Intent Scoring: Build and Test a Practical Priority Model

Intent scoring ranks eligible people or accounts for a defined next action using research and engagement evidence. It helps a team decide what to review first. A priority score is not proof that someone will buy, and an arbitrary 80-point score is not an 80% purchase probability.

Start with eligibility, choose a few explainable inputs, set a review threshold that fits your capacity and test the results against your existing process. Keep fit, research activity, identity and permission as separate questions.

Gate eligibility before assigning points

Remove or hold records that do not qualify for the proposed action. Examples include an unsupported geography, poor company fit, an unresolved identity or a suppression requirement. A high activity score should not override a hard exclusion.

Define the unit too. Are you prioritizing individual contacts, companies or workspaces? Account-level research should not become a claim that a particular employee performed a search. Missing information should be labeled as unknown instead of silently becoming zero intent.

Choose inputs you can explain

• Fit: Whether the company or role matches your intended market.
• Research relevance: Whether the available topic relates to your offer.
• Recency: How old the available evidence is.
• Direct engagement: Relevant first-party actions where available and permitted.
• Identity context: Whether the record can support the intended action.
• Negative evidence: Reasons to hold, reject or suppress the record.

Use only fields actually supplied by your sources or verified through your own process. Do not invent confidence values, exact search queries or event timestamps because your proposed model would benefit from them.

A demo request and a third-party research signal are different evidence. Preserve that distinction in the reason codes and handling. An explicit request for contact should not be ignored because an inferred priority score is low.

A simple, original scoring example

The following is a hypothetical rule set for a research queue, not a BrandWell scoring feature, recommended universal weighting or predictive model:

• Relevant topic confirmed during review: 30 points.
• Evidence falls inside the team’s defined recent window: 20 points.
• Role is relevant to the task: 20 points.
• Relevant direct engagement is available: 30 points.

Eligibility and suppression are gates outside the score. A missing input goes to review. The team initially investigates eligible records scoring at least 50, then adjusts the rule after examining outcomes and workload.

A record with a relevant topic, recent evidence and relevant role scores 70. That means it meets this example’s review rule; it does not mean the person is 70% likely to buy. The point weights and threshold would need testing in the actual business.

Cap repeated contributions and avoid counting one event several times. If a pricing-page visit is already included in direct engagement, adding a separate journey-stage score derived from that visit may double-count it.

Set thresholds around a real action

Define what happens above and below the threshold. “High priority” should identify an owner and next step, such as research or approved follow-up. It should not automatically authorize contact.

If the team can research 40 records a week and the rule produces 400, investigate eligibility and ranking before creating an unmanageable queue. Track the age of unreviewed work. Letting tasks expire without an outcome hides whether the model is useful.

Use current evidence and a written expiry rule. The useful lifetime of a signal depends on the source, task and buying process; there is no universal decay period in this guide.

Test whether the priority rule works

Define an outcome you can label consistently, such as “accepted for account research” or “qualified opportunity within the stated period.” Evaluate top-ranked records and a sample below the threshold. Looking only at records you selected hides missed cases.

Google’s classification guide explains precision and recall as different measures. Precision is the share of positive predictions that are correct; recall is the share of actual positive cases found. Threshold changes affect both.

For an original example, a rule flags 20 records. Review finds 12 meet the defined positive label, while a complete labeled evaluation set contains 30 positive cases. Precision is 12 / 20 = 60%, and recall is 12 / 30 = 40%. These are results for that label and test set, not proof of buying intent or general data quality.

If you have not labeled the broader evaluation set, you cannot calculate recall from the selected records alone. Likewise, a higher reply rate does not establish that the rule found everyone who might respond.

Separate ranking quality from business impact

High-priority records may have been more likely to convert before the new workflow. To evaluate whether using the score improves outcomes, compare it with the existing prioritization process under a documented design.

Keep the market, offer, review capacity and measurement window comparable. Where feasible, use random assignment. If you use a before-and-after or matched comparison, describe selection and timing limitations rather than claiming causation from associated revenue.

Measure review effort, action completion, qualified outcomes and costs. A score that sorts records well can still fail operationally if no one acts or the next action is irrelevant.

Use MarketPulse as an input to a review process

MarketPulse provides topic audiences and ICP filtering, with intent-data plans starting at $500/mo. A pull retrieves available data from the previous 24 hours. Counts, fields and overlap vary. Review the actual output before designing a scoring rule around it.

Starter includes one topic with weekly refreshes and 25,000 monthly contact exports. Growth includes three topics with daily refreshes and 100,000 monthly exports. Scale includes ten topics with daily refreshes and unlimited exports.

Growth and Scale include AIMEE for the initial user. With configured Automations and tools, assign research or follow-up work and preserve review where needed. This does not establish that MarketPulse supplies a native predictive score or every field in a proposed model.

Keep a model record you can maintain

• Eligibility rules, inputs and missing-value handling.
• Weights, caps, recency rules and thresholds.
• Model version and effective date.
• Reason codes, owner and permitted next action.
• Evaluation labels, sample, period and results.
• Change history, corrections and stop conditions.

Revisit the rules when the offer, audience, source or sales process changes. For client work, keep each client’s definitions and outcomes separate. A rule useful for a software seller may not fit an IT services company.

Intent-scoring questions

Is intent scoring the same as lead scoring?

Intent evidence can be one input to lead or account prioritization. A broader lead model may also include fit, explicit requests and lifecycle context. State what your score actually represents.

Do I need machine learning?

No. A small explainable rule set may be sufficient for a review queue. Additional complexity needs appropriate outcome data, validation and maintenance.

What should I do first?

Choose one action and define eligibility and evidence. Review a relevant MarketPulse topic, then test a transparent priority rule before expanding.

Reviewed and updated October 2, 2026.

Written by

Farnaz Kia