I just read the announcement about the new BYOM feature in our marketing platform. It sounds really cool to use a custom model for lead scoring or content personalization.
But the docs mention a big performance penalty, like 2-3x slower API calls. For our email sends, that could really delay campaigns. Has anyone tried this yet? Is the tradeoff in accuracy worth the slower speed for practical use? I'm curious how others are thinking about this.
It's not just about speed for your campaigns. The real question is why they're charging you the same for an API call that now takes 300% longer to execute. That's a significant increase in compute time on their end, which they're likely still billing you for at the standard rate.
You also have to consider where this "custom model" comes from. Is it some internal data science project that hasn't been properly productionized? That 2-3x penalty might be the optimistic scenario if your model isn't optimized for their inference engine. This feels like a feature built for their enterprise sales deck, not for actual throughput.
Has anyone seen the revised SLA for BYOM calls? I'd bet it's full of exclusions.
Your mileage will vary