Engagement scoring model
Definition
Engagement scoring model
An engagement scoring model gives a prospect or customer a running number based on what they actually do: pages viewed, emails opened, product sessions, meetings attended. Sales and marketing teams use that score to decide who gets worked next and in what order.
This is a sales and marketing measure. It has nothing to do with employee engagement surveys, which ask staff how they feel about their own jobs and produce a completely different kind of number.
The model answers one question. Of the thousands of accounts in the database, which handful is showing enough behaviour right now to justify a human picking up the phone?
Scores are relative, not absolute. A score of 78 means nothing on its own — it means something only against the threshold your own closed deals crossed last quarter.
Key takeaways
- Score observed behaviour, never stated intent or demographic fit alone.
- Weight signals by how well they predicted closed revenue in your own history.
- Decay points over time, or dormant accounts keep their score forever.
- Recalibrate quarterly against actual outcomes, not against how the numbers feel.
How it works
An engagement scoring model assigns points to specific behaviours, decays those points over time, and sums them into a single figure per contact or account. A threshold then routes the record to a human, a nurture track or nothing at all.
| Signal | Typical weight | Decay |
|---|---|---|
| Pricing page view | High | 30 days |
| Demonstration request | Very high | 90 days |
| Email open | Low | 14 days |
| Webinar attended to the end | Medium | 60 days |
| Product login after trial | High | 30 days |
| Support ticket raised | Low | 30 days |
Weights should come from your own history, not a vendor template. Pull the last two hundred closed deals, look at which behaviours appeared before the first meeting, and weight those above the ones that merely correlate with being busy.
Define the denominator and the window before you score anything.
Public measurement practice is strict about this. The UK Government Digital Service manual calculates digital take-up by dividing completed digital transactions by the total across all channels in the same fixed period.
Behavioural measurement at scale is well established in government too. The Digital Analytics Program, run by the US General Services Administration, measures traffic and engagement across thousands of federal websites under one shared analytics account.
Scoring sits downstream of demand generation and upstream of the sales conversation. The campaigns create the behaviour; the model reads it and decides whose turn it is.
Keep the model separate from fit. Firmographic fit says whether an account should ever buy, while the engagement score says whether now is the moment — two different questions that break badly when merged into one number.
Agree the threshold with sales before launch, in writing. A score nobody acts on is a reporting exercise, and reps who were not consulted will carry on working the accounts they already liked.
Negative signals deserve points too — an unsubscribe, a bounced email or a cancelled meeting should subtract, and most templates forget this. A model that only ever counts up keeps promoting accounts that are actively pulling away.
Watch what the score does to lead qualification rate. If reps start rejecting more scored leads than before, the weights are rewarding activity that looks like interest without being interest.
Examples
Four settings show the model doing real work — in each one the signal set differs, because the buying behaviour differs and the thing being predicted is not the same.
A business software vendor scores pricing page visits, trial logins and documentation reads. A free trial account that logs in five days running scores higher than a chief executive who opened three newsletters, because usage predicts purchase and reading does not.
Outsourcing buyers behave differently. A firm comparing providers downloads pricing guides, reads case studies in one sector and books a discovery call, so lead generation teams weight sector-specific content far above general traffic.
Existing accounts get scored too. A customer success team runs a customer engagement process where falling logins and unanswered check-ins drag the score down, flagging renewal risk months before the contract date.
Publishers and media businesses score for a different outcome. Their model predicts subscription renewal rather than first purchase, so the engagement rate on individual articles matters more than any single visit.
Professional services firms score meeting behaviour rather than clicks. A prospect who brings a second decision maker to the second call outranks one who downloads every white paper and still turns up alone.
Related terms
- Revenue per lead: the outcome measure that tells you whether the scoring thresholds are set correctly.
- Cost per lead: the acquisition cost a scoring model is meant to make go further.
- Lead qualification rate: the fastest early warning that weights have drifted.
- Demand generation: creates the behaviour the model reads.
- Engagement rate: a single channel ratio; the scoring model combines many of them into one figure.
FAQ
What is the difference between engagement scoring and lead scoring?
Lead scoring usually blends fit and behaviour into one number. Engagement scoring isolates behaviour, which makes it easier to see whether interest is rising or fading on an account that already fits.
Which signals matter most?
The ones that preceded your own closed deals. In most business markets that means pricing page visits, demonstration requests and repeat product usage, with email opens near the bottom.
How often should scores decay?
Give most signals a 14 to 90 day half-life depending on how long your sales cycle runs. Without decay, an account that was interested last year outranks one that is interested today.
Can an engagement scoring model work without a marketing automation platform?
Yes, at small scale. A weekly export, a weighted spreadsheet and an agreed threshold will outperform no model at all, though the manual version breaks past a few thousand records.
How do you know the model is working?
Compare win rates and cycle length for leads above and below the threshold over a full quarter. If the two groups convert at similar rates, the score is sorting noise.
Compare outsourced demand generation partners and their reporting on Outsource Accelerator.
Independent
