There's a moment in every SaaS team's life when "the dashboard says X" and "the AI assistant says Y" disagree. That's the moment you find out whether your AI analyst actually knows your business — or whether it's pattern-matching on training data.
Before you trust an AI revenue analyst with your board pack, it should answer five plain-English questions on day one. Not "let me think about it." Not "I'll need to pull more data." Answers, with numbers, in seconds.
Here's the checklist.
1. What's our exact MRR, ARR, and NRR right now?
The most basic test, and the one most analytics tools fumble. MRR isn't a single number — it's a chain of definitions. Is it annualized ARR ÷ 12, or last month's recurring? Are you including expansion before or after the period boundary? What about contraction MRR from downgrades?
An AI revenue analyst should:
- Return MRR/ARR with explicit definitions ("annualized recurring revenue from active subscriptions as of [date]")
- Distinguish gross vs net MRR, with all four components broken out — new, expansion, contraction, churn
- Compute NRR for any period without you specifying the formula — and explain its assumptions when asked
If your AI tool can't tell you NRR for "last quarter, enterprise segment only" in one prompt, you're going to spend the rest of the year fighting it. Bonus: a good answer to "what is our MRR?" includes the date the answer is good for and which source systems were last synced.
2. Where is growth coming from?
Growth has four ingredients: new logos, expansion, pricing, and churn (which subtracts). An AI analyst that returns "MRR grew 12% MoM" without decomposition is worse than no analyst — it's giving you a metric without the story.
The answer should look like:
Revenue grew 12.4% MoM. Drivers: $42k new logos (60% of growth), $28k expansion from existing customers (40%), partially offset by $12k churn. Top contributor: enterprise segment, +$31k from the Stark Industries upsell.
Notice the structure: percentage + dollar amounts + attribution + a specific account. That's a sentence you can paste into a board update. "MRR grew 12%" is a sentence you have to defend.
For RevOps teams, a derivative of this question is "where is growth coming from by channel / segment / plan?" — same decomposition, different cut. A capable AI analyst handles both without re-explaining the schema.
3. Which customers are at risk this quarter?
This is where most BI dashboards go silent. They'll show you a churn rate; they won't tell you which 8 customers are about to churn.
An AI revenue analyst should fuse three signals to answer this:
- Behavioral: usage drop, login frequency change, feature adoption decline
- Commercial: NPS slip, support ticket spike, plan downgrade, late payment
- Lifecycle: renewal within 90 days, customer health score change
And it should produce a list, not a percentage:
8 customers at risk this week. Total exposure: ~$42k MRR. Top three: Acme ($12k, usage down 60% MoM), Globex ($9k, NPS dropped 2 points), Initech ($7k, renewal in 31 days + open support escalation).
That's actionable. "Churn rate is 2.1%" isn't.
4. How concentrated is our revenue?
Customer concentration is the question your board will ask in the next funding round. Investors run quick math: if your top 10 customers represent 38% of revenue, the loss of any one of them is 3–4% of ARR — material.
An AI revenue analyst should answer "top 10 customers as % of revenue" without you teaching it the formula. It should also surface:
- Largest customer (% of MRR)
- Top 5, top 10, top 20 — concentration at each tier
- The Pareto cutoff: how many customers make up 80% of revenue?
A healthy SaaS distribution has the top 10 customers under 25% of revenue. If your top 10 are over 50%, you're a few churn events away from a catastrophic quarter. Use our Quick Ratio calculator to stress-test the math on your own numbers.
This question is also where a good AI analyst offers follow-ups: "want to see the growth trajectory of just these top 10?" or "what's the renewal status of these accounts?"
5. What does the next quarter actually look like?
The hardest one. Forecasts are notoriously fragile because they're rarely revisited until the next quarter ends and the board says "you said $2.5M; you did $2.1M."
A good AI revenue analyst gives you a forecast with:
- The base number ("$2.7M next quarter")
- The components (existing recurring + booked renewals + pipeline at expected conversion − expected churn)
- The confidence interval ("80% probability between $2.5M and $2.9M")
- The sensitivity ("if churn holds at current rate, +$80k; if Stark Industries doesn't renew, −$340k")
And critically, it should remember its prediction. When the quarter ends, the analyst should automatically reconcile and tell you which input was off — was pipeline conversion below plan? Was churn higher than expected? Was expansion lower than modeled?
That's the difference between a forecasting tool and a forecasting partner. The tool gives you a number. The partner closes the loop and gets better next time.
The day-one bar
If your AI revenue analyst can answer those five questions — with numbers, in plain English, in under a minute each — you have a real analyst. If it can't, you have a dashboard with a chatbot grafted on.
Day one shouldn't be "let's onboard the data first." Day one should be: connect Stripe and HubSpot, ask the five questions, get five clean answers. Anything else is a demo, not a product.
Want to try the questions yourself? atSpark unifies your billing, CRM, and subscription data in 14 minutes — and then it's your turn to interrogate it. Get early access →