How an AdTech Company Built an AI DataOps Team for Ad Operations

By Khoushik Reddy Ananth

Analysts at an AdTech data and operations company were spending 40 to 60 hours a week on manual data work.

By the time the numbers were ready, the window to act had often passed, so yield problems leaked revenue every day. Tools were spread across many data sources with no single control, and the reasoning behind decisions lived with individual people.

Today the company runs an AI DataOps team on LeafCraft, under AI governance that audits every action on the ad server. It went from its first workshop to production in six weeks, and it now recovers more than $12,000 a day in revenue.

Why ad operations were a hard problem

Yield problems have to be caught while there is still time to act. With analysts spending 40 to 60 hours a week on manual data work, the numbers often arrived too late, and revenue leaked every day.

Tools were spread across many data sources with no single control. The reasoning behind each decision lived with individual people, so it could leave with them.

The company needed enterprise intelligence for ad operations: agents that read the data every day, people who make the judgement calls, and one record of why each call was made.

What LeafCraft did: one AI agent management platform for ad operations

The company now runs an AI DataOps team on LeafCraft, set up across all five pillars.

Human-AI collaboration, with AI and people as one team

Agents read ad and revenue data from the ad server and Google BigQuery, spot yield problems and recommend what to do. The ad ops team approves those actions in Slack, so nobody has to learn a new portal.

Exceptions go to analysts, and their decisions go back to the agents, so the next similar case is handled the same way. This human-AI collaboration keeps analysts on judgement calls instead of manual data work.

AI governance that audits every action

AI governance covers every agent. Each one works under clear rules, and every action it takes on the ad server is audited.

Low risk actions are handed to the agents as trust in their record grows, while everything else waits for a person's approval.

AI cost control within a known budget

AI cost control caps every agent's cost before it runs, so agents can work across every data source, every day, within a known budget.

One view, and AI agent management on one platform

Work that was spread across many tools and data sources now runs on one platform, under one control. AI agent management happens in one place, and every action is tracked in Jira until it closes, so the team always knows what is open and who owns it.

Enterprise memory and institutional knowledge the company owns

Every decision is kept with its reason. The logic behind yield calls becomes institutional knowledge that belongs to the company rather than to individual analysts.

How a typical case moves through the team

Take one case: a fill rate that drops overnight.

1. What the agent did

Overnight, an agent notices that fill rate on a premium ad unit has dropped sharply after a floor price change. Using data from the ad server and BigQuery, it works out the likely cause and recommends lowering the floor for that unit.

2. Why it went to a person

Changing a floor price affects revenue and is outside the low risk actions the agent may take alone, so it posts the recommendation in Slack for the ad ops team, with the data behind it.

3. What the person did

An ad ops lead agrees with the cause but chooses a smaller cut, because a large advertiser's campaign starts that week. She approves the adjusted change in Slack.

4. How it was resolved

The agent applies the approved floor on the ad server, updates the Jira ticket and closes it once fill rate recovers.

5. What LeafCraft recorded

The drop, the agent's recommendation, the lead's change and her reason, the action on the ad server and the outcome are all recorded on LeafCraft. The next time a floor change hits fill rate during a big campaign, the agent recommends with that context.

From knowledge retention to decision intelligence

The reasoning is kept as enterprise memory, which is what makes knowledge retention work: when the lead moves on, her judgement stays with the company.

It also builds decision intelligence, because each new recommendation starts from the decisions already on record.

The result: less manual work and revenue recovered every day

Analysts now do about 50 to 60% less manual data work. The company recovers more than $12,000 a day in revenue, saves about $200,000 a year, and went from its first workshop to production in six weeks.

Figures are as reported by the customer.

How the LeafCraft platform helped

Each problem the company started with was answered by a part of the platform.

About LeafCraft

LeafCraft is the platform for the Hybrid Intelligent Enterprise, where AI agents and people work as one team, each with a job description, under one set of rules. Enterprises run the agents they've built or bought (from SaaS or AI agent vendors), or build new ones on LeafCraft. It works with the workflow and automation tools they already use, whether built in house, bought as software or custom developed.

Human-AI collaboration on LeafCraft rests on four pillars:

People and AI as one team. You own the rules, memory and budget.

If you are planning the move from AI tools to AI teammates, we would be glad to show you how it works. Book a conversation.

Frequently asked questions

What does Human-AI collaboration look like in ad operations?

Agents read ad and revenue data, spot yield problems and recommend what to do. The ad ops team approves those actions in Slack, and exceptions go to analysts, whose decisions go back to the agents.

How does AI governance work when agents act on an ad server?

Each agent works under clear rules, and every action it takes on the ad server is audited. Low risk actions are handed to the agents as trust in their record grows. Everything else waits for a person's approval.

What does an AI agent management platform do for an ad ops team?

It brings work that was spread across many tools and data sources onto one platform, under one control. AI agent management happens in one place, and every action is tracked in Jira until it closes.

How does AI cost control work for an AI DataOps team?

Every agent's cost is capped before it runs. The agents can work across every data source, every day, within a known budget.

How do the agents support institutional knowledge and knowledge retention?

Every decision is kept with its reason, so the logic behind yield calls belongs to the company. When an ad ops lead moves on, her judgement stays.

Where does decision intelligence come in?

Each new recommendation starts from the decisions already on record. The next time a floor change hits fill rate during a big campaign, the agent recommends with that context.

How long did it take to go live?

The company went from its first workshop to production in six weeks.

What results did the company report?

Analysts do about 50 to 60% less manual data work. The company recovers more than $12,000 a day in revenue and saves about $200,000 a year.

Is this kind of enterprise intelligence only for AdTech?

No. LeafCraft is a horizontal platform, so the same five pillars apply in any industry and any process.

What is LeafCraft?

LeafCraft is an AI agent management platform for the Hybrid Intelligent Enterprise. AI agents and people work as one team, each with a job description, under one set of rules, and the enterprise owns the rules, memory and budget.

Does LeafCraft work with the agents and tools an enterprise already has?

Yes. Enterprises run the agents they've built or bought, from SaaS or AI agent vendors, or build new ones on LeafCraft. It works with the workflow and automation tools they already use, whether built in house, bought as software or custom developed.

What happens to the decisions agents and people make on LeafCraft?

Every decision and its reason is recorded and stays with the enterprise. That record is the enterprise's institutional knowledge, and it is what makes knowledge retention and decision intelligence possible.