01
Billing, recoverable
The provider billed more than the rate card. Reconciliation works out what the bill should have been, token by token. Accepted findings become refund claims.
AI Spend Audit
Y o u r A I b i l l g r e w 1 0 x . Y o u r v i s i b i l i t y d i d n ' t .
See where your AI spend goes across every provider, cut the waste, and recover overcharges where we find them. Connected to Anthropic and OpenAI, with more coming.
Read-only API keys. No SDK, no code changes. Free to run.
AI Spend agent
Audited spend
$148,200
last 30 days
Annual savings
$214,000
18% of run-rate
Recovered
$4,120
3 claims filed
Open violations
11
ranked by impact
Top findings by annual impact
Summarization on oversized model Efficiency $4,200/mo
Prompt cache misses, prod-api Efficiency $2,800/mo
Untagged keys, no owner Anomaly $900/mo
Provider billed above rate card Billing $310/mo
Daily cost last 30 days
Cost by
Claude Sonnet 62% $91,884
GPT-4o 21% $31,122
Claude Opus 12% $17,784
GPT-4o mini 5% $7,410
Anthropic 74% $109,668
OpenAI 26% $38,532
prod-api 48% $71,136
batch-jobs 23% $34,086
eng-sandbox 19% $28,158
untagged 10% $14,820
Reconciliation compares list price against the cost the provider reports. Genuine overbilling is rare today and grows as we connect your invoices.
Provider billed above rate card 1 finding $310/mo
Model substitution, summarization 6 findings $4,200/mo
Prompt cache misses, prod-api 14 findings $2,800/mo
Context inflation, chat workload 3 findings $1,150/mo
Workspace concentration 1 finding $640/mo
Retry storms, agent loops, and duplicate calls.
Coming soon.
Action plan ranked by monthly impact
Move summarization to a smaller model $4,200/mo
Fix prompt caching on prod-api $2,800/mo
Consolidate and tag stray API keys $900/mo
Claims in flight
2
submitted
Pipeline value
$3,200
under review
Confirmed credits
$4,120
applied
Claim lifecycle, Anthropic billing finding
1 Draft
2 Submitted
3 Acknowledged
4 Accepted
Recovery is the edge we are building. We file the claim on your behalf and track it to credit.
- Anthropic + OpenAI
- connected today, more coming
- Daily
- sync, always current
- Read-only
- API keys, no code changes
- Free
- to run, recover what is real
The problem
Your provider dashboard shows usage, not waste.
AI bills depend on the model, prompt size, caching, and input versus output tokens, none of which finance can see. The waste, runaway scripts, cache misses, oversized models, leaked keys, and billing errors, shows up at month-end, after the money is gone.
Our Anthropic bill was $50K last month. I can't tell you which team, which workload, or which model drove it.
How it works
Connect, audit, act.
We surface the recommendation; your team implements it. We never touch your code or your accounts. For billing findings you accept, we file the claim on your behalf.
-
Step 01
Connect
Paste read-only admin API keys for Anthropic and OpenAI. Keys are encrypted. No SDK, no code changes.
-
Step 02
Audit
The agent pulls usage and cost, reconciles it against the provider's billing, runs the detectors, and keeps syncing daily.
-
Step 03
Review
Findings appear as violations and anomalies, each with a dollar impact and a plain explanation. You accept or dismiss each one.
-
Step 04
Recover and act
Accept a billing finding and we file the refund claim with the provider. Accept an efficiency finding and it joins your Action Plan.
What it catches
Every finding, a dollar figure and a plain explanation.
Each one carries an impact, a severity, and an explanation a non-engineer can read. You accept or dismiss every finding.
02
Model substitution
A workload running on a heavier, pricier model than the job actually needs.
03
Prompt caching
Prompts that should hit the cache and don't, cache writes paid for without matching reads, and low cache-hit rates.
04
Context inflation
Prompts growing across turns and inflating per-request cost. OpenAI today.
05
Workspace concentration
A single workspace or project taking an outsized share of spend, so the cost owner is obvious.
06
Anomalies
Off-hours and weekend bursts, untagged keys with no owner, and stale credentials still spending.
07
Advisory
Likely flat-fee seats the API can’t see, day-level spikes, and effective discount versus list. Context, not a claim.
08
Execution
Retry storms, agent loops, and duplicate calls. Coming soon.
Two kinds of money
Savings and recovery are not the same thing.
We track them as separate numbers, because they work differently and you should know which is which.
Future spend you cut
We surface the recommendation, ranked by dollar impact. Your team implements it. We do not change your code or your accounts. This is where most of the value is today.
Past overbilling, clawed back
When you accept a billing finding, we file the refund claim with the provider and track it to credit. Genuine overbilling is rare until we ingest your actual invoices, so we are honest: recovery grows over time.
The differentiator
The Action Plan your engineers can actually run.
From your accepted findings, we build a prioritized brief: each recommendation ranked by urgency, with the monthly dollar impact, the steps to take, and an owner.
Your team tracks each item from pending to complete and rates how accurate it was, which tunes the next plan. This is the report you hand to engineering.
Action plan
- Move summarization to a smaller model $4,200/mo
- Fix prompt caching on prod-api $2,800/mo
- Consolidate and tag stray API keys $900/mo
Why us
Ramp and Brex stop at visibility. We cut the waste, too.
They sell you a dashboard of your AI spend. We do that, then cut the waste on top, and recover overcharges where they are real.
AI spend
Visibility, savings, and recovery on Anthropic and OpenAI.
This agent Ad spendRecover overcharges and block invalid traffic on Google and Meta.
AI Ad Audit ShippingLate fees, dimensional weight, and rate creep on every carrier.
AI Shipping Audit Pricing
Free to run. You keep what you save.
Visibility and savings recommendations are free with Dash.fi. For recovery, we only earn a success fee on credits we actually win back from the provider.
FAQ
Questions, answered straight.
Anthropic and OpenAI today, with more coming. A few detectors are provider-specific because of what each API exposes.
Your fastest-growing cost deserves more than a dashboard.
Connect Anthropic and OpenAI with read-only keys and see where the money goes, what you can cut, and what you are owed.
- Read-only keys
- No code changes
- Free to run