Background
A financial research firm came to Dataforge with a familiar problem: AI had become part of daily work, and everyone wanted their own subscription.
Roughly twenty staff each carried a personal Claude or ChatGPT account at about $50 per month. That added up fast—and most of those seats sat idle while a handful of power users burned through their limits.
Business Challenge
Per-user AI subscriptions do not scale when usage is uneven. Twenty accounts at ~$50 each meant roughly $1,000 per month, but only a few people were heavy users on any given day. The rest paid for access they barely touched.
There was no central view of spend, no way to shift capacity to whoever needed it, and no shared guardrails. Staff picked whichever vendor they preferred, and IT had no lever to standardize access or tune models for research workflows.
They wanted one pool of tokens the whole team could draw from—without giving up model choice or locking into a single chat interface.
Solution
Dataforge designed and installed an Open WebUI platform in our datacenter. Open WebUI gives the team a single web interface to every model we connect—frontier models from major providers plus smaller, capable options that are often faster and less expensive for everyday tasks.
Usage draws from one shared token budget instead of twenty separate bills. Administrators control who can access which models, set limits where needed, and customize or fine-tune models to fit how the research staff actually work.
The platform runs on Dataforge infrastructure, so the customer gets uptime, monitoring, and support without standing up their own AI stack.
Implementation
We had the full system running in about three weeks. Dataforge provisioned the Open WebUI environment in our datacenter, connected the model providers they wanted, and configured their favourite models for day-to-day research work.
User accounts and access rules were set up before anyone logged in—power users and occasional askers on the same pool, with permissions matched to role. After a short handoff, staff were off and running in the shared interface.
Project Considerations
Moving from personal subscriptions to a shared platform meant a behaviour shift: one login, one place to pick a model, and usage that counts against the team pool—not an individual cap.
We front-loaded model selection and access tiers so researchers kept choice without twenty parallel accounts. Custom model settings were tuned for their typical prompts rather than generic defaults.
Because the stack lives in the Dataforge datacenter, model keys and usage data stay under managed infrastructure rather than scattered across personal billing accounts.
Environment Comparison
| Previous Environment | Current Environment |
|---|---|
| ~20 separate Claude or ChatGPT subscriptions at ~$50/user (~$1,000/month) | One shared Open WebUI platform at ~$200/month on pooled usage |
| Most accounts idle; heavy users hit individual limits | Single token pool flexes to whoever needs capacity that day |
| No central access control or usage visibility | Per-user permissions and model access managed in one place |
| Model choice locked to each vendor's consumer plan | Multiple providers and custom-tuned models from one interface |
| Personal billing accounts scattered across staff | Platform hosted and monitored in the Dataforge datacenter |
Key Outcomes
- ~$1,000 down to ~$200 per month — twenty per-seat subscriptions replaced with one pooled platform
- 3 weeks to production — Open WebUI provisioned, models configured, users onboarded in the datacenter
- Any model, one interface — frontier and cost-efficient models available without picking a single vendor
- Access under control — shared pool with per-user limits and custom models for research workflows
Long-Term Track Record
Monthly AI spend dropped from roughly $1,000—twenty seats at ~$50 each—to about $200 on the shared system. That held while the team gained access to more models than any single subscription offered, including powerful options that are inexpensive for bulk research tasks.
Idle seats stopped being a line item. When one researcher is in a heavy week and another is quiet, the pool absorbs the difference instead of leaving paid capacity unused.
Dataforge continues to manage the platform: new models, access changes, and infrastructure behind the scenes.