How should I build an email marketing engagement scoring model for opens and clicks?

Hi! A solid “best-practice” engagement scoring model for email marketing usually boils down to two ideas: (1) weight behaviors by how strongly they indicate real intent (clicks and conversions > opens), and (2) weight recency heavily (what they did in the last 7–30 days matters a lot more than 6 months ago). If your model does those two things and you calibrate thresholds against your own list’s send frequency, you’re on the right track.

A practical setup that works well for newsletters + automations:

1) Define the time windows first (because thresholds depend on send cadence)
Pick windows that match how often you mail:

• “Very recent”: last 7–14 days
• “Recent”: last 30 days
• “Aging”: last 60–90 days
  After ~90 days, most lists treat activity as weak evidence unless you send very infrequently.

2) Choose event weights (typical relative weighting, not magic numbers)
Because opens are increasingly noisy (privacy features + image prefetching), many teams treat opens as a light signal and clicks as the core engagement metric.

• Click (unique, per send): strong signal (often 3–10× the value of an open)
• Open (unique, per send): weak signal (helpful, but don’t over-trust it)
• Reply / forward / “add to contacts” / preference-center update (if you can track it): very strong
• Conversion / purchase / key site event (if applicable): strongest

If you only have opens + clicks, a common “sane” rule is: 1 open = 1 point, 1 click = 5 points (or 1 and 3, depending on how click-heavy your content is). The exact numbers matter less than the ratio and the recency weighting.

3) Add recency weighting (simple and effective)
Instead of giving the same points forever, apply a multiplier by age, for example:

• Activity in 0–14 days: ×1.0
• 15–30 days: ×0.6
• 31–60 days: ×0.3
• 61–90 days: ×0.1
  Or use a smooth decay (e.g., each week older is worth ~20–30% less). This prevents someone who clicked twice six months ago from still looking “engaged.”

4) Normalize by opportunity (optional but often worth it)
People can’t click if they didn’t receive many emails. If your send frequency varies a lot across segments/automations, consider scoring an engagement rate in addition to raw points, e.g.:

• click-rate-per-delivered over the last 30/60/90 days, or
• points divided by emails delivered in the same period

This keeps “highly engaged” from just meaning “we emailed them a lot.”

5) Set thresholds using your own distribution (don’t guess)
The best way to pick “highly engaged / warm / inactive” is to score everyone, then look at the score distribution and outcomes (complaints, unsubscribes, conversions, inbox placement trends). A common pattern is:

• Highly engaged: top 10–25% of scores or anyone with a click in the last 30 days
• Engaged/warm: some activity in last 60 days (opens and/or at least one click in last 60–90)
• At-risk: no clicks in 60–90 days (opens only, or nothing)
• Inactive: no measurable engagement in 90+ days (adjust longer if you mail monthly)

If you want a very practical “deliverability-friendly” rule of thumb: use clicks for your “engaged” definition when possible, and treat opens as a secondary signal (especially on Apple-heavy audiences).

6) Add a few “negative” signals (recommended)
To protect deliverability and reduce spam-folder risk, you can subtract points or flag subscribers for suppression when they do things like:

• mark as spam / complaint (usually an immediate suppression)
• repeated bounces (hard bounce = remove; soft bounce patterns = monitor)
• long-term no engagement despite frequent sends (risk of future complaints)

7) Validate and tune with a lightweight experiment
After you pick initial weights/thresholds, run a 2–4 week test:

• Send your normal cadence to “highly engaged”
• Reduce frequency or shift to a re-engagement automation for “at-risk”
• Suppress or sunset “inactive” (or send a final permission pass)
  Then compare click-through rate, unsubscribe rate, spam complaints, and conversions. Adjust weights so the score better predicts the outcomes you care about.

If you tell me roughly (a) how often you send newsletters, (b) how many automated emails a typical subscriber gets per month, and (c) what you want the score to optimize for (deliverability vs. revenue vs. retention), I can suggest a concrete point table + recency decay and a clean set of thresholds that fit your cadence.