Ask any data team and they’ll tell you: maintaining analytics is a nightmare. Especially at scale. And the easier analytics becomes to create, the harder it becomes to maintain.

Almost everyone at Ollie has access to Lightdash. People can explore governed data and build what they need, but that also means they’re creating alternate versions of existing dashboards, one-off charts, and copies of copies of copies.

Cleaning them up used to mean reviewing each chart and deciding whether it still mattered. In practice, it rarely happened. So Ollie switched on Autopilot and let an agent do the maintenance instead.

Here’s what happened after four months of running Autopilot at Ollie:

• 1,183 maintenance actions logged

• 489 stale charts and dashboards retired

• 57 broken charts repaired

• 71 fixes written out and handed to the team

And only 3 actions needed to be reversed by a human.

Autopilot turned cleanup into a four-minute nightly job

Lightdash Autopilot is a scheduled AI agent that maintains a Lightdash project so you don’t have to. It finds stale or broken content, fixes what it safely can, and flags the rest for your data team to review. Anything it removes is soft-deleted, so admins can restore it, and every action appears in an activity log with the reason it was taken.

Ollie configured Autopilot around a 90-day stale-content rule, meaning Autopilot could delete or modify anything that hadn’t been viewed or touched by a member of the team in 90 days. It has run every night since mid-May: 127 runs, roughly 4 minutes each.

The ability to easily recover content made it easy for the data team to trust Autopilot. Now, the team could use the agent to do the cleanup instead of creating a queue of dashboards for someone to review.

In just one run, Autopilot surfaced about 43 stale charts. Across four months, Autopilot retired 489 charts and dashboards: finished campaign analyses, duplicated dashboards, and one-off explorations nobody had opened in months.

By September the backlog was gone, and the nightly report had shrunk to just a one-line health check.

“Before Autopilot, dashboard cleanup was a manual job that rarely happened. Now it runs in the background. We can restore anything it removes, and after months of scheduled cleanup, nobody has complained that anything important is missing.”
- Ben Pannell, Tech Lead, Data Engineering @ Ollie

Then a dbt change broke 50 charts overnight

Company-wide access creates another maintenance risk: when someone renames a field in the model, saved content can break silently across the business.

Back in August, a routine dbt change renamed two fields at Ollie and broke about 50 charts. Most of them on a single dashboard. Autopilot’s next report was able to identify both root causes, and it repaired 42 charts across the week.

But it didn’t fix everything. One chart bucketed data on the renamed field in a way that might have changed its meaning, so Autopilot flagged it for human review instead.

The thing is, a renamed field isn’t a hard fix. It’s 50 easy fixes, chart by chart, but a lean data team rarely has the time for it. With Autopilot, most of these changes happened overnight, while Ollie’s data team slept.

Lightdash replaced a Looker setup Ollie could no longer iterate on

Ollie became a Lightdash customer more than two years ago, when the team was refactoring its data lineage. Looker’s development felt deprioritised, and overlapping Explores and duplicate metric definitions had made its LookML difficult to untangle.

“It was clear that we weren’t able to iterate as quickly as we wanted, and the self-service was confusing.”
- Ben Pannell, Tech Lead, Data Engineering @ Ollie

Lightdash fit the stack Ollie already had. Instead of maintaining a separate LookML layer, the team keeps metric logic with its dbt work and uses one governed semantic layer for charts, dashboards, and exploration.

The Lightdash team translated Ollie’s existing LookML into starter YAML, putting around 80% of the initial migration in place. Ben described it as “a much easier lift for a lean team.”

“Having an opportunity to create a semantic layer and have one definition for a metric was a game changer.”
- Ben Pannell, Tech Lead, Data Engineering @ Ollie

One governed metric layer now serves the whole company

Today, business users across Ollie explore governed data and share findings without touching SQL, while analysts keep edit access and inspect the generated SQL when they want to go deeper.

“Almost everyone at Ollie has access to Lightdash, and we hear almost no complaints. That's the best sign it's working: people can explore trusted data without bringing every question back to the data team.”
- Glenn Schechter, Senior Director of Data & Analytics @ Ollie

Ollie’s next step is letting more people ask questions conversationally. That means more content, created faster, by more people and more agents - and it only works if maintenance scales with creation.

Interested in automating your dashboard cleanup? Get in touch with our team today to try Autopilot for free.