Dogfooding

In Brief

Dogfooding is an unstructured product research method where team members use their own product in real daily work, just as a customer would. You go through the actual workflows, note where things break or feel awkward, and record ideas for improvement. The output is a firsthand list of bugs, friction points, and unexpected use cases that formal testing often misses.

Description

Using one’s own product is standard practice among many technology startups and entrepreneurs. In many cases, entrepreneurs are building a product to solve their own pain points, so it is common to then use the product.

There is no predefined script for dogfooding, and it is not a formal quality assurance (QA) process. The main advantage to this method is that it may reveal unorthodox use cases that were not covered in the requirements or QA tests. Dogfooding should primarily be considered generative research and not an experiment.

AI has minimal impact on dogfooding, and that is precisely the point. You cannot delegate this to an AI agent or automated testing script — the value comes from personally encountering friction, confusion, or delight while trying to accomplish a real task. If your product includes AI features, dogfooding becomes even more critical: you need firsthand experience of how the AI behaves in real scenarios, where it fails, and where it produces unexpected results. Automated QA and AI-powered testing tools are valuable for catching bugs, but they are not dogfooding.

Companies that do not use their own products and services are sometimes criticized, in some cases very publicly.

How to

Prep AI Prompt

1. Set a clear scope. Pick the workflows and personas you intend to walk through (new-user signup, daily power-user task, admin path) so the team is exercising the same surfaces and you can compare notes afterward.
2. Set up a low-friction capture channel. Make sure you have a place to take notes that is easily accessible and won’t interrupt your workflow too much — a shared issue tracker, a single Slack channel, or a running doc.
3. Decide who participates. Pull in teammates outside the build team where possible; engineers and designers carry too much prior knowledge to notice every friction point alone.

Analysis AI Prompt

1. Be careful to interpret the results as generative and not evaluative. The makers of a product have an intimate knowledge of the product design and likely cannot capture the uneducated user’s perspective. This is especially true in dogfooding the new-user process, where the makers of the product have a massive amount of prior information and expectations regarding the signup and onboarding.
2. Watch for missing edge cases, particularly when the team is not diverse. A team may not analyze the product for use by handicapped or minority users and thus overlook substantial aspects of its user experience. This becomes more of a problem as a product scales beyond an initial niche audience.
3. When multiple team members dogfood their product, collect and sort notes via card sorting, stack ranking, or other standard UX methods to surface the patterns rather than reacting to individual gripes.