How Pixieset achieved 35% AI feature adoption by solving the right problem with Amazon Bedrock
Amazon Web Services Kinman Lam
Pixieset used Amazon Bedrock to add AI alt text for photos, and 35% of users stuck with it. They didn’t chase flashy AI; they fixed a boring problem photographers actually hate.
Based on reporting by Amazon Web Services, Kinman Lam — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Pixieset found a rare AI success by aiming straight at a nuisance instead of a novelty. The photography platform used Amazon Bedrock to build an image alt text generator, and 35% of the users for whom it applies are still using it 16 months later. That matters because the market is crowded with AI features people try once and ignore.
The company’s starting point was simple: photographers spend too much time on work that isn’t photography. Alt text fit that test. It is tedious to write, especially when a portfolio can contain hundreds or thousands of images, but it also affects whether images can be found by search engines. Pixieset saw a chore that users should do but often postpone, not a creative task to automate away.
That distinction shaped the product. Instead of blanketing entire websites with AI output, Pixieset started one image at a time. A photographer can review a single suggestion, accept it, edit it, or reject it. Only after that do they expand the feature, and even then every caption stays editable. The idea was to let trust build gradually, not demand faith up front.
The company says it went from concept to production in four months and shipped the feature to millions of users at the beginning of 2025. In the first week alone, Pixieset generated alt text for more than 750,000 photos, and the feature drove significant subscription upgrades right away. For a company that hosts over 8 billion photos and serves millions of photographers worldwide, the scale is real, not a demo.
Technically, the integration was deliberately light. Pixieset already ran on an event-driven setup with Amazon EC2, AWS Lambda, and Amazon SQS. It added one more step: send the uploaded image to Amazon Bedrock, generate the caption, store it, and show it inside the website builder photographers already use. No new infrastructure. No GPU provisioning. No model hosting. The company also used Cross-Region inference and a fallback model to keep the feature available, and says it has had zero downtime since launch. And the lesson is pretty plain: if you automate the wrong thing, users notice. If you automate the chore they were avoiding anyway, they may even thank you.
My take — AI-written commentary, not fact-checked reporting
This is the kind of AI story that should embarrass the industry a little. The winning move wasn’t louder models or a prettier demo; it was resisting the urge to automate the thing users actually care about. Most AI products fail because they solve the vendor’s itch, not the customer’s, which is a very expensive hobby.
Read more about this at: Amazon Web Services