Mass email is dying
Open rates at 20%. CTR at 1%. List burnt out in 18 months. Every send erodes your domain’s reputation and reduces the deliverability of the next one. Traditional mass email no longer scales.
Predictive email doesn’t send more. It sends less — but when it matters. A Bayesian model calculates the probability of opening, clicking and converting for each recipient and each moment. If the probability exceeds a threshold, it fires. If not, it waits.
How we build it
1. Historical open rate analysis
We need at least 6 months of send history with recipient-level data: who opened what, when, and what they did afterwards. If you don’t have it, we can help you start collecting it (3–6 months of baseline).
2. Bayesian model
For each recipient × email type × time of day combination, the model predicts the probability of opening and conversion. It readjusts automatically with each new send. Naive Bayes + simple features = fast and explainable.
3. Integration with your tool
Mailchimp, ActiveCampaign, Customer.io, Brevo. Whatever platform holds your templates and domain, we plug into it. Your team keeps using the tool they already know.
4. Trigger rules
We define thresholds with you: “fire only if open probability > 35%”. We fine-tune it in production over 4-6 weeks. We start conservatively and scale when the model proves it’s accurate.
What changes when this works
- Sending volume -50% (only to those who will open).
- Effective open rate +40-60% amongst those who do receive it.
- Domain health (sender reputation) recovers and improves the deliverability of ALL your sends.
- Your list stops burning out at the same rate.
When we do NOT recommend this
- If your list is small (<5,000 contacts): there is not enough statistical signal.
- If your product is ultra-transactional (confirmation sends, password reset): predictive here makes no sense — they must always arrive.
- If your business depends on mass communication due to regulation (banking, health, legal): you may have to send to everyone whether you like it or not.
Stack we use
- scikit-learn (Naive Bayes + lightGBM) for the model.
- Python pipeline with MLflow for versioning.
- Mailchimp / ActiveCampaign / Customer.io API direct.
- Metabase for weekly model performance report.
We start with a diagnostic session
Before quoting the full setup we run a 90-minute session. We look at your current list, your open-rate history and your email tool together, and we leave with an honest recommendation: whether this service fits where you are right now, or whether it makes more sense to start with something else.
We don’t charge for that session. If you’re interested, get in touch.