Okay, I’m deep in a vendor bake-off and need some real talk. We’re seriously considering Claw (the real-time feature flag service) because their cloud pricing looks great on paper. But our security team is pushing for an on-prem/local deployment for certain flags.
The vendor’s TCO sheet only lists license fees and hardware specs. I know from past experience that’s maybe 60% of the actual cost. The rest is hidden in our own infra and team hours.
I’m trying to build a realistic model. Here’s what I’ve got so far for annual overhead:
* **Baseline:** Their recommended VM config (8 vCPU, 32GB RAM per node, 3-node cluster for HA).
* **Direct Costs:** Our internal cloud team charges back ~$X per vCPU/month and $Y per GB RAM/month. That’s easy math.
Where I get stuck is the **indirect, ongoing stuff**. My list feels incomplete:
1. **Orchestration:** If we containerize it (we should), that’s ongoing K8s/OpenShift management overhead. How do you even quantify that? A percentage of a platform engineer’s time?
2. **Monitoring/Alerting:** We can’t just use their provided dashboard. We need to pipe logs/metrics to our central Grafana, set up PagerDuty alerts, define SLOs. That’s a chunk of DevOps time for setup and occasional tuning.
3. **Backup/Disaster Recovery:** They provide a backup script. But we need to integrate it into our scheduled backup system, test restores, manage retention. More storage costs, more script maintenance.
4. **Security Patching:** The base OS, the runtime, the Claw software itself. Each update needs testing in lower environments. That’s a recurring engineering task they don’t mention.
5. **Internal Support Burden:** When something goes wonky, our team is first line, not their support. That’s a real time cost.
Does anyone have a framework or even rough percentages they use for this? Like, “take the vendor’s hardware estimate and add 40-70% for operational burden”? Or a way to map these tasks to actual sprint hours per quarter?
I love a good experiment, but I need to get the baseline costs right first. Otherwise the ROI calc is garbage in, garbage out 😅
—kc
Sample size matters.