How Do You Score Leads When You Have Under Fifty Customers?

How do you score leads when you only have a few dozen customers?

You mostly don't, at least not with weighted points. With under fifty customers, a points model encodes guesses as math. I use three simpler layers instead: hard disqualifiers, a three-level fit grade built from won deals you know by name, and engagement used only as a timing trigger.

Most lead scoring advice is written for companies with thousands of contacts and hundreds of closed deals. At that scale, patterns in the data are real, and a model can learn from them. At thirty or forty customers, the "pattern" is usually three deals that happened to close in the same quarter. Put weights on that and you get a confident number that means very little.

I work on go-to-market systems for B2B SaaS teams, and the early-stage version of this problem comes up constantly. The founder wants a score because the CRM has a field for it. The real need is simpler: tell the one or two people selling which leads to call today, and which to ignore.

Why does traditional lead scoring break at small numbers?

Traditional scoring assumes enough closed deals to tell a real signal from noise. With a small customer base, one unusual deal can swing a weight, job titles repeat by accident, and engagement points reward people who click a lot rather than people who buy. The score looks precise but rests on guesses.

Think about what a weight claims. Giving "VP title" 15 points and "visited pricing page" 10 points says one is worth exactly one and a half times the other. Nobody at an early-stage company can defend that ratio. It came from a blog template or a gut feeling, and the decimal places make it look like research.

There is also a feedback problem. With few deals, you cannot easily check whether the score worked. If thirty leads scored above 70 last quarter and four became customers, is that good? You have no baseline. So the score never gets corrected, and the sales team learns to ignore it. Once reps ignore a score, it is worse than no score, because it still drives automations nobody is watching.

What should replace points in an early-stage scoring model?

Replace points with three plain layers. First, disqualifiers that remove leads you would never sell to. Second, a fit grade of A, B, or C based on your real customers. Third, an engagement trigger that flags timing, not quality. Each layer is a rule a rep can read and argue with.

I call this disqualify, grade, trigger. Disqualifiers are binary: wrong country for your billing setup, student email, competitor domain, company size you cannot serve. A lead that hits one is out, no matter how engaged. This alone removes a surprising share of noise from a small team's queue.

The fit grade is a judgment written down. A means "looks like the customers you would love more of." B means "could work, needs a conversation." C means "technically possible, not a priority." The grade comes from firmographics you can see: industry, team size, the tool stack, and the job the buyer is trying to do.

The trigger layer watches for recent intent, like a demo request, a pricing page visit, or a reply to outbound. It does not change the grade. It changes when someone acts. An A-grade account with no recent activity goes into a nurture or an outbound list. An A-grade account that just hit the pricing page gets a call today.

How do you build a fit grade from a small customer list?

Pull your won deals into one view and review each one by hand. Mark the five or ten you would clone if you could. Write down what they share in plain words. Those shared traits become your A criteria. Lost deals and churned accounts often tell you what belongs in C.

This is where small numbers become an advantage. With forty customers, you can know every one of them. You remember why each one bought, who championed it, and which ones were painful. A model trained on that data would miss the context you already have in your head. So write the context down instead.

I like to keep the A definition to three or four traits. For example: a B2B software company, a sales team of a certain size, a CRM already in place, and a specific trigger such as a recent funding round or a new revenue leader. If your best customers share fewer traits, that is useful too. It tells you your ICP is still forming, and the grade should stay loose.

If you have not done the ICP work yet, start there. I wrote a separate guide on defining an ICP when you have twenty customers. The fit grade is the ICP turned into a field your CRM can filter on.

Where does engagement data fit when you have few customers?

Engagement should answer "when," not "how good." Email opens and page views are cheap actions that many non-buyers take. Treat engagement as a short-lived timing flag with an expiry date, so a lead who read your blog three months ago does not stay hot forever. Recent, high-intent actions matter most.

HubSpot's own documentation draws the same line between the two ideas. It describes fit scores as qualifying records by demographic information through property values, and engagement scores as qualifying records by their actions and interactions, such as visiting a website or opening a marketing email. It also supports combined scores that grade fit with a letter and engagement with a number, from A1 to C3.

That letter and number grid is a good mental model even if you never build a points score. A1 is high fit and high engagement. C3 is low on both. With a small customer base, I would keep the letter as a manual judgment and let the number be the only automated part.

The same documentation describes score decay, which reduces an event's score based on how long ago it happened. Whatever tool you use, build in some kind of expiry. A demo request from last week means something. One from last spring usually does not.

Should a small team use AI lead scoring?

Not yet, for most teams under fifty customers. AI scoring learns from past conversions, and a small set of conversions gives it little to learn from. It may also sit on a plan tier you do not need yet. Revisit it once you have enough won and lost deals that a pattern could actually exist.

HubSpot, for example, says you can create contact engagement and fit scores with AI, and that the model trains on contacts whose lifecycle stage changed within a timeframe you choose. Its knowledge base lists Marketing Hub Enterprise for that feature. Plan access changes, so check HubSpot's current documentation before you budget around it.

The deeper issue is not price. It is that any model, AI or manual, inherits the quality of its inputs. If lifecycle stages are updated inconsistently, which is common in young CRMs, the model learns your data hygiene, not your buyers. Clean stages first. Score second.

How do you connect the score to routing and outbound?

Make the grade drive one decision each. A-grade leads with a fresh trigger go straight to a person. A-grade leads without a trigger go into targeted outbound. B-grade leads get a nurture sequence and a quarterly review. C-grade leads stay in marketing. Disqualified leads never reach a rep.

This is where the system earns its keep. A grade nobody acts on is decoration. Tie each grade to a specific owner and a specific next step in your CRM. If you route inbound leads in HubSpot, my notes on routing inbound leads to the right owner in HubSpot show how to hang that logic on a single property.

For outbound, the grade works as a filter in tools like Clay or Apollo. Build your list from accounts that match the A definition, enrich them, and only then look for signals. You will send fewer emails, and the ones you send will be to companies that resemble buyers you already have.

How do you know when to switch to a real scoring model?

Switch when you can test the model against outcomes. That means enough won and lost deals per quarter that a difference in conversion between grades is visible, consistent lifecycle stages in the CRM, and a rep team large enough that manual triage no longer scales. Until then, rules beat weights.

A simple test tells you when you are close. At the end of each quarter, compare how A, B, and C leads converted to opportunities. If A leads clearly beat B and C, your grade is working, and you may be ready to add weighted detail. If the grades convert about the same, your definition is wrong, and adding points would only hide that.

I also watch the sales team's behavior. When reps start asking for finer distinctions inside the A group, that is a sign the volume has grown enough to justify more structure. Let the need pull the model forward. Do not push a model onto a team that is still learning who its buyers are.

What should you do next?

Pull your won deals into one list this week, pick the ones you would clone, and write a three-trait A definition. Add four or five hard disqualifiers. Then set engagement to flag timing with an expiry. Review the results each quarter before you add a single point.

If you want help turning that into a working system across your CRM, enrichment, and outbound tools, this is the kind of GTM engineering I do every week. Reach out at pravinkumar.co and let's chat about where your leads are getting lost.