When you receive five requests a week, you read them all. At fifty, you handle the most recent and let the others go cold. Automated qualification exists so that processing order no longer depends on arrival time.
Definition
Automated lead qualification means analysing incoming requests to estimate which best match your offer, and in what order to handle them.
Technically it is a classification task: the model reads the message and available context, then assigns a category or a score according to criteria you defined.
The decisive point is that those criteria must come from you, not the model. "Qualify this lead" gives unusable results. "Does this lead match a company under twenty people, in these three sectors, with a need within three months" gives a workable ranking.
What you can ask a model
| Reliable | Unreliable |
|---|---|
| Spotting the sector | Estimating an unmentioned budget |
| Detecting expressed urgency | Guessing purchase intent |
| Identifying the stated need | Predicting probability of closing |
| Discarding off-topic requests | Scoring a prospect's quality |
| Summarising in three lines | Deciding alone not to reply |
The left column rests on what is written in the message. The right column requires guessing what is not there, and that is exactly where a model produces plausible, wrong answers.
How to proceed
Write your criteria in plain words first. If you cannot say what makes a good prospect, no system will know for you.
Sort, do not eliminate. A system that files into three piles is useful. A system that deletes requests will lose you customers without you ever knowing.
Ask for the reasoning. A ranking with its one-sentence reason is verified in three seconds and makes errors immediately visible.
Test on history. Replay the system on the last six months of requests and compare with what actually closed. It is the only honest way to know whether it sorts well.
Watch for bias. A model tends to underrate badly written messages, which are not the least serious. See Algorithmic bias.
If your system scores individuals or decides alone to discard a request, you enter regulated territory. Assisted sorting is not a problem, an automated decision affecting someone is.
Frequently asked questions
Score or categories?
Categories, almost always. A score out of a hundred gives a false impression of precision, whereas three piles honestly reflect what the system can distinguish.
Can you enrich automatically before sorting?
Yes, and it clearly improves sorting: reading the company's site gives far more context than the form alone. Just watch the time and cost that enrichment adds to every request.
What about requests ranked at the bottom?
A reply, always, even a standard one. A misclassified request left unanswered is a lost customer, and you will never get the feedback that would have let you correct it.
How do you build this sorting concretely?
By connecting the form to a Workflow that enriches, classifies and notifies, with the reasoning displayed. Our n8n course builds that complete chain, testing on history included.