Lead scoring for a small business: a score that sorts a queue.

The score exists to answer one question: who gets a human today.

Most lead scoring advice was written for businesses with a queue and a team to work it. Copy the apparatus without the queue and you get scoring theatre — a number on a dashboard nobody opens, feeding a routing decision nobody makes.

A small business needs routing far more than it needs scoring. Routing is the act of getting the right person in front of a human while they still care. Scoring is one mechanism for deciding the order, and it only becomes worth building on the day the number of people showing intent exceeds the number of conversations you can hold. Before that day, the score is a project that feels like progress.

Three questions before you build anything

1. 01Is there more intent than capacity? More people show a buying signal in a week than you can have conversations with. If you can currently reply to everyone who does something interesting, a score adds a number to a decision you are already making correctly by reading.
2. 02Is there a specific human action on the other side? A personal message, an offer of a call, a task with a name on it. “Mark as qualified” is not an action; it is a label, and labels do not produce revenue. If nothing happens when the score crosses the line, the score is decoration.
3. 03Will somebody actually do it, this week? The most common failure is not a badly weighted model. It is a well-built score feeding a queue nobody works, which is worse than no score at all because it looks like coverage.

If any answer is no, build visibility instead. A saved view of everyone who did something meaningful in the last seven days, checked each morning with a coffee, is most of the value of a scoring model for a fraction of the work — and it is also the thing you will keep doing while you are busy, which the model is not.

Score what people did, not what they told a form.

Enterprise models split into demographic and behavioural halves. For a small business selling to individuals, the demographic half is mostly self-reported, rarely updated, and restates a decision you already made when you chose who to sell to. Drop it and score behaviour.

1. 01You can observe it reliably. Not “we think this fires”. Something you can check on your own record in ten seconds. A signal you cannot verify is a signal you cannot debug when the queue fills with the wrong people.
2. 02It happens often enough to separate people. A behaviour that fires for two percent of your audience cannot sort a queue — it just produces a very short list. Sorting requires signals that a reasonable number of people trip.
3. 03It is not a proxy for how long they have been around. Anything cumulative — total opens, total sessions attended, total pages seen — quietly measures tenure. Your most loyal reader is not your most likely buyer, and a score that keeps confusing the two loses everyone’s trust within a month.
4. 04You would be comfortable saying it out loud. “You looked at the pricing page twice this week” is a normal thing to know. If a signal would feel strange to mention on a call, it will feel strange in the follow-up it triggers.

Five to seven is the working range, and the reason is maintenance rather than accuracy. Every signal is a definition somebody has to remember, a piece of tracking that can silently break, and a line in the explanation you give when a colleague asks why a contact is at forty. A model with fifteen inputs is not more accurate at this size; it is merely harder to argue with, which is not the same thing.

Each of those signals also has to come from somewhere you can query. That is the constraint that decides your list in practice — the same one that governs behavioural triggers, and for the same reason: a score can only count what one record can see.

Weight by how close the behaviour sits to a purchase.

Not by how much effort it took the person, and certainly not by how much effort it took you to set up the tracking. Booking a call takes ninety seconds and outranks sitting through a sixty-minute session, because one of them is a request for a conversation and the other is an afternoon that was free.

Those numbers are illustrative arithmetic, not findings — a starting set with editable assumptions. The ratios are the part worth keeping: the top of the list should be roughly five times the bottom, so that one genuine buying signal outweighs a week of mild interest. If a person can reach your routing threshold purely by clicking things, the spread is too flat.

Three rules that keep the arithmetic honest

• →Keep the scale small and whole. Integers, nothing above about twenty-five, no decimals. Precision you cannot justify is precision that will mislead you later, and a score of 63.5 implies a rigour that a seven-row table does not have.
• →Cap the repeatable signals. Page visits, link clicks and session attendance can all be repeated. Cap each at two or three occurrences, or your highest-scoring contact will be whoever left a tab open.
• →Calibrate against a plausible week. Add up what one genuinely interested person could do in seven days. That total should land near your routing threshold. If a realistic enthusiastic week comes to half the threshold, nobody will ever cross it.

Without decay you have built a loyalty metric by accident.

A score that only ever goes up measures tenure. Give it a year and your top of queue is the person who has been reading you since the beginning and has never once tried to buy anything — ranked above somebody who opened your checkout this morning.

1. 01A rolling window — count only recent behaviour. Score the last sixty or ninety days and let everything older fall out. Self-cleaning, requires no scheduled job, and the score always describes the person as they are now rather than as they once were. This is the one to pick.
2. 02Periodic subtraction — lose points over time. Deduct a fixed number of points per quiet week. It works, but it needs something to run on a schedule and it produces numbers nobody can explain: when somebody asks why a contact is at thirty-four, the honest answer becomes “arithmetic”.

Choose the window by your own sales cycle, not by convention. If the typical gap between a first meaningful signal and a purchase is three weeks, a ninety-day window is mostly history and sixty is generous. If people take four months to decide, ninety days is the floor. When you genuinely do not know, start at ninety and shorten it once you have watched a few purchases happen.

Three resets that matter more than the decay curve

• →Reset on purchase. A buyer’s prospect score is meaningless — they are in a different queue now, and the questions about them are about delivery and the next offer, not about whether to call. Leaving buyers in the prospect queue is how a small team ends up pitching somebody who bought last Tuesday.
• →Hold after a real conversation. If you spoke and they said “not until the new budget”, the score should stop nominating them every week. Put a hold date on the record and let it re-enter the queue then. Otherwise the queue keeps handing you a decision you already made.
• →Suppress rather than subtract. Unsubscribed, bounced, refunded, asked not to be contacted — these are removals from the queue, not minus five points. Negative points only delay the same wrong answer by a week.

Those three are where small teams lose credibility with their own queue. A model that keeps nominating a customer who bought last week, or a person who politely explained their timing on the phone, teaches everyone to stop trusting the list — and a queue nobody trusts is a queue nobody works.

Two lines. Everything between them carries on as normal.

A score with five tiers gives you five decisions to make and five things to maintain. A score with two lines gives you the only two decisions a small business actually acts on: talk to this person, and stop spending on that one.

Setting the upper line without guessing

Work backwards from capacity. Decide how many conversations a week you want to have — five, ten, whatever is real — then look at where people actually land and put the line where roughly that many cross it. If forty people cross and you can hold five conversations, you do not have forty hot leads; you have a line in the wrong place. Move it, and keep moving it monthly at first. The threshold is a capacity dial, and the most common mistake is treating it as a truth about buyers.

Then make crossing it do something specific. “Becomes a qualified lead” is not an action. The strongest version is an offer of a conversation with the booking link already in it — the webinar-to-call funnel covers how to make that ask without it reading as a sales trap, and it is the same move whether the signal came from a session or from a score.

The lower line is a cost decision

People below the floor for a full window are not neutral. They cost you contact capacity, they dilute every open rate you use to make decisions, and at scale they cost you inbox placement with the people who do read you. Give them one honest re-engagement attempt, then remove them from active marketing — the re-engagement campaign covers the attempt, and deliverability basics covers why the removal is the part that pays for itself.

Review it quarterly, and prune more often than you add.

Scoring models rot in a specific way: they accumulate. Somebody adds a signal for a campaign, the campaign ends, the signal stays, and two years later the number is produced by machinery nobody can explain. Half an hour a quarter prevents all of it.

1. 01Which signals fired at all? A signal that fired four times this quarter is not a signal; it is a line of configuration you are maintaining for nothing. Delete it.
2. 02Which signals fired for people who never bought? Weight it down or remove it. This is where the click-based signals usually end up, and where the honest version of a small-business score gets smaller each year.
3. 03Which purchases came from people who never crossed the line? The most valuable of the four questions. Every one of those is a signal you are not capturing — go and look at what those people did instead, and consider scoring that.
4. 04Is the route threshold still matched to capacity? If the queue has been consistently longer or shorter than the conversations you can hold, move the line. The threshold is a capacity dial, and the dial needs turning as the business changes.

One rule holds the whole thing together: never add a signal without removing one. It sounds arbitrary and it is the reason the model still fits on a page in year three. An unexplainable score gets quietly ignored, an ignored score is worse than no score, and the reason it is worse is that it looks like coverage — everyone assumes somebody is working the queue.

Write the definition down where a person can read it, not only where the software stores it: the signals, the weights, the window, the two lines, the owner, and the date it was last changed. One page. If your model does not fit on one page, that is the finding, and the third question in the review above is usually where the excess came from.

A score sorts a queue. It does not predict a purchase.

This is the sentence to keep saying out loud, because everything about a number invites the other interpretation. Scores look like measurements. Yours is an opinion with arithmetic attached, and it is useful precisely to the extent that you remember that.

The two ways small teams get burned

The first is trusting it. Somebody at twelve points sends you a long email with a real question in it and gets a templated reply, because the queue said they were cold. A written question is the strongest signal in this entire guide and it will often carry almost no score, because writing to you is not a click. Read your inbox; the model is for the people who have not written.

The second is over-fitting it. With a dozen purchases behind you there is no statistical basis for rebuilding weights every month, so a monthly rebuild is fitting to noise while feeling like rigour. Change weights when the business changes — a new offer, a different price, a new way people arrive — and otherwise leave them alone and go and have the conversations. On a small list the whole job of the score is to make sure this week’s five conversations are with the five most likely people. That is worth having. It is not a forecast.