ICP matching and feedback

Every person Embers finds is checked against the ideal customer profile you defined, and the verdict is the single biggest lever on their score. This article covers how the verdict is reached and how to overrule it.

What the match is based on

Matching reads the structured criteria produced when you defined your ICP, not your raw description: target job titles, departments, industries, locations, company size ranges, buying signals, and the negative titles and industries to exclude. Against those it weighs seven dimensions:

Two things carry a veto rather than a weight. A hit on your negative titles or negative industries overrides every positive. So does being a peer or competitor: someone selling what you sell, a vendor in your category, industry media, or a coach or course creator in your space is not a lead, however senior their title.

Beyond the profile itself, the classifier also sees how the person engaged, which topics and comment intents were involved, and, once the profile has been enriched, their work history, skills, education, and certifications. A role started within about 90 days is treated as a strong buying signal. Missing data never counts against someone, it just returns a neutral verdict on that dimension.

Confidence, and what the badge means

Each verdict comes with a confidence from 0 to 1. The lead panel turns the pair into one label:

• Strong: a match at 0.8 confidence or higher.
• Likely: a match below 0.8.
• Unlikely: a non-match below 0.8.
• No match: a non-match at 0.8 or higher.
• Pending classification: not yet evaluated.

Confidence also feeds scoring. A match at high confidence pushes the ICP score toward 100 and lifts the composite; a confident non-match crushes it. An unclassified lead sits at the neutral middle and takes a small discount for the uncertainty, so it neither floats to the top nor disappears.

Correcting a match

1. Open the lead

   From the leads list or the Priority Queue, open the lead and find the ICP Match card.

2. Tag the fit

   Choose Tag ICP fit, then Is ICP or Not ICP. Add a short reason if you have one. It is optional, and it is the part that teaches Embers the most.

3. Save

   Your tag replaces the automatic verdict. The card gets a coloured edge and reads “You tagged as ICP” or “You tagged as not ICP”.

Remove tag clears the override and returns the lead to the automatic verdict with its usual discount for being unclassified.

What changes after you correct one

Three things happen, on three different clocks.

• Immediately, for that lead. The manual verdict wins: the ICP score is set to 100 or 0, confidence to 1.0, and the composite score is recalculated on the spot. A manual approval also satisfies the ICP half of the destination rule, so a lead you approve by hand can be delivered to Slack, HubSpot, or a webhook even when the classifier disagreed.
• On the next classification pass, for everyone. Up to ten of your corrections, five positive and five negative, are shown to the classifier as examples, so it calibrates against your judgement rather than the generic criteria.
• Once five corrections have built up. Embers rewrites the ICP criteria themselves, adjusting titles, departments, negative exclusions, and buying signals to match who you actually keep saying yes to.

Your feedback always takes precedence. It is checked before the deterministic guardrails run, so an override is never quietly reversed by a title or industry rule.

When the automatic verdict is usually wrong

• The profile is thin. A headline with no company and no work history gives the classifier almost nothing. Enrichment usually fixes this by itself.
• Your negative titles are too wide. A negative entry matches inside current side roles as well as the main title, so a broad word can veto people you want.
• You sell services. Service businesses get aggressive peer exclusion by design, which is right for competitors and occasionally wrong for a specific buyer. Tag them and move on.

Related