Dropbox · AI product · Mid · 45 to 60 minutes

How should Dropbox add AI to file previews?

Dropbox is a cloud storage and collaboration platform where people store, share, and preview many kinds of files. Answer this product management case and get it graded line by line against a weighted rubric.

Product strategy
AI product sense
Metric design
Tradeoff analysis
Execution planning

The situation

You are a PM on Dropbox's web file preview experience. The engineering team has shown that Dropbox can already convert many file types, including documents, audio, video, and complex formats, into text through its existing preview pipeline. Leadership wants to use that capability to help knowledge workers get the gist of a file faster and ask questions without manually searching through long documents or meeting recordings. You need to recommend a v1 approach for AI powered summaries and question answering inside file preview on the web. The hard part is that this is not just a feature design problem. It is also a product economics and systems problem. Long files, many file types, and follow up questions can quickly drive up model cost and latency. At the same time, weak answers or shallow summaries will damage trust. There is also pressure to think ahead to collections of files, not just one file at a time, but shipping too much in v1 could slow the team down. Your manager wants a clear product recommendation that balances user value, quality, speed, and cost, and explains how you would learn after launch.

What to produce

Write a 1 to 2 page product strategy memo.

Your answer must

Define the primary user problem and the top 1 to 2 use cases to support first in web file preview.
Recommend the v1 scope, including whether to launch summaries, Q&A, or both, and whether to handle single files only or plan for multiple files from the start.
Set success metrics for quality, latency, adoption, and unit economics, including at least one metric that captures ongoing AI cost per use.
Propose a high level retrieval and context selection approach, including how you would think about chunking content and choosing what to send to the model.
Explain your launch and iteration plan, including experiments, trust or safety safeguards, and how you would decide what to improve next.

Constraints

The feature must work within Dropbox's existing web file preview flow and reuse the current file to text conversion system rather than requiring a brand new ingestion stack.
Users preview many file types and file lengths, so the solution cannot depend on a single narrow format or short documents only.
Model calls and context windows are limited, so the proposal must explicitly manage latency and API spend.
Users or admins may want control over whether files are sent to AI systems, so the design cannot assume all content is always available for processing.

Format

Use sections for goal, users and jobs, v1 recommendation, success metrics, approach and tradeoffs, launch plan, and risks.

How Dropbox actually played it

Dropbox added summaries and question answering to file previews and had to balance quality, speed and cost. Clustering approaches made summaries cover about 50 percent more topics, and later work cut cost per summary by 93 percent and per query by 64 percent. A concrete look at product tradeoffs in a first generation AI feature.

Quality, latency and cost trade off, pick targets early.
Cost per use is a product metric for AI features.
Iterate on how content is chunked, it changes what users get.

Built from Bringing AI-powered answers and summaries to file previews on the web by Dropbox Tech.