.
- Co-defined the prompt design with the engineer, and together wrote the tests and safety limits for what the AI generates.
- Negotiated a scalable first version with the team behind Miro's AI assistants that combined both teams' visions.
The hardest call
There were two ways to build it. A direct line to the AI would have generated activities faster, but it needed a new loading design for the slide and more build time. Building on Miro's existing AI assistant was slower to generate but quicker to ship, so we started there.
Generating activities from a ready-made flow was the other route in. I left it out of the first release, so the team could ship one route properly. A smaller first release also carried less risk.
Who benefited
Presenters and facilitators. They run sessions for a whole audience, and keeping that audience engaged is one of their hardest jobs, so anything that helps them reaches many people at once. The Sidekick infers what they are trying to achieve and places several relevant activities in the deck to engage the audience, so they don't have to build each one by hand. The results show they keep what it generates.
Their audiences. They get activities to respond to during the presentation, such as polls.
Result
- People keep what the AI generates, more than what they make by hand. AI-generated polls (one kind of activity) are deleted less often than polls made by hand: only 40–45% are deleted within two days.
- People who try activities become more active. In the 28 days after someone first used an activity, they were active on 27% more days, took 89% more of the key actions that signal getting value, and used AI features 23% more. This isn't a controlled test, so I read it as a strong signal, not proof of cause.
Want to talk about this? Get in touch →