Lead generation · 5 min read
Lead Scoring as a System, and When It Is Not Worth It
Lead scoring gets talked about as a capability, something a tool either has or lacks. It is better understood as a decision about your own time: you are choosing to spend attention building a rule so that later you can spend less attention deciding.
That trade only pays off past a certain volume. Below it you are maintaining machinery to sort a list you could have read.
The threshold, stated honestly
If you can open your new leads in the morning and read every one of them before your coffee goes cold, you do not need scoring. You need a habit of doing that. Any rule you write will be less accurate than your own eyes on twelve messages.
The signs that you have crossed the line are behavioural rather than numerical, which is why nobody can tell you the number:
- You have started skimming. Not reading, skimming — scanning for the ones that look important. That is an unwritten scoring model, already running, with no rule and no review.
- Leads arrive faster than you close them. The list grows week over week and you have stopped opening the bottom of it.
- You no longer recognise the names. When every lead was someone whose comment you remembered, you had context. When they are strangers, you need structure.
- Somebody else answers your DMs now. The moment a second person is involved, "who should I answer first" needs an answer that lives outside your head.
Cross one, and scoring is premature. Cross three, and the absence of it is already costing you conversations.
The four parts, of which most setups build two
A scoring system is inputs, a rule, an action, and a review. Almost every implementation has the first two, calls itself done, and quietly becomes decoration.
Inputs are the fields you already maintain, not new ones you now have to. If a score depends on somebody manually tagging every lead, the score dies the first busy week.
The rule turns inputs into a small number of buckets. Three is the right number. Ten is a spreadsheet.
The action is what actually changes in your day because of the bucket. If nothing changes — if you would have worked the list in the same order anyway — you have built a label, not a system.
The review is the part everyone skips, and it is the only part that makes the other three trustworthy.
A score nobody has ever compared against an outcome is not a measurement. It is a preference, written down.
Inputs worth using, and inputs that flatter you
Cheap, honest, already there
Feels sophisticated, misleads
The right column is the reason so much lead scoring reads as astrology — it scores affection, and affection correlates with buying much less reliably than anyone building the model expects. The longer argument about what actually makes a lead hot is worth having before you write a single rule, because the rule will inherit whatever definition you started with.
Build buckets, not points
A 100-point scale invites a false claim: that 71 and 68 are meaningfully different. Nobody has ever behaved differently towards those two people, and pretending otherwise means every argument about the model becomes an argument about weights.
Three buckets, defined by what you will do:
- Answer now. Replied, asked something specific, window closing today.
- Answer today. Captured recently, has not replied yet, still inside the window.
- Leave it. Window closed, or nothing beyond a keyword. Not rejected — just not where the next hour goes.
That last point is worth being direct about. This product has stages, tags, a replied flag, interaction counts and a value field. It has no lead score, and adding one is not on a roadmap somewhere, because a filter over real fields tells you why a person is in the list and a number tells you only that they are.
The review loop, which is the whole thing
Once a conversation ends, mark the lead: Converted, or Not a fit. It takes a second per lead and it is the only input to the review.
Then, once a month, run one comparison. Take the people who converted and look at which bucket they were in the day before. If your top bucket is not over-represented among them, the rule is wrong.
Rules decay, quietly
A rule written during a launch encodes what mattered during a launch. Three months later the launch is over, the audience has shifted, half your inbound is coming from one reel that unexpectedly travelled, and the rule is still confidently sorting for conditions that no longer exist.
Nothing alerts you to this, because a decayed rule keeps producing output. Put a date on it. When you write the rule, write the month you will re-derive it, and treat it as expiring rather than permanent.
The version that is almost always enough
Sort by time remaining in the window, then read the ones that replied. That is not a compromise for people who cannot build a real system — it uses the two most predictive signals available and needs no maintenance at all.
Add complexity only when you can name the decision it improves. What signals genuinely exist in a thread to build anything more elaborate on, and where a model helps rather than hallucinating confidence, is worth knowing before you design the rule. And if the reason your list is unmanageable is that it is full of people who never wanted anything, the fix is not scoring at all — it is the offer that filled it.
