HubSpot AI: Ask Your CRM Questions in Plain English

What it takes for a plain-English question about your HubSpot data to return a number you can put in front of a board, and the handful of questions that are worth asking first.

By Ed · Published October 25, 2026 · 7 min read

Twenty-five years starting companies, consulting, owning a multi-million-dollar P&L and leading business intelligence for large enterprises.

You can ask HubSpot questions in plain English, but the question is the easy part. The hard part is that "win rate" or "pipeline" has to mean the same thing every time someone asks, and you have to be able to see how a number was built. This post covers the questions worth asking first, what an answer needs underneath it, and how Quiriz's HubSpot connector handles it.

The problem is not the typing

Most CRM reporting pain is not that people cannot find the report builder. It is that three people ask for "pipeline" and get three numbers. One includes deals marked closed lost that nobody cleaned up. One uses create date and another uses close date. One multiplies by stage probability and one does not. Putting a chat box on top of that does not fix it. It produces the same three numbers faster.

So the useful test of any AI layer on a CRM is not how natural the question sounds. It is whether the same question returns the same figure on Monday and on Friday, and whether you can inspect the filters when someone challenges it.

Write the definitions before the questions

Before connecting anything, write one line for each of the numbers your leadership actually discusses. A reasonable starting set:

Open pipeline: sum of deal amount where the deal is not closed. Say which date field you bucket by.
Win rate: closed-won deals divided by all closed deals, by close date. Open deals are not in the denominator.
Average deal size: mean amount of closed-won deals.
Sales cycle: mean days from create to close for closed-won deals.
Closed-won revenue: sum of closed-won amount, by close date.

One detail matters more than it looks: use HubSpot's own closed and closed-won flags rather than a stage name. Custom pipelines call their stages whatever the team likes, so a rule like "stage equals Closed Won" quietly misses a pipeline where the winning stage is called "Signed". If you want the long version of the win-rate argument, read Why Your Win Rate Keeps Changing.

The questions worth asking first

Once definitions exist, these are the questions that tend to pay for the effort, roughly in the order to try them:

What is our win rate by deal owner this quarter? Quick check on whether owner assignment is clean.
How much pipeline was created each month this year? Shows whether the top of the funnel is growing before it shows up in revenue.
Which open deals are past their close date? The cheapest hygiene report there is.
Why are we losing deals this quarter? Grouped by the closed lost reason field, if your reps fill it in.
How long do deals sit in each stage? Needs stage history, covered below.
How has open pipeline changed since last Monday? Needs daily snapshots, because HubSpot only stores the current state of a deal.

If you build dashboards from these, HubSpot Dashboard Examples shows which ones are worth keeping on a screen.

How Quiriz's HubSpot connector answers them

The connector is read-only and refreshes daily. It brings HubSpot in as separate datasets: [HubSpot] Deals, [HubSpot] Contacts, [HubSpot] Companies, [HubSpot] Deal stage history and [HubSpot] Pipeline snapshots. Custom properties you choose can be added as extra columns on deals, contacts and companies.

On top of those, an admin defines metrics once in Company Context. The connector ships with suggested definitions for Pipeline value, Win rate, Average deal size, Sales cycle (days), Closed-won revenue, Weighted pipeline, Closed-won ARR, Lost deals, Average days in stage and Deals entering stage. Win rate, for example, is the average of the closed-won flag over closed deals by close date, so it is a decimal where 0.25 means 25 percent, and it never includes open deals.

You then ask in ordinary language: "What's our win rate by deal owner this quarter?" or "How long do deals sit in each stage?" The answer comes back with the query and filters shown, so you can confirm it used the close date and the closed flag. Two things this setup buys you: the same question returns the same figure for everyone, and a challenged number can be traced to a filter rather than argued about.

Stage time and conversion need stage history

Average days in stage and stage-to-stage conversion cannot be computed from the Deals table alone, because a deal row only knows its current stage. The Deal stage history dataset has one row per deal per stage with the date it entered and exited. Deals entering a stage, divided into deals entering the next stage, is the stage conversion rate. This is the part of CRM analysis most people never get to by hand because the raw property exports are painful, so it is a good reason to use a connector rather than a CSV.

Honest limits

Garbage in stays garbage. If amounts are blank, owners are unassigned or lost reasons are empty, the answer will say so or show a gap. It cannot invent the missing data.
It can still be wrong. Governed definitions make answers consistent and showing the work makes them checkable. That reduces errors. It does not make an AI infallible, so verify any figure that is going into a board deck.
Daily refresh. The data is as of the last daily refresh, not live to the minute.
Pipeline history starts when you connect. Daily snapshots begin accumulating from connection. HubSpot does not hand back a historical pipeline, so there is no retroactive "pipeline as of last March".
Weighted pipeline is a stage-probability calculation, not a forecast. See Weighted Pipeline Is Not a Forecast.

For the broader picture of AI on business data, what an AI data analyst is and is not is a useful companion, and the Quiriz page for CRM teams lists the workflows built around this.

Ask HubSpot a question and see the work

Connect HubSpot read-only and ask about win rate, pipeline or stage time in plain English. Free to start.

Try Quiriz free →

Frequently asked questions

Can you ask HubSpot questions in plain English?

Yes. HubSpot has its own AI features inside the product, and you can also connect HubSpot to a separate analysis layer like Quiriz and ask questions such as which open deals are past their close date or how long deals sit in each stage. The useful difference is not the typing, it is whether the metric definitions behind the answer are fixed and visible.

How do I get HubSpot data into an AI tool?

The usual routes are a CSV export, which is manual and goes stale, or a read-only connection that refreshes on a schedule. Quiriz connects to HubSpot read-only and refreshes daily, bringing in deals, contacts, companies, deal stage history and pipeline snapshots as separate datasets.

Is it safe to connect HubSpot to an AI analyst?

Check three things: that the connection is read-only, that you can see what data is imported, and that the tool shows how it computed each answer. The Quiriz connector reads and never writes. Beyond that, apply the same review you would give any vendor that touches CRM data.

Why does the AI give a different win rate than my HubSpot report?

Almost always a definition difference: open deals in the denominator, a different date field, or a stage name treated as closed-won instead of the closed-won flag. Compare the filters, not the arithmetic. See why your win rate keeps changing for the three common formulas.

What HubSpot questions should I ask first?

Start with win rate by owner, pipeline created per month, open deals past their close date, and why deals are lost. They use fields most portals already fill in, and each one shows quickly whether your stage and owner hygiene is good enough to trust the rest.