Build and test a lead scoring model

Task 1: fit points

Only L1 reaches 50. L5 loses 15 points because Tailspin Toys has more than 2,000 employees.

Task 2: engagement points

• L1: webinar 15 + pricing 10 + guide 5 = 30. Nothing is older than 90 days.

• L2: demo 30 + guide 5 halved to 2.5 (120 days ago) = 32.5.

• L3: guide 5 + six opens at 0 points = 5.

• L4: webinar 15 + two guides 10 + pricing 10 + reply 10 = 45.

• L5: webinar 15, pricing 10 and guide 5, all older than 90 days, so halved: 7.5 + 5 + 2.5 = 15.

• L6: reply 10 + pricing 10 + webinar 15 = 35.

• L7: demo 30 + pricing 10 + webinar 15 + two guides 10 = 65.

• L8: webinar 15 + guide 5 = 20. 70 and 60 days are within 90, so no decay.

Task 3: totals and ranking

Task 4: handover

L1 (80) and L2 (72.5) are MQLs. After the calls, only L1 is an SQL: Anna Berger has the need, the authority as safety manager, a budget and a date. Jonas Weber is interested, but he does not decide and the timing is next year. L2 stays an MQL; a good next step is to ask who in purchasing should be involved, and to nurture the lead until then.

Task 5: decay

Without decay, L5's engagement would be 15 + 10 + 5 = 30, and the total 35 + 30 = 65: an MQL. With decay it is 50, below the threshold. Decay is useful because Erik Lindqvist's interest is more than three months old; calling him now as a "hot" lead would probably be too late, or too early again. Proseware should also check the fit itself: with 2,400 employees Tailspin Toys is bigger than the ICP, and the team should ask whether that matters for this product or whether the size limit is too strict.

Task 6: test the threshold

• SQL rate at 60 points or more: 12 / 40 = 30 %.

• SQL rate below 60: 8 / 160 = 5 %.

• Share of all SQLs that came from MQLs: 12 of 20 = 60 %. The score separates good and weak leads clearly (30 % against 5 %), but 40 % of the SQLs came from leads below the threshold.

• With a threshold of 45: 40 + 50 = 90 handovers with 12 + 6 = 18 SQLs, a rate of 18 / 90 = 20 %. Only 2 of the 20 SQLs (10 %) would come from below the threshold.

• Recommendation: 90 handovers are more than the 60 sales can handle. Any reasoned answer counts. For example: keep 60 as the threshold, but send demo requests to sales straight away whatever the score; or lower the threshold to 45 only if sales gets more capacity; or look at which leads between 45 and 59 became SQLs and add points for what they had in common. Note that 200 leads is a small sample.

Task 7: improve the model

• L7 works well: the competitor scores highest on engagement (65) but ends at -5. Without negative points, Contoso Learning would have been the top lead.

• L4 works well, mostly: a student researching a thesis is very active but will not buy. The -30 keeps her out. The model would be better if it recognised her by a question in the download form, rather than by a manual note.

• L6 is misleading: Lotte Visser replied to an email and visited the pricing page last week. Unsubscribing from the newsletter does not mean she has no interest in buying; she may simply not want a newsletter. The -20 pushes an active conversation below the threshold. Better: no negative points for unsubscribing, but no more newsletters either. That second part is not optional.

• Other good changes: demo requests go to sales directly whatever the score; check bounces quickly, because a bounce often means the contact has left the company, as may be the case at Alpine Ski House; review the model every quarter against real results.

GDPR. Giving people scores based on their behaviour is profiling in the sense of Article 4(4) GDPR, the General Data Protection Regulation. Proseware must:

• tell people about it, for example in the privacy notice shown on the download form (Articles 13 and 14);

• stop profiling a person for direct marketing as soon as they object (Article 21(2) and (3));

• keep to data minimisation (Article 5(1)(c)): score only what the model really needs.

This is an overview for teaching, not legal advice.