AI for RevOps: Where It Helps and Where It Doesn't Yet

A RevOps person's job is stitching together answers that live in four different systems: the CRM has the deal, the billing system has the invoice, the product has the usage, and the spreadsheet has whatever reconciliation someone did last time this exact question came up. AI for RevOps, done honestly, is mostly about closing that stitching gap. It is not, yet, about replacing the judgment calls that make RevOps a real job rather than a report-generation function.

There's almost nothing written specifically on this question. Search for it and you mostly get generic "AI for revenue operations" listicles that could apply to any go-to-market function, not anything grounded in what a RevOps team actually spends its week doing. That's the gap this piece is trying to close.

Where it helps today

Cross-system querying is the clearest win. A RevOps person asking "which accounts expanded usage but haven't been upsold" today means pulling a CRM export, a usage export, and reconciling them by hand, or waiting for someone in data to run a query. An AI layer that can query billing and CRM data together in natural language collapses that into one question, answered directly.

Lago’s operations solution is built around exactly this kind of cross-system billing and usage visibility.

Anomaly surfacing is close behind. A renewal at risk, a usage pattern that doesn't match the contracted tier, a segment where churn is creeping up before it shows up in the monthly numbers. This works well specifically because RevOps sits on structured, well-modeled data (accounts, subscriptions, usage events), which is a much better substrate for this than messier qualitative work.

Drafting routine outputs, QBR prep, renewal risk summaries, segment performance write-ups, is mature and low-risk, the same way it is in general finance use cases. A human still reviews before anything goes to a customer or a leadership deck.

Where it doesn't work yet

Judgment calls about account strategy. Whether to push a renewal conversation now or wait, whether an at-risk signal reflects real churn risk or a temporary blip, whether a discount is worth offering, these depend on context (relationship history, what happened on the last call, competitive pressure) that isn't fully captured in any system, and AI tools confidently answering these questions anyway is the single biggest way a RevOps AI pilot goes wrong.

Cross-system data quality problems don't get fixed by adding an AI layer on top. If the CRM's stage definitions and the billing system's plan names have drifted apart, an AI assistant answering questions across both will just surface the mismatch faster, and sometimes will paper over it with a plausible-sounding answer instead of flagging the inconsistency. This is worth explicitly testing before trusting an output.

The dependency RevOps has on billing specifically

A meaningful share of what RevOps struggles to answer quickly isn't a CRM problem, it's a billing-data problem: usage that isn't cleanly connected to the account record, invoicing history that lives in a separate export nobody's kept current, revenue recognition that doesn't map cleanly to what the CRM calls a "closed-won" deal. RevOps for SaaS covers the function-building side of this. The AI layer on top only works as well as the billing data underneath it, which is the same grounding argument that shows up in Generative AI in Finance: a fast, wrong answer is worse than a slow, right one, and RevOps is exactly the kind of role where a wrong answer gets acted on before anyone double-checks it.

Key takeaways

AI for RevOps works well today for cross-system querying, anomaly surfacing, and drafting routine outputs, all cases where a human still reviews the result. It doesn't yet work for the judgment calls that make up the harder half of the job, and it doesn't fix underlying data-quality problems between CRM and billing systems, it just answers questions about them faster. The quality ceiling for any RevOps AI tool is set by the billing data underneath it, not by the AI layer itself.