How to Actually Build Company Scoring Criteria (With AI's Help, Not AI's Guesswork)

Building scoring criteria isn't about picking point values that feel right. It's a three-part process: find the patterns in your existing data, confirm which of those patterns actually matter with the people closest to the sale, then translate that into rules. AI is genuinely useful for the first part. It can't do the second part for you, and that's by design, not a limitation.

If Post 1 was about recognizing you already have the data, this post is about turning that data into criteria you can actually build on.

Start With What Won, Not What's Possible

The instinct when building an ideal customer profile is to describe who you wish you sold to. The right starting point is who you've actually sold to, and won.

Pull your closed-won deals. Look at the companies behind them: industry, size, source, how long the deal took to close. Then pull your closed-lost deals and losses that stalled entirely. The contrast between those two lists is where real criteria comes from, not from a hypothetical "dream client" description built in a strategy meeting.

This matters because criteria built from imagination tends to describe the customer you'd like to have next quarter. Criteria built from history describes the customer you actually know how to close. Start there.

Where AI Actually Helps in This Process

This is where AI earns its place in the process, and it's a narrower job than the hype suggests. AI is good at scanning a large set of closed-won and closed-lost records and surfacing patterns faster than a person manually cross-referencing spreadsheets. It can point out things like:

• Which firmographic traits — industry, company size, employee count — show up disproportionately among your best customers
• Which engagement behaviors tend to appear before a deal closes, versus behaviors that show up on deals that stall
• Combinations of traits that wouldn't be obvious scanning the data by eye, like a specific industry-and-size pairing that closes faster than either factor alone would suggest

That's real, useful work. It turns a database of hundreds or thousands of records into a shortlist of patterns worth paying attention to. What it doesn't do is tell you which of those patterns are actually meaningful for your business, versus which are coincidence. That judgment call belongs to a person.

Where AI Runs Out of Judgment

Every dataset has exceptions the data alone can't explain. The deal that closed despite "bad fit" data, because of an existing relationship. The segment sales or marketing leadership already knows to avoid, regardless of what the pattern shows, because of a past experience that never made it into a CRM field. The industry that looks promising in the data but that your team knows, from experience, tends to churn within six months.

This isn't a flaw in the process. It's the expected, normal boundary of what pattern-finding can do. AI can tell you what happened. It can't tell you why, and it doesn't know what you know from actually being in the room for these deals.

This is why the second half of building criteria isn't optional: taking the patterns AI surfaces back to the leadership closest to those customer relationships — sales, marketing, or both, depending on how your business is set up — and asking a simple question about each one: does this hold up, or is there context here the data doesn't capture? Some patterns will be confirmed outright. Some will need an exception built in. That conversation is what keeps the resulting criteria something your team actually trusts, instead of a model they quietly ignore because it doesn't match what they know to be true.

Turning Patterns Into Actual Criteria

Once a pattern is confirmed, it needs to become something usable: a weighted fit or engagement criterion.

In practice, this means assigning relative value to each confirmed factor. A strong industry match might carry more weight than company size. A high-intent engagement behavior, like requesting a demo, should carry more weight than a lower-intent one, like opening a marketing email. The exact point values matter less than getting the relative order right — the criteria that matter most to a real "yes" should outweigh the ones that are just nice to have.

Keep this step simpler than it feels tempting to make it. A scoring model with twenty finely-tuned variables looks impressive and gets ignored, because no one on the team can explain why a given company scored the way it did. A model built on the five or six factors that actually separate your best customers from everyone else gets used, because the team using it can look at it and understand the logic immediately.

A Simple Way to Sanity-Check Your Criteria

Before this criteria goes anywhere near a live HubSpot build, run it through a low-stakes gut-check first.

Pull a handful of companies you and your team — sales, marketing, or both — already agree were great fits, and a handful you already agree were poor fits. Run your draft criteria against both groups by hand. If the great-fit companies score high and the poor-fit companies score low, the criteria is doing its job. If a known poor fit scores unexpectedly high, or a known great fit scores low, that's a sign a weight needs adjusting or a pattern needs revisiting before you build anything.

This step takes an afternoon, not a data science background, and it catches the kind of mismatch that's expensive to find later, after the model is live and the team has stopped trusting it.

Once criteria passes this check, it's ready to move from a working document into HubSpot itself — building the actual fit and engagement scoring logic, connecting it to workflows, and making sure it's something your team uses instead of something that lives quietly in a spreadsheet. That's exactly what the next post in this series walks through.

FAQ

Do I need a data scientist to build lead scoring criteria? No. AI tools can help surface patterns in your existing CRM data, but the process of confirming and weighting those patterns is a business judgment step, not a technical one. It requires input from sales and/or marketing leadership, not a data science background.

What's the difference between AI scoring and rules-based scoring in HubSpot? Rules-based scoring uses criteria you define manually — specific traits and behaviors assigned specific point values. AI-assisted predictive scoring, available on HubSpot Enterprise, analyzes historical data automatically to identify high-converting patterns. Both can work well; rules-based scoring is available on Professional and gives you more direct control over the logic.

How do I know if my scoring criteria is actually accurate? Test it before building it into HubSpot. Run your draft criteria against a small set of known good-fit and known poor-fit companies by hand. If it sorts them correctly, the criteria is likely sound. If it doesn't, adjust the weighting or revisit the underlying pattern before building further.

Should sales leadership be involved in building scoring criteria? Yes, and marketing leadership too, depending on your business. Whichever team is closest to the customer relationship holds context that data alone can't capture — relationship history, known exceptions, and lessons learned that never made it into a CRM field. Criteria confirmed by the people who'll actually use it is criteria that gets trusted and used.