How TripleLift let every team start running without waiting on engineering

Something is always broken

TripleLift is a global ad exchange focused on creative, serving trillions of impressions a day. Each one connects thousands of publishers with tens of thousands of advertisers, and every connection is a chance to make money or lose it. Engineering goes into the product. Keeping everything running across all those systems falls to people.

Pilots that never became a team effort

TripleLift adopted AI early and widely, enough that Carolin calls those first years a "Wild West" of shadow AI. But AI assistants made individuals faster, not teams. Everyone built their own thing, and the results sat in individual download folders where no one else could reuse them. The toughest jobs, the shared processes that run across teams, needed workflow tools and engineers, so teams waited their turn and automation pilots stalled before reaching the rest of the company.

With Convey, the people who own the work train digital teammates themselves, and those teammates work for the whole company, not one person. After a 30-day proof of concept, TripleLift stopped lining teams up for engineering time and let every team start at once.

A governed team, by design

Carolin set up two kinds of digital teammates. A handful of company-wide teammates each own one critical responsibility: Felix creates all external collateral, Sasha owns customer intelligence, and Otto is the only teammate that touches internal APIs. About a dozen team teammates, for groups like finance, rev ops, sales enablement and CTV, hand off to them whenever work touches those areas. Sensitive access stays in a few places, which gave TripleLift the confidence to let every team build on its own.

That confidence comes from how Convey is built. Every digital teammate is set up like an employee, with its own identity, its own credentials and access scoped to what its job needs. One teammate can't see another's data or history, admins decide who can create, change and direct each teammate, every credential retrieval is logged, and prompts and outputs are never used for model training. AI pilots usually stall when leaders can't say which agents exist, what they can reach and who is using them. Carolin can answer all three.

In Slack, where everyone already works

Rob had seen plenty of AI rollouts stall for a simple reason. People are busy, and a new tool is one more thing to learn.

So every TripleLift teammate lives in Slack. People ask for help the way they'd message a colleague, which lowers the bar for anyone less technical or a little wary of AI. Behind the scenes, teammates hand work to each other, so the person asking never needs to know which one does what.

Shared channels also give Carolin a live view of demand. She can see who's asking which teammate for what, and a concierge teammate sends her a summary of those requests with ideas for new teammates to build. That shows her where people need better information and which process to train a teammate on next.

The results

AI at TripleLift is now a team effort. Work that used to take people hours across multiple interfaces now runs on its own, with fewer manual errors. Adoption started in services and has spread to legal, accounting and nearly every other department. One early project consolidated 54 Confluence spaces into nine. And because reporting is easy, Rob can measure ROI process by process.

For Rob, the real payoff is where that time goes.