How PetScreening turned a dbt pipeline into a pet tag factory

At dbt Summit 2026, Nick Fernandez, manager of data analytics at PetScreening, and William Guicheney, principal analytics engineer at Aimpoint Digital, walked through one of the more unusual builds on dbt: a pipeline that ends not in a dashboard, but in a physical object landing in someone's mailbox.

PetScreening runs pet policy management software for property managers, handling vaccination records, risk assessments, and accommodation requests for assistance animals. Alongside that core business, the company runs FIDO Alert, a free service often described as "Amber Alert for pets." Members get a QR-coded pet tag linked to their phone number; if a pet goes missing, the network pushes an alert to everyone within a 10-mile radius.

FIDO Alert has grown to 1.6 million users and 2.2 million pets, and has helped reunite more than 75,000 lost pets with their owners.

Behind that tag is a dbt pipeline, and the core story is what happened when a software team decided to manufacture hardware in-house.

From outsourced and slow to in-house and instant

FIDO Alert's tag fulfillment didn't start with dbt. The team first outsourced manufacturing to external vendors, a process built on manually downloaded CSVs, emailed back and forth with no verification. The result was a three-month backlog just to get a free tag into a member's hands.

"Unreliable, expensive, and slow, which is a very bad trifecta of qualities," Fernandez said of that early vendor setup.

The team moved manufacturing in-house and connected it directly to their existing dbt project, treating the tag-printing machine the way they'd treat any other exposure: as a destination for modeled data, no different in principle from a BI dashboard, just with a mechanical arm on the other end instead of a chart.

Three years, three storefronts, one pipeline

FIDO Alert has gone through three major iterations, from a bolted-on storefront inside PetScreening's core product, to a dedicated Shopify storefront, to the current setup that now prints over 1,000 tags a day across two machines. Each transition swapped out the data source and changed the shape of the incoming data entirely.

That's where the team leaned hardest on dbt's modularity. Because the pipeline was built on a standard medallion architecture, models could be swapped in and out without ever interrupting production.

"By the time we got to V2, we had removed that issue, so now we didn't have tags that were orphaned. It was really an overnight switch of updating the stream of data going into the actual machine," Fernandez explained, describing how a fuzzy-matching step built to patch bad customer data in V1 became unnecessary once a cleaner storefront replaced it.

When bad data means wasted material, not just a bad chart

The team built the case for three capabilities that mattered most as FIDO Alert scaled: flexibility, reliability, and efficiency. Reliability carried different stakes here than it does for most dbt projects.

"For us, bad data doesn't result in bad numbers in a dashboard. It results in missed prints, reprints," Fernandez said.

A duplicate join that inflates a revenue number is an easy fix behind the scenes. A duplicate join that reprints the same tag 100 times is a few hundred wasted tags and a lot of postage.

dbt tests became the team's line of defense: uniqueness checks, non-null checks, and custom tests at each stage of the lineage, wired into Slack so failures get seen and diagnosed fast. As the pipeline matured, the team noted, those alerts started firing less and less, freeing up a data team of just five people to support every function in the business.

One model, two very different jobs

The efficiency story came down to reuse. The same gold-level model that feeds the tag-printing machine also powers the dashboard the growth team uses to monitor FIDO Alert day to day.

"It's truly one source of truth, so any work we do to improve any aspect of it improves all the exposures and downstream use cases," Fernandez explained. That single-source approach, paired with dbt's standard developer workflow of isolated PRs and dedicated dev schemas, let a five-person team ship changes to a live manufacturing pipeline without babysitting it around the clock.

The numbers back it up: PetScreening now runs more than 1,000 dbt models, has seen a 32x increase in data freshness, and cut the time from a new data request to an actionable answer by more than 90%.

What's next

The team is now building toward a semantic layer, with an eye toward letting AI agents work directly against well-documented, well-tested pipelines like this one, both to calculate metrics and to help monitor and debug operational data flows that, unlike most, have a physical machine waiting on the other end.

Watch the full session, plus every keynote and breakout from dbt Summit 2026, on demand. Watch dbt Summit 2026 on demand →