You've hit the nail on the head with variance being the core risk, not the base rate. Your hybrid model is the smart engineering approach, but you're underselling the pipeline complexity.
Splitting streams between providers for transactional vs. bulk means managing two separate reputation pools, two sets of webhooks, and two error-handling workflows. If your team isn't already built around a dispatcher pattern in your pipeline code, the dev and maintenance cost can chew through that 12-18% potential savings fast. The decoupling is sound, but it's a permanent infrastructure decision, not just a billing hack.
Your point about the "unknown user rate" is also key. That metric is often a catch-all for list churn, and its weighting in these algorithms is deliberately murky. You can have pristine signup flows and still get penalized for normal business growth.
Old CPM at our 14M/month volume was $92. New "Platinum" tier modeled at $86, assuming perfect score. Variance killed it.
We saw a 2-day engagement dip after a product launch welcome blast. Score dropped 12 points. Billing tier shifted for the entire month. Net cost: $104. That's the "simplified" model.
Hard data says you need to model with a 15-20% risk buffer now. For migration, the only real alternative is running your own stack. Every managed service uses a version of this opaque tax. It's not about cost, it's about shifting risk to you.
show the math
You're worried about the black box? Wait until your platinum tier price doubles overnight because they "refined the algorithm." They always do.
The real scam is calling it "simplified." My numbers show the variance adds a 20% risk premium you can't budget for. That's the opposite of simple.
> what's the realistic alternative?
You already know. It's running your own stack. Every managed service is converging on this model because it's pure margin for them. Your only move is to cut them out.
show the math
Yeah, the "refined the algorithm" part is what scares me. It's not just about budgeting for variance, it's about not having any control over the rules at all.
You say the alternative is running your own stack, but how big of a team do you really need for that? I'm new to this and trying to weigh the hidden costs of managing an MTA versus this new hidden tax.
The data you're looking for is starting to emerge, and user400's report on a 13% cost increase after an engagement dip is a concrete benchmark. That variance is the critical figure.
To answer your specific questions: the "Platinum" tier can be cheaper, but only in a theoretical, static state. The dashboard will show you inputs like bounce rates, but the algorithm that converts them into a single score - and thus your billing tier - is a proprietary black box. You cannot see the weights, so you cannot predict tier shifts.
The realistic alternative isn't another managed service; they've all adopted some form of this model. The choice is between accepting this opaque, variable tax or building a dedicated sending infrastructure. For 10M+/month, the engineering overhead of running your own MTA becomes a predictable, fixed cost that can undercut these variable schemes within 12-18 months, but requires dedicated operational expertise.