Turn AI experiments into revenue systems

TL;DR

• Start where you can see the problem, not where AI looks shiny. Tie automation to measurable bottlenecks like funnel leaks or support lag.

• Run AI pilots like experiments. Define success metrics, own the data loop, and document learnings before scaling.

• Treat data as infrastructure. Fix naming, lineage, and governance first — messy data kills speed and trust.

• Train your team to manage exceptions. Automation fails when people don’t know what to do when it’s wrong.

• Connect AI to business metrics early. Track conversion, cost-per-signup, or engagement lift — not model accuracy.

Identify where automation drives revenue, not efficiency

Anuradha’s operating philosophy is simple: start where automation can change the P&L, not just the process.

When she introduced data-informed automation at Expensify, the first target wasn’t “AI strategy” — it was a high-cost funnel leak: onboarding drop-offs. By mapping every customer touchpoint, her team found that qualified leads were engaging but not completing setup.

“We didn’t start with a model. We started with a measurable business problem. That’s what made the data useful.”

They used lightweight automation to trigger follow-ups from onboarding specialists when certain behaviors — like partial form completion — occurred. Within two quarters, qualified lead engagement doubled and signup costs dropped from triple digits to single digits.

Lesson: AI needs a revenue hypothesis, not a tech objective.

Run AI pilots like scientific experiments

Expensify doesn’t “launch” AI features — they run controlled experiments that follow a three-step structure:

1. Define success clearly. Choose one metric (for example, support resolution time or conversion rate).

2. Control the variables. Run a small pilot with limited scope and fixed data input.

3. Close the loop. Document every edge case and update the process before expanding.

   
   

“Most AI projects fail because no one owns what happens when it’s wrong.”

Every pilot has a single DRI — the person responsible for learning, not just launch. That owner writes the playbook for scaling once success is proven.

This turns experimentation into a habit, not a headline.

Build the data foundation before adding intelligence

AI magnifies whatever data you feed it — good or bad. So Anuradha treats data hygiene as the first milestone in any automation effort.

Her framework for making data usable:
• Ownership: assign a data steward for every key metric.
• Structure: document what every field means and how it’s updated.
• Access: ensure teams can pull data without manual exports.

“You can’t automate chaos. If you skip the cleanup, you’ll spend six months fixing what the AI broke.”

When Expensify unified naming conventions and cleaned pipeline data, marketing suddenly saw which campaigns drove paid conversions — insights that had been invisible for years.

Design for human-in-the-loop workflows

AI isn’t about replacing people. It’s about moving humans to higher-value judgment work.

At Expensify, operations teams run what Anuradha calls “exception playbooks”: predefined checklists for what to do when automation fails or looks suspicious.

Examples:
• When fraud detection flags an outlier, a specialist follows a 3-step escalation flow.
• When onboarding AI misclassifies a lead, the rep can correct and retrain it instantly.

“If humans don’t know what to do when the machine is wrong, adoption stops.”

The rule: automate confidence, not control. Keep humans in the loop where trust is earned, not assumed.

Measure business impact, not model accuracy

The first AI projects at Expensify weren’t evaluated on how accurate the algorithms were — but on whether they moved real metrics.

Each automation tracked at least one of the following:
• Conversion rate improvement (e.g., from lead to paid user)
• Cost-per-signup reduction
• Time-to-resolution in support workflows

“Accuracy doesn’t matter if it doesn’t show up in your metrics dashboard.”

This discipline let her teams scale automation safely, because every AI system had a clear ROI story before it touched production.

Make AI part of your GTM system, not your side project

AI’s value compounds when it connects to how you go to market. At Expensify and Breakout, GTM teams use AI-generated insights to refine messaging, segment leads, and prioritize channels dynamically.

Tactically, that means:
• Automating qualification based on behavior patterns.
• Using predictive analytics to time follow-ups.
• Feeding back real conversion data into creative briefs.

The payoff is alignment. Marketing doesn’t just generate leads — it feeds the intelligence that makes every next touch smarter.

The operating principles for AI-led growth

Anuradha’s approach to scaling intelligent systems can be summarized in five operating principles:

1. Solve for clarity first. Know what “good” looks like before you automate it.

2. Instrument everything. You can’t improve what you can’t measure.

3. Close the feedback loop. Every automation output should teach you something.

4. Keep humans visible. Ownership builds trust and accountability.

5. Translate results into learning. Document every success and failure publicly.

This is how Expensify turned experimentation into a scalable operating model — not a lab project.

Final word

AI becomes a revenue engine only when it’s wired into how your business learns. Start small, tie every experiment to a real metric, and build feedback into every layer — from the data to the humans who run it.

The winning loop isn’t build → deploy → repeat. It’s observe → learn → operationalize → scale.

About the COO

Anuradha Muralidharan is the Chief Operating Officer at Expensify.

At Expensify, she has scaled data-informed operations, rebuilt onboarding and payments systems across 200+ countries, and driven measurable revenue growth through automation. Her background spans Citi, Marqeta, and Oracle, giving her a unique blend of product, finance, and GTM expertise.

She is known for operationalizing intelligence — building systems that learn faster than competitors.

About the series

This is from Breakout Sessions, where marketing leaders unpack how AI is changing GTM and the way buyers buy.