AI Spend Control Plane
Meter every token. Cap every dollar. Block no one.
One gateway across Claude, ChatGPT, Gemini, and Bedrock. Per-employee budgets that degrade gracefully to cheaper models — your team never hits a wall.
AI spend · October
$48,211
Quota event: Engineering hit 90% of premium allowance — auto-routed to efficient models. Zero downtime.
The problem
Four gaps no provider dashboard will close
Organizations are spending on AI faster than they can track it. These are the gaps SpendFriend was built to close.
18%
Nobody can attribute it
AI spend arrives on three rails — SaaS seats, API tokens, and cloud-hosted infra. Roughly 18% of it can't be tied to any team or outcome.
11%
Nobody can forecast it
Only ~11% of businesses can accurately forecast AI spend. Classic cloud FinOps tools weren't built for LLM economics.
32%
Cheaper isn't cheaper
In ~32% of enterprise use cases, the cheaper model costs more — once retries, longer chains, and failed first-pass answers are counted.
78%
It's hiding in other tools
Usage-priced AI add-ons ship inside SaaS IT already pays for — up to 78% of IT leaders report surprise charges from these hidden features.
How it works
From scattered AI spend to one controlled ledger
01
Issue virtual keys
Every employee gets a per-person key that maps to provider credentials held in your vault. They paste it into any OpenAI-compatible tool — or use the built-in chat.
02
Meter every request
Tokens, cost, latency, and cache hits land in one ledger — whether the request came through our dashboard or a passthrough API key in someone's own tool.
03
Enforce budgets, never block
Set allowances at org, team, or employee level. When premium quota runs out, requests degrade to mid-tier or free models instead of stopping.
04
Optimize on real numbers
The routing advisor replays your prompt history and prices cost per successful outcome per model — so every routing change comes with a dollar figure attached.
Metering
Every request metered — whatever surface it came from
Requests flow through the SpendFriend gateway — via the built-in dashboard chat, or per-employee passthrough keys your team pastes into the tools they already use. Token counts, cost, latency, and cache hits append to one ledger.
Metering is metadata-first: prompt and response bodies are stored only if your org opts in.
Daily consumption · all providers
Refreshed dailyBudgets & Fallback
Cap the spend. Never block the work.
Set premium allowances at org, team, or employee level — daily, weekly, or monthly. When quota runs out, requests don’t stop: they degrade transparently to mid-tier or free models, mid-conversation.
Employees see a badge — “Running on efficient mode” — and keep going. Admins keep hard-block available as the exception, not the rule.
Engineering · monthly premium allowance
Efficient mode$4,982 / $5,000
Quota crossed → transparent downgrade mid-conversation. The employee keeps working; the budget holds.
Routing Advisor
The cheapest model is the one that finishes the job
Cost per request lies. The advisor replays your actual prompt history and measures cost per successful outcome per model — retries, long chains, and failed first passes included.
Recommendations arrive with real numbers — “route this workload, save $Y/mo” — and nothing changes until an admin approves it.
Cost per successful outcome · support drafts
Recommendation: route “support drafts” to Sonnet — est. save $4,120/mo.
Awaits admin approvalConsolidation
Three rails of AI spend. One ledger.
Gateway-metered API tokens, SaaS seat billing, and cloud spend from Bedrock and Vertex — normalized to the same org, team, and workflow so anyone who needs the number sees the same number.
Finance, IT, and engineering stop reconciling partial spreadsheets that each claim to be the AI budget.
Total AI spend · one ledger
$48,211
Every spend source normalized to org → team → member → workflow, so the ledger answers “what does the support team cost” no matter which rail the spend arrived on.
Leak Audit
Find the spend hiding in context windows
Provider logs show what was spent, not why. SpendFriend tracks cache-hit rates and context size per request, then flags the agents stuffing un-cached blocks into context windows.
You get a per-application leak report quantifying wasted spend — not just a spike on a chart.
Context & cache leak report · this week
2 leaks founddoc-qa re-sends a 90k-token spec uncached on every call — $1,870/mo wasted.
Shadow AI
Catch the AI features hiding inside tools you already pay for
Salesforce, Zoom, Notion, and Jira all ship usage-priced AI add-ons. A lightweight browser extension flags billable AI consumption outside sanctioned channels and feeds it into the same ledger.
It observes request patterns — never prompt content — and hands IT a review queue, not a surveillance feed.
Shadow AI candidates · pending IT review
Observed by the browser extension — usage patterns only, no prompt content captured.
Unit Economics
Turn tokens into a P&L line
Meter events carry workflow tags — ticket resolved, contract reviewed, asset generated — so reporting computes cost per outcome, benchmarked against the human labor it replaced.
“AI spend went up” becomes “support cost per ticket dropped 38%.”
Cost per outcome · workflow tags
Tags attach to meter events — token totals become a P&L line finance can actually read.
Governance
Enterprise controls, built in — not bolted on
SSO / SAML + SCIM
Provision employees into your org → team hierarchy from your IdP. RBAC for admins, team managers, finance viewers, and employees.
Virtual keys & revocation
Employees never touch provider credentials. Per-person keys make offboarding and incident response instant — one revoke, one person.
Policy & audit trail
Budgets, request-class rules, and routing policies are versioned and audit-logged, because procurement and compliance will ask.
Privacy-first metering
Token counts and cost are logged by default; prompt bodies only if your org opts in. Custom endpoints and DLP hooks on Enterprise.
One control plane for finance, IT & team leads
For Finance
A continuously refreshed ledger across every provider, cost per workflow outcome, and forecasts built on actuals — not invoice archaeology.
See finance solutionsFor IT & Platform
One gateway, per-employee virtual keys, budgets at every scope, and instant revocation — with an audit trail on every policy change.
See IT solutionsFor Team Leads
Set allowances your team can't blow through and they'll never feel — premium quota exhausts into cheaper models, not dead ends.
See how fallback worksSee every AI dollar. Control where it goes.
Connect the gateway in minutes. Per-seat platform pricing plus metered consumption — and employees who never hit a wall.