Most lead scoring articles describe a system with fourteen weighted variables, a decay function and a threshold that triggers an automated workflow. Then they include a screenshot of a dashboard from a company with a data team.
If you are a founder or a two-person sales team, you need something else: a model you can build before lunch, explain to a new hire in one minute, and act on without opening a spreadsheet.
Here it is.
The only question a score has to answer
Not "how likely is this lead to convert", expressed to two decimal places. Just: who do I contact first this morning?
That reframing kills most of the complexity. You are not building a prediction engine, you are building a sort order. A sort order needs to be roughly right and immediately usable, which is a far lower bar and a far more useful one.
Three dimensions, ten points each
Fit: does this business match who I sell to?
Score against the segment you defined. Concrete criteria only.
| Signal | Points |
|---|---|
| Exact industry match | 4 |
| Right size band (staff, locations, revenue proxy) | 3 |
| In your service area or timezone | 2 |
| Uses a technology or supplier you complement | 1 |
Fit is the easiest dimension and the least interesting, because everyone already does it. It stops you wasting time; it does not tell you who is ready.
Need: is there evidence they have the problem I solve?
This is the dimension that separates a working model from a demographic filter, and it is the one most teams skip because it takes actual looking.
The signals are specific to what you sell, but the shape is always the same: something publicly observable that implies the gap.
| Example signal | What it implies |
|---|---|
| No website, or a site that is visibly dated | They know they have a problem |
| No online booking or ordering | Direct gap if that is what you sell |
| Reviews complaining about a specific failure | Named, evidenced pain |
| Hiring for a role that overlaps with your service | Budget exists and is being spent |
| Recently opened, expanded or relocated | Buying window is open |
| Running ads | Marketing budget confirmed |
Score up to 10 by counting the signals that apply, weighted by how strongly each implies your particular solution.
The reason need matters so much is not really the scoring. It is that a need signal gives you the first line of your email. "I noticed you take bookings by phone only" is a sentence that could only have been written to that business, and specificity is what earns replies.
Reachability: can I actually get to the decision maker?
A perfect prospect you cannot contact is worth nothing.
| Signal | Points |
|---|---|
| Verified email for a named decision maker | 5 |
| Verified generic address at a small business | 3 |
| Verified business phone or messaging number | 3 |
| Prior interaction of any kind | 2 |
Note that reachability is the dimension most improved by tooling. Verification turns a maybe into a yes or no before you waste a send, and a validated messaging number opens a channel with far less competition than the inbox.
Adding it up
Total out of 30. Three bands:
| Band | Score | What you do |
|---|---|---|
| Hot | 22 to 30 | Contact this week, personally, with a message written for them |
| Warm | 14 to 21 | Enrol in a sequence with light personalisation |
| Cold | Below 14 | Leave it. Revisit if a new signal appears |
Resist the urge to add a fourth band. Three maps cleanly onto three behaviours, and behaviour is the point.
The rules that keep a model honest
Score on facts, not vibes. If a criterion cannot be checked by looking, it does not belong in the model. "Seems like they would be interested" is not a signal.
Do not score on engagement you manufactured. Opens and clicks feel like intent signals and are increasingly noise, thanks to privacy proxies that open everything. Replies are real. Opens are not.
Recency beats magnitude. A need signal from last month is worth more than a stronger one from two years ago. If you keep leads for a long time, decay the need score.
A low score is not a rejection. It is a "not now". The most valuable thing an old lead list can do is get rescored when a new signal appears, which is why signals should be refreshed rather than captured once.
Why most models die within a month
Three reasons, in order of frequency.
Nobody uses it. The score lives in a column somebody has to remember to sort by. If it does not change what appears at the top of the work queue automatically, it will be ignored by week three. The fix is structural, not motivational: the score has to drive the list order and the sequence enrolment, not sit beside it.
It was never calibrated. The weights were guessed once and never tested. The fix is a monthly review that takes twenty minutes: list the deals that closed, look up the score each had when you first contacted them, and see whether your winners were concentrated in the hot band. If they were scattered evenly across all three, your model is measuring something that does not matter.
It got too complicated. Someone added negative scoring, decay curves, sub-weights per industry, and now nobody can explain why a given lead sits at 17. A model nobody can explain is a model nobody trusts, and a model nobody trusts gets overridden by gut feel, which is where you started.
The monthly recalibration, concretely
- Pull every deal closed in the last month.
- Note the score each lead had at first contact.
- Note the score of twenty leads that went nowhere.
- Ask which single criterion best separates the two groups.
- Increase that criterion's weight. Decrease anything that shows no separation.
- Delete any criterion that has never once separated a winner from a loser.
Six steps, twenty minutes, once a month. That is the entire maintenance burden, and it is what turns a guess into a model.
What good looks like after three months
You open your workspace, filter to the hot band, and there are between ten and thirty businesses in it. Each one has a visible reason for being there that you could read aloud. You write those personally. Everything in the warm band is running through an automated sequence in the background. The cold band is quietly waiting for a signal to change.
That is the whole system. It fits on an index card, it takes an hour to build, and it will outperform a fourteen-variable model that nobody looks at.