Avoid Usage-Based Surprise Bills: Predictable Cost Planning for Self-hosted Feature Flags
SaaS feature flag platforms commonly include usage-based billing variables that are hard to forecast at the start of a fiscal year. This page catalogs those variables, explains how they inflate unexpectedly, and shows how self-hosting converts them into a capped, forecastable infrastructure line.
TL;DR
- ▸SaaS feature flag billing commonly includes 4–6 variable meters: seats or feature tiers, MAU, request volume, service connections, logs or event traffic, and enterprise add-ons.
- ▸Any one variable growing unexpectedly can trigger a tier bump, adding hundreds to thousands of dollars per month.
- ▸Self-hosting converts those billing variables to infra cost + ops time +, if needed, a flat annual FeatBit enterprise license — all forecastable at the start of the year.
- ▸When building your annual budget, the self-host model has a visible ceiling: adding more engineers, services, or environments does not create new per-seat, per-request, or per-MAU vendor line items.
SaaS Billing Variables That Drive Unpredictability
The following meters appear across major SaaS feature flag platforms. LaunchDarkly publishes service connections and MAU, GrowthBook publishes seats plus CDN requests and bandwidth, Flagsmith publishes requests and team members, ConfigCat publishes config downloads and network traffic, DevCycle publishes MAU plus cloud/server config requests and events, and Unleash publishes seat pricing plus API traffic. Not all meters apply to every vendor; consult current pricing pages directly for your plan.
Seats / feature tiers
Medium-High riskCharged per engineer or embedded in a higher tier once your team needs SSO, approvals, RBAC, or audit workflows. As teams grow, seat and feature-tier cost compounds quickly.
Monthly Active Users (MAU)
High riskApplied to client-side SDK flag evaluations. If you roll out feature flags to your end-user product (not just internal tools), MAU can spike with user growth campaigns, seasonal traffic, or new product launches — in ways engineering cannot predict from their desk.
HTTP requests / CDN traffic
High riskSome platforms meter config downloads, API polling, long polling, CDN requests, or bandwidth. A microservices architecture with many services and regions can create a very large monthly request footprint long before engineering notices it on paper.
Service connections / backend SDK activations
Medium riskSome platforms charge by backend SDK instance or service connection. Container-based deployments with horizontal scaling multiply counts automatically. LaunchDarkly, for example, publishes $12 per service connection per month.
Logs, flag eval events, experiment metrics
Medium-High riskPlatforms increasingly charge for evaluation data, event export, experimentation metrics, or log volume. This grows with product traffic and rollout sophistication rather than with headcount.
Enterprise add-ons
High riskSSO, audit logs, RBAC, data export, and uptime SLA guarantees are commonly gated behind enterprise tiers. Once your InfoSec team requires them, cost jumps to an enterprise tier — often a step function from $1,500/month to $5,000–10,000+/month.
The Cost of Unpredictability
Budget overruns from SaaS metered billing have two costs: the direct dollar amount and the engineering management overhead of explaining, justifying, and re-forecasting the overage.
Common scenario: A team budgets $2,000/month for feature flags. MAU doubles during a promotional campaign. The SaaS bill jumps to $4,200. Finance flags it. Engineering spends 4 hours investigating, forecasting, and justifying. The ops overhead on a billing surprise often exceeds the cost of the overage itself.
In a self-hosted model, traffic spikes hit your infra bill (Postgres queries and FeatBit service CPU), not your vendor bill. Infra overages are visible earlier, scale more gradually, and are entirely within your engineering team's control and forecasting ability.
Self-hosted Cost Model: Fixed by Architecture
A self-hosted model eliminates all per-unit billing. Your annual cost has three fixed-ish lines:
Cloud VM/container cost grows only when you add regions or materially increase traffic — not when vendor billing adds a new meter.
Un-automated: 6–15h/mo. Fully automated: 2–4h/mo — mostly review-and-confirm. With AI coding agents skills and cloud MCP servers, routine tasks can drop further, well below 2h/mo. Adding 20 engineers doesn't meaningfully change ops hours.
Community Edition is free. FeatBit Enterprise starts at $3,999/year, which is still predictable and does not change with seats, MAU, requests, or service connections.
Budget Planning Framework
Use this framework to produce a defensible annual budget for self-hosted feature flags:
FAQ
What if infrastructure costs spike unexpectedly?
Unlike SaaS billing spikes, infra cost spikes are visible in real time via cloud cost dashboards, alert on budget thresholds, and are entirely within your architecture's control. You can scale down resources within minutes, add caching, or tune queries — options not available when SaaS metered billing runs over.
Does self-hosting expose us to more SLA risk?
FeatBit's application services are fully stateless — With managed Postgres (RDS Multi-AZ, Database for PostgreSQL HA, Cloud SQL HA) and a Redis cluster with replica failover, FeatBit can support 99.999%-tier SLA targets. Self-hosting lets you exceed that ceiling at infrastructure cost, not license cost.
How do we forecast infra cost as we scale?
FeatBit's infra cost scales primarily with Postgres write volume and the request rate handled by its self-host services — both directly tied to flag change frequency and active SDK connections. For teams not making hundreds of flag changes per hour, infra cost is nearly flat between 10 and 500 engineers.