There is a simple experiment that illustrates the problem of mass email marketing. Take your last newsletter and look at the open data by the hour. You will see a peak around the send time and almost zero afterwards. Most opens happen in the first 90 minutes. Those who did not open it then will never open it.
Now ask yourself: is Tuesday at 10:00 the right time for a lead who gets up at 5:30 to go to work at the factory to open your email? For the self-employed person who checks their email at 22:00 when their children are in bed? For the finance director who only checks their professional email on Mondays and Thursdays?
Predictive email solves this by sending to each contact within their optimal individual window, calculated by a model that has learnt their historical open behaviour.
How the Bayesian optimisation model works
For each contact in the database, the model analyses the following:
- Open history: which days and times the contact opened emails in the past
- Click history: correlation between the time of opening and the probability of clicking
- Recent trends: if behaviour has changed in recent months, the model gives more weight to that than to older history
- Email type: the same contact can have different open patterns for newsletters vs transactional emails vs prospecting emails
The Bayesian model does not wait until it has perfect data to make predictions. It starts with an initial estimate (this can be the average behaviour of similar contacts) and updates that estimate with each new piece of data about the specific contact’s behaviour. The more emails a contact receives, the more accurate the prediction will be.
Expected results
Research from platforms such as SendTime (a specialised tool for optimising sending time) and analyses from platforms such as Klaviyo, with the Smart Send Time functionality, consistently show:
- Open rate improvement: on average between 15% and 30%
- CTR improvement: between 10% and 20% (lower than the open rate, as the content itself remains the same)
- Reduction in unsubscribes: between 5% and 15%, as emails are received when people are more receptive
The impact is greater in lists with high profile heterogeneity (with different B2B sectors and roles) and smaller in very homogeneous lists, where all contacts have similar behavioural patterns.
How to implement it without changing platform
Most modern email marketing platforms have a version of send time optimization built in: Klaviyo, ActiveCampaign, Brevo and Mailchimp all offer the functionality of optimal sending time per contact in their professional plans.
For more sophisticated implementations with proprietary models, the architecture is as follows: the email marketing platform receives sending instructions with precise timestamps for each contact, triggered by the prediction system, running on n8n or a proprietary service. It requires API-based integration.
BAI’s Email Predictive implements this advanced architecture with its own Bayesian model.