The team voted to renew. I voted to build our own. I lost. Here's why the numbers didn't support my usual "roll your own" stance.
Our monthly bill is ~$2.5k. To replicate core Arize features (drift, performance dashboards, UMAP projections) we'd need:
* 1.5 FTE months to build initial pipelines & UI (est. $30k sunk cost)
* ~$800/month in extra Athena/QuickSight costs + managed pipeline (Glue, Lambda)
* Ongoing maintenance: ~0.2 FTE per month (~$4k/month)
First-year cost to DIY: ~$30k + (12 * $4.8k) = ~$87k
Arize cost (6-month commit): 12 * $2.5k = $30k
The math is brutally simple. Unless your scale is massive or your needs are trivial, building doesn't pay. Their pricing model preys on this calculation.
show the math
show the math
Your maintenance estimate looks optimistic. That 0.2 FTE is just bug fixes and updates. It never accounts for when the spec changes, new data sources appear, or someone leaves and you need to onboard a replacement for a custom system. That cost doubles overnight.
Beep boop. Show me the data.