Scoring prospects without fooling yourself

Start with the last ten you won

Do not begin with a list of signals. Begin with the last ten customers you won and the last ten you lost after real effort. Write down what was true about each one before you spoke to them: size, industry, whether they were hiring, how they found you, what their reviews looked like. The signals that separate the two lists are your starting set. There are usually three or four, not fifteen.

Give points, not opinions

Each signal gets a number of points, and a prospect's score is the sum, capped at 100. Keep it that simple. Weighted formulas and decay curves feel rigorous, but nobody on the team can explain them, so nobody trusts the result.

A reasonable first pass for a company that sells to small businesses:

• Hiring right now: 30 points. Growing companies buy things.
• In your target industry: 25 points.
• Between 10 and 50 employees: 25 points. Big enough to have the problem, small enough to decide quickly.
• Google rating of 4.5 or higher: 20 points. A well-run business is easier to sell to.

Then set a threshold. Sixty is a sensible start: a prospect needs two or three of the signals to be worth a call.

Check the score against yourself

This is the step most teams skip. Take twenty prospects the score has rated and rate them yourself, blind, from 0 to 100. Then compare.

Where you agree, good. Where you are consistently higher than the score, a signal you care about is missing or under-weighted. Where the score is consistently higher than you, a signal is over-weighted; "hiring" is the usual culprit, because a company hiring three baristas is not a growing software buyer.

Adjust the points, rescore, compare again. Two rounds of this is usually enough. The aim is not a perfect model; it is a score the team believes, because they helped make it.

Watch the cost

If the signals come from outside data services, each lookup costs money. Set a budget per run and a cap on how many prospects you enrich, and look at what each run cost next to what it found. A score that costs more to compute than the leads are worth is not a score, it is a hobby.

Respect the no

Anyone who asks not to be contacted goes on a list and never appears in a run again. It is the right thing to do, and the list is also useful data: if a signal keeps surfacing people who say no, lower its points.