Split cost vs self-hosted GrowthBook: where the money actually goes
If your Split or Harness FME renewal is due, the cost question has a short answer: Harness FME is sold only on a written quote as part of a Harness Enterprise plan, while the self-hosted Open Source edition of GrowthBook carries no license fee. The real comparison is therefore the Harness subscription against your own infrastructure, warehouse compute, paid seats for single sign-on and approvals, migration work and ongoing operations. Harness pricing is not published, so ask for a written quote that states the modules, seat or usage basis, term and what is included, and compare it line by line with the items below.
On the GrowthBook side, the free self-hosted Open Source plan allows 1 project with unlimited users and unlimited API requests, which is often enough to evaluate the tool but rarely enough for a portfolio of teams. Self-hosted Enterprise, quoted separately, adds SSO and SCIM, advanced permissioning, approval workflows, safe rollouts and holdouts. Those are the features that usually drive a paid conversation, so confirm with GrowthBook which of your current Split controls you need before you size the license. For contrast, Unleash lists Pay-As-You-Go at $75 per seat per month with a 5-seat minimum for self-hosted, a per-seat model that scales differently from a quoted platform fee.
Two costs recur whatever the license decision. GrowthBook runs experiment analysis as read-only SELECT queries in your own warehouse, so compute for those queries is billed by your warehouse provider, and the frequency of analyses drives that spend. Infrastructure is also yours: the GrowthBook image, a MongoDB or compatible database (the docs suggest a managed service for production) and file storage all need hosting, patching and backups. Estimate these from your own environment and ask your database and cloud teams to price them. One-time migration work, including rebuilding metrics and running Split and GrowthBook side by side until results match, is the largest variable, and it grows with the number of experiments and custom targeting rules.
Which cost components change between Harness FME and self-hosted GrowthBook
This table compares each cost component of running Split's successor, Harness FME, against self-hosting GrowthBook, separating what is paid to a vendor from what is paid to your own team and cloud providers. Read the one-time and recurring column first, because the largest recurring differences are usually seats for SSO and approvals and the warehouse queries that experiments generate. The main takeaway is that the Harness side is a single quote, while the self-hosted side is spread across several line items you control.
| Cost component | Commercial side (quoted Harness Enterprise module) | Self-hosted side (GrowthBook Open Source or Enterprise plus hosting) | One-time or recurring |
|---|---|---|---|
| Subscription or license | FME sold as a module of a Harness Enterprise quote; ask for the module, unit basis and renewal terms in writing | GrowthBook Open Source is free to self-host; self-hosted Enterprise is a separate paid agreement, so confirm its scope and price with GrowthBook | Recurring |
| Infrastructure | Little or none on your side, since Harness hosts the service | Compute, load balancing and storage for the GrowthBook image, plus a MongoDB or compatible database, often a managed service | Recurring |
| Warehouse query compute | Depends on how Harness handles experiment metrics; confirm where analysis runs and who pays for it | GrowthBook runs SELECT queries in your warehouse, so experiment volume and metric count drive your warehouse bill | Recurring |
| SSO and approval seats | Typically bundled into the quoted tier; confirm which governance features are included | SSO, SCIM, approval workflows and advanced permissions sit in self-hosted Enterprise, not the free Open Source plan | Recurring |
| Implementation | Limited to Harness onboarding and internal configuration effort | Deployment, Helm chart setup, identity integration and metric definitions, either by your team or a fixed-scope partner | One-time |
| Migration | Export flags and rebuild targeting in Harness, or stay put with no move | Export definitions, recreate targeting and segments in GrowthBook, rebuild metrics, swap SDKs and run both systems until results match | One-time |
| Support | Covered by the Harness contract; confirm response terms and escalation paths | Internal on-call or a paid support or retainer agreement, scoped to the platform and its database | Recurring |
| Upgrades and maintenance | Handled by Harness as part of the hosted service | Your team tracks GrowthBook releases, tests upgrades and patches the database and container host | Recurring |
| Internal skills | Admin training on the Harness platform | Staff who can operate Kubernetes or Docker, MongoDB and the warehouse connection, and own experiment statistics | Recurring |
What moves the comparison most for your team
Most of the cost difference between Split and self-hosted GrowthBook comes from a few variables: how many people need paid access, how much warehouse work your experiments create, and how much operational effort your team can absorb. Work through these before you request a quote, because each one changes which GrowthBook tier you need and what the platform costs to run.
Seat count against the Enterprise tier
The free self-hosted Open Source plan allows unlimited users, but SSO and SCIM, advanced permissioning, and approval workflows sit in self-hosted Enterprise. If your company requires single sign-on or approval gates on flag changes, the license cost depends on that feature set rather than on headcount. Ask Harness for a quote that lists the Split modules and seats you use today, and ask GrowthBook for Enterprise pricing for the same feature list. Confirm which of your current Split controls depend on features only available in the paid tier.
Experiments and metrics driving query volume
GrowthBook runs read-only SELECT queries against your own warehouse and sends only aggregated statistics back. Warehouse compute therefore scales with the number of experiments, metrics, and how often results refresh, and that cost lands on your data platform bill rather than on GrowthBook. Count active experiments and metrics per quarter, and ask your data team to estimate warehouse spend for a typical refresh schedule. Split's hosted metrics do not add that line item, so the saving on the license may be partly offset here.
Projects against the free plan limit
The self-hosted Open Source plan allows one project. A company that uses Split for several product areas, business units, or separate applications will likely need more than one project, which pushes the design toward a paid tier or toward consolidating projects. Map your current Split workspaces to the projects you would need in GrowthBook, then confirm the project count with GrowthBook before you commit to a deployment layout. Consolidation changes permissions and naming, so plan that work into the migration estimate.
MongoDB availability and hosting
GrowthBook stores metadata in MongoDB or a compatible database, and the documentation suggests a managed service such as MongoDB Atlas or AWS DocumentDB for production. Choosing a managed database adds a recurring service fee and a dependency on your cloud account, while a self-managed replica set moves backup, failover, and patching onto your team. Decide which option fits your availability target before you size the platform, and ask your cloud or database team to price both paths in your own accounts.
Data residency and warehouse access
Self-hosting keeps the GrowthBook application and its metadata inside your environment, and warehouse analysis keeps raw event data where it already lives. Check whether Split currently stores any experiment or event data in a region or account you must keep, and whether your residency rules cover the aggregated statistics that return to GrowthBook. Compliance review and evidence collection are your team's work, or the work of a partner preparing evidence for an external assessor, and they add effort on top of the license comparison.
In-house skills for upgrades and incidents
Self-hosting makes your team responsible for upgrades, security patches, backups, and incident response for a stack that bundles a Next.js front end, an Express API, and a Python statistics engine. Each GrowthBook release, such as v5.1.0, needs testing against your warehouse connections and SSO configuration before rollout. Ask whether you have staff who can operate Kubernetes or Docker, MongoDB, and your warehouse credentials, or whether you need a retainer for ongoing support and upgrades.
How to build the cost model with your own numbers
This sequence turns a vague renewal comparison into a cost model built from your own seats, warehouse usage, hosting bills and staff time, so both options are priced over the same term. Each step produces a figure you can defend in a budget review.
Inventory current Split seats, modules and environments
Pull the contract, order form and admin console data to record every paid seat type, each FME or Split module you license, the number of environments and projects in use, and the renewal date. Ask the account owner which features are contractually bundled and which are add-ons. The step is done when one sheet lists each item with its unit (seat, module, environment) and its current annual line in your own currency.
Request a written Harness quote for the same scope
Send Harness the inventory from the previous step and ask for a written quote covering the same seats, modules, environments, support level and term, with each line item separated. Harness prices FME as a module of its Enterprise plan on quotation, so ask which components are required to get SSO, approvals and warehouse-connected analysis. The step is done when you hold a quote that names each module and states the term, so you can compare it line by line.
Map warehouse sources and experiment volume
Work with your data team to list every warehouse or database that would hold experiment exposure and metric data, such as BigQuery, Snowflake, Redshift, Postgres or Trino, and estimate how many experiments run each month and how many metrics each one uses. GrowthBook runs SELECT queries against your own data source, so the query load lands on your warehouse account. The step is done when you have a table of sources, monthly experiment counts and metric counts that a cost estimate can reference.
Price GrowthBook self-hosted Enterprise against its pricing page
Compare the free self-hosted Open Source plan with the self-hosted Enterprise tier on the GrowthBook pricing page. Enterprise is needed for SSO and SCIM, advanced permissioning, approval workflows, sticky bucketing, holdouts and safe rollouts, so list which of those your Split usage depends on. Enterprise pricing is custom, so ask GrowthBook for a written quote for your seat count and confirm the license terms in its LICENSE file. The step is done when you know which features force an Enterprise quote and have that quote in hand.
Estimate hosting, database, storage and warehouse compute
Use your cloud bills to price the infrastructure that GrowthBook needs: compute for the bundled Next.js, Express and Python statistics services, a MongoDB or compatible database (managed services are suggested for production), file storage for uploads, and the extra warehouse compute from experiment queries. Tag the usage by month and check whether your team already pays for a database or Kubernetes cluster. The step is done when each infrastructure line has a monthly figure taken from your own billing data, with the assumption behind it written next to it.
Add migration and operations effort
Estimate the internal and partner hours for exporting Split definitions, recreating targeting and segments, running the LaunchDarkly-style importer equivalent or manual rebuilds, swapping SDKs, rebuilding metrics on your warehouse and running both systems side by side until results match. Also estimate recurring operations work such as upgrades, backups, access reviews and on-call coverage. Treat one-time migration work and recurring operations as separate lines. The step is done when each effort line has a named owner and an hourly or day-rate basis, and a DeployOpen or other partner quote, if you request one, follows from the written scope.
Compare totals over the same term and flag dependencies
Place the Harness quote and the GrowthBook totals side by side over the same contract term, with one-time costs and recurring costs in separate columns. Mark every figure that depends on an assumption, such as seat growth, experiment volume, warehouse pricing, Enterprise quote terms or staff rates, and show how the total moves if each assumption changes. Share the model with finance and the data team for review. The step is done when the comparison shows both totals, the dependencies list is complete, and each owner has signed off on their inputs.
When self-hosted GrowthBook costs more than staying on Harness
The total cost rises in three situations. The first is the self-hosted Enterprise tier: the free self-hosted Open Source plan does not include SSO and SCIM, advanced permissioning, approval workflows, holdouts or safe rollouts, according to the GrowthBook pricing page. If your Split or Harness FME use depends on SSO enforcement, SCIM provisioning or approval gates before production changes, you will be buying the Enterprise license, and that price is set by quotation. Ask for it in writing alongside a Harness FME quote for the same user count, seat model and feature list, so both sides are compared on the same scope. The second is operations. The GrowthBook image needs MongoDB or a compatible database, and the docs suggest a managed service for production. If your team has no MongoDB or Kubernetes operators, the cost of hiring, training or buying managed database and cluster services belongs in the comparison. The third is warehouse spend. GrowthBook runs SELECT queries against your own warehouse, so every experiment adds compute on Snowflake, BigQuery or a similar platform. Heavy experiment volume shows up on your warehouse bill, not on GrowthBook's.
The free self-hosted tiers also carry limits that can push costs up. GrowthBook's self-hosted Open Source plan allows one project. If you run several product areas as separate projects in Split, you will either consolidate them, which takes planning, or move to the Enterprise tier. Unleash Open Source allows one project and two environments per instance, according to the Unleash pricing page. A typical setup with development, staging and production, plus a separate project per team, will exceed those limits quickly, and the paid Unleash options are priced per seat with a five-seat minimum for self-hosted. Count your projects and environments today and compare them with each tier's limits before choosing a platform.
Treat these as the checks to run before committing. Export your current Split or Harness FME seat count, project list and experiment volume. Confirm which governance features your change process uses. Price each option against that scope, including Enterprise licensing, the database and cluster run cost, the warehouse query cost at your experiment volume, and internal or partner time for migration and upkeep. Ask each vendor to state which of these are one-time costs and which recur each year. Those answers determine whether staying on Harness is cheaper at your next renewal.
Questions buyers ask before comparing Split and GrowthBook costs
How much does a self-hosted GrowthBook deployment cost to run?
The software itself is free under the MIT license outside GrowthBook's enterprise directories, so the recurring cost is the infrastructure and the people who run it. Your costs come from compute for the GrowthBook image (Next.js front end, Express API and Python statistics engine), a MongoDB or compatible database (the docs suggest a managed service such as MongoDB Atlas or AWS DocumentDB for production), file storage, backups, and staff time for upgrades and monitoring. Warehouse query costs for experiment analysis are billed by your data warehouse provider and grow with experiment volume. Ask your team to price each line against your own cloud accounts, and confirm current license terms in the GrowthBook license file before you commit.
Is the free self-hosted GrowthBook plan enough for our team?
The free self-hosted Open Source plan includes unlimited users, one project and unlimited API requests, so it fits a single product area with a small set of flags and experiments. It does not include SSO and SCIM, advanced permissioning, approval workflows, multi-armed bandits, safe rollouts, ramp schedules, sticky bucketing or holdouts, which the self-hosted Enterprise tier adds. Check your current usage against each of these items. If you rely on approvals or SSO for compliance, the free plan will not cover those controls, and you should price the Enterprise tier from the GrowthBook pricing page or ask GrowthBook for a written quote.
How long does a migration from Split take, and what does it need from the team?
The duration depends on how many flags, targeting rules, segments and experiments you have, and on how many metrics must be rebuilt on your warehouse. The GrowthBook Statsig importer documentation shows the general approach for other platforms, and the LaunchDarkly importer reads flags into a preview before importing them. Split has no equivalent importer in the supplied sources, so expect export and recreation work. Your team will need to export definitions, map treatments and targeting, rebuild metrics against warehouse tables, swap SDKs in the application, and run both systems side by side until results match. Ask any implementation partner for a written timeline based on your inventory.
Can our engineering team do the migration without outside help?
A team with warehouse access, deployment experience and someone who owns the experiment metrics can usually do the work in-house, provided the scope is small. Outside help is most useful for inventory of complex Split setups, metric definitions that depend on warehouse modeling, hardening, backups, and the parallel-run validation that proves results match. The GrowthBook LaunchDarkly documentation warns that advanced setups can include cases the importer cannot handle, and that applications still need to change SDKs after an import, so plan for those steps regardless of who performs them.
What should we request in a Harness quote for Feature Management and Experimentation?
Harness does not publish FME prices in the sources available here, so ask for a written quote that separates each component. Request the FME module price on its own, the Harness Enterprise plan it is sold within, the number of seats or units used for billing, what SSO, approvals and audit features are included, the renewal uplift, and the support tier. Also ask what happens to data and configurations on exit, and whether a migration or export is included. Compare the quote line by line with the self-hosted components described above, and confirm all terms with Harness in writing.
What recurring costs should we budget for after the migration?
The recurring costs are infrastructure, the managed database, warehouse query charges, backups, upgrades and the staff time to operate them, plus any paid GrowthBook Enterprise seats or tiers you choose. One-time costs include the export and recreation work, metric rebuilding, SDK changes, parallel runs and training. Upgrades recur, since GrowthBook publishes new releases such as v5.1.0 on September 22, 2026, and each one needs testing in a non-production environment before rollout. Build a yearly operations line item from these components so it can be compared fairly with a subscription renewal.
Does self-hosting GrowthBook keep our experiment data inside our environment?
GrowthBook analyzes experiments against your own data source and runs only SELECT queries, transmitting aggregated statistics rather than raw event rows. Supported sources include BigQuery, Snowflake, Redshift, Databricks, ClickHouse, Postgres, MySQL and Trino, among others. Your security and compliance teams should still confirm the data flows, network paths, credentials for the warehouse connection and where metadata lives in MongoDB, using the GrowthBook data source documentation as the starting point. DeployOpen can help document and harden these controls for your compliance assessment, but it does not issue audit reports or certifications.
How GrowthBook's components and your warehouse fit into the bill
Self-hosted GrowthBook costs less to start than it looks, because most of its moving parts are one deployment. The front end, the Express API and the Python statistics engine ship in a single Docker image, so the application tier is one container image to run, patch and scale. Metadata sits in MongoDB or a compatible database, and the GrowthBook documentation suggests a managed service such as MongoDB Atlas or AWS DocumentDB for production. That database is usually the largest line of new infrastructure spend, and it recurs every month whether you pay a managed provider or your own platform team. Ask your team which database option fits your hosting standard before you compare totals with Split's quote.
The optional Kubernetes deployment uses the official Helm chart, which adds cluster capacity and platform labor rather than a license fee. Uploaded files can live on local disk, in Amazon S3 or in Google Cloud Storage. Local disk is the cheapest to start but ties backups and high availability to the host, while object storage adds a small recurring storage bill and a configuration task for credentials and bucket policies. Confirm which storage option your backup policy and disaster recovery target require, because that decision changes both infrastructure and operations effort.
The largest cost driver for experiments is your warehouse, not GrowthBook. GrowthBook runs read-only SELECT queries against sources such as Snowflake, BigQuery, Databricks or Postgres, and those queries consume compute in the warehouse you already pay for. Only aggregated statistics leave the warehouse, so raw event data stays under your controls, which matters for data residency reviews. Before you finalize the estimate, ask your data platform team how often experiment queries will refresh, how large the experiment tables are and whether a dedicated compute pool or warehouse size is needed so analysis does not slow other workloads. Those choices, not the GrowthBook license, usually set the recurring query cost.
Which GrowthBook edition and license terms affect the business case
The first decision is which GrowthBook edition the business case assumes, because the license split decides what you can run without a paid contract. GrowthBook's LICENSE file states that code outside its enterprise directories is MIT Expat licensed, while the directories packages/back-end/src/enterprise, packages/front-end/enterprise and packages/shared/src/enterprise fall under the GrowthBook Enterprise License. Self-hosted Enterprise adds SSO and SCIM, advanced permissioning, approval workflows, sticky bucketing, holdouts and safe rollouts, so every Split capability you rely on must be mapped to a feature on the free side or the Enterprise side before you compare totals. Ask GrowthBook in writing which of your required features sit inside the Enterprise scope and whether any licensing terms apply to your deployment model.
Confirm the version you would deploy as well. GrowthBook v5.1.0 was released on September 22, 2026, and the feature set and license file can change between releases, so check the GrowthBook GitHub releases page on the day you finalize the design and record the version in the business case. Plan a review at each upgrade, since a new release can move a capability between editions or change the terms that apply to modified code.
If the fallback option is self-hosted Unleash, the license question is different. Starting with Unleash v8.0.0, the source in the main repository and the unleash-server npm package moved from Apache 2.0 to AGPLv3, while the official open-source Docker images remain under Apache 2.0. Unleash's own guidance says that running the unmodified official images is not affected, but a team that modifies Unleash and offers it as a public or commercial service must publish its source or buy a commercial license. Have counsel confirm whether any internal customization, or any hosting arrangement with a partner, would trigger those obligations before you put Unleash in the cost model.
How DeployOpen scopes a fixed-price GrowthBook implementation
DeployOpen is a Chainscore Labs company that plans, implements, integrates, hardens, documents and supports self-hosted open-source platforms inside your own infrastructure. For a Split or Harness FME replacement, we start with a written scope that separates the GrowthBook deployment, the warehouse analysis work and the migration of flags and experiments, so you can compare it line by line with your current contract.
How we scope
- Inventory every Split or Harness FME flag, experiment, segment and metric that the replacement must cover
- Confirm which GrowthBook edition fits, since self-hosted SSO, approvals and holdouts sit in the Enterprise tier
- List the warehouse sources GrowthBook must query, with the read-only access and aggregate-only output you require
- Agree deliverables, exclusions and timelines in writing before any work starts, as engagements are fixed-scope
How we build
- Deploy GrowthBook in your cloud account or data center, with its metadata database and file storage chosen by your team
- Configure SSO, roles and approval workflows to match your existing governance rules and access policies
- Build warehouse data source connections and metric definitions, then validate them against results you already trust
- Test the self-hosted deployment, backups and upgrade path before any production traffic depends on it
How we hand over
- Document architecture, configuration, data source connections and the runbook for routine operations
- Provide a parallel-run plan that compares GrowthBook and Split results until your analysts accept the differences
- Hand over SDK swap steps for each application, with a rollback path for every flag change
- Deliver the evidence pack your internal or external assessors need for change control and access reviews
How we support
- Ongoing support, upgrades and continuous testing are available as a retainer after handover
- Retainer scope, response expectations and upgrade cadence are agreed in writing and priced from that scope
- Ask us to confirm what each quote covers, including warehouse query costs that stay in your own data platform
- We prepare systems and evidence for compliance assessments, but independent auditors issue any audit reports
