Lead scoring for B2B marketing: keep fit and engagement separate
Quick answer
Keep fit and engagement separate in B2B lead scoring. A company can be a great match without showing much interest, while an enthusiastic reader may never become a customer. Your score should help sales choose whom to follow up with, not pretend to know who is ready to buy.
A busy contact is not automatically a good prospect
A student can read every article on a website. A competitor can download a pricing guide. A customer can visit a support page repeatedly. Engagement is real in each case, but the commercial meaning differs.
Start with fit: organisation, market, role and the problem your business can solve. Then assess behaviour in context. A request for a specific meeting is usually more actionable than repeated visits to introductory content.
HubSpot’s lead-scoring documentation distinguishes fit and engagement scores. The distinction is useful even if you use another platform or a spreadsheet. A single blended number can conceal a prospect who scores highly for the wrong reason.
Make the first model deliberately explainable
Choose a small set of criteria that sellers recognise. Assign points only when the data can be obtained and maintained reliably. Record the source of each field and how missing values are handled.
Use hard exclusions for conditions that make outreach inappropriate or commercially irrelevant. Do not let frequent activity cancel out an exclusion. Keep current customers and open opportunities in the correct operational route rather than treating them as new leads.
For behaviour, distinguish explicit requests, evidence of evaluation and general interest. Place limits on repeated low-value events so ten page refreshes do not outweigh a meaningful enquiry. Apply time decay where an old action should lose relevance.
| Fit | Behaviour | Possible route |
|---|---|---|
| High | Explicit relevant request | Named seller with context |
| High | General educational interest | Development content |
| Unclear | Strong activity | Research fit before routing |
| Low or excluded | Any activity | Respect exclusions; avoid inappropriate handoff |
Route by the combination
A simple matrix can be more useful than a score ranked from one to a hundred. High-fit contacts with a clear commercial request go to the responsible seller. High-fit contacts with general interest receive relevant development content. Low-fit contacts should not consume the same sales capacity merely because they are active.
Document what happens after routing. Who owns the record? What information must be included? What response is expected? What happens if the seller rejects it? Our CRM handoff checklist addresses the process that scoring alone cannot supply.
If several contacts belong to one account, preserve their individual actions while considering the account context. Do not simply add every contact’s points and assume the largest organisation has the strongest buying intent.
Validate precision and missed opportunities
Look at a recent group of contacts once sales has had time to follow up. If the model passes over 100 and sales accepts 30, the acceptance rate is 30%. Agree what “accepted” means first. Otherwise you’re comparing one seller’s judgement with another’s and calling it model performance.
Also review opportunities that the model failed to route. A model can look precise by passing very few contacts while missing much of the useful demand. Compare both workload and coverage with the previous process.
Use an older period to develop the rules and a later period to evaluate them. Repeatedly tuning the score against the same successful deals risks overfitting. Our machine-learning evaluation guide explains the underlying problem, even when the scoring model is manual.
MarTech Logic’s view: scoring is a queue-management tool
The score should help people decide where to spend attention. It should not become a substitute for a conversation or a claim that someone has budget.
Keep a monthly review of accepted, rejected and missed cases. Read the rejection reasons. If most failures come from incorrect company size or duplicate records, fix data quality before adding more scoring sophistication. If sellers ignore well-qualified records, investigate ownership and capacity.
For AI-assisted scoring, ask what outcome the model was trained to predict. Historical wins may reflect past targeting and seller attention rather than an objective map of the market. Require an explanation of inputs and limitations, and retain a way to review individual decisions. A modest, understood model that produces a better queue is more useful than an opaque score nobody trusts.
Related reading
ABM account selection: build a target list sales can defend · B2B intent data: verify the signal before you buy the story · CDP vs CRM vs data warehouse: choose around the customer task