Skip to content
Audon

Know which client’s agents eat the margin. Before the retainer invoice goes out.

Audon reads your OpenAI, Anthropic and xAI (Grok) cost data by project, workspace and model, so each client engagement has its own spend line, its own alerts and evidence you can put in front of the client.

Free 30-day audit. Read access only, no card.

SpendDemo agencySynthetic example
27 Sep – 26 Oct 2026

Observed spend

$14,935.65

30 of 30 days

Findings

4

1 anomaly to review

Largest project

42.9%

client-atlas

Verified savings

$1,063.42

1 approved experiment

Daily spend, all projectsObservedSpike window
Spend spike on project client-cobalt+$339.62
By projectShare · amount
  • client-atlas42.9%$6,412.80
  • client-birch33.3%$4,980.25
  • client-cobalt14.9%$2,226.70
  • internal-rnd8.8%$1,315.90

Mapped as one project per client engagement.

One provider bill, many clients. Splitting it fairly is where margin goes.

  1. Per-client attribution

    One OpenAI organization, eleven clients, one invoice. Splitting it by hand each month means guesses and awkward conversations.

    Give each client or engagement its own project or workspace and Audon breaks spend down by client, model and day. Export the totals to CSV for your billing system.

    • Breakdown by project, workspace and model
    • Report and data export
  2. Margin on fixed-fee retainers

    The fee is fixed; the model bill isn’t. A client whose usage triples halfway through the month quietly eats the margin on that retainer.

    Sustained growth flags a client project growing over 20% a week for three weeks. Daily and monthly spend alerts per project tell the account lead in time to talk about scope.

    • sustained_growth
    • Daily spend alerts
    • Monthly spend alerts
  3. Rebilling evidence clients accept

    Pass-through AI costs get questioned, and “our records say” is weaker than the provider’s own numbers.

    Every amount reconciles to the provider’s cost API and is labelled observed spend, with the record count behind it. Recent days are marked provisional until the provider finalises them.

    • Evidence class on every amount
    • Report and data export
    • Detector versions and snapshot hash
  4. Runaway spend on prototypes and evals

    Pitches, prototypes and eval runs use the agency’s own key. Nobody bills them, and nobody notices when one runs all weekend.

    The spike detector flags any project, including internal ones, above 2.5× its trailing median; the concentration detector flags one model or project dominating spend.

    • spend_spike
    • spend_concentration

One month across four engagements. Which client to call, and which to rebill.

A synthetic example for an agency with one project per client. The classifications are the ones the real detectors assign.
FindingsDemo agencySynthetic example
  • FND-0607Medium

    Sustained spend growth on project client-birch

    Observed spend$644.20 weekly increase

    Weekly spend $802.10 → $975.60 → $1,188.45 → $1,446.30

  • FND-0608High

    Spend spike on project client-cobalt

    Anomalous spend$339.62 above median

    12–13 Oct at 4.2× and 3.3× the trailing 14-day median · 44 cost records

  • FND-0609Info

    78.2% of attributed spend is on model claude-sonnet-4

    Observed spend$11,694.61 model total

    Share of attributed spend over 30 days · 3 models with attribution

  • FND-0610Info

    No outcome data: cost per successful task is unknown

    Observed spendUnavailable cost per success

    $14,935.65 of spend with no matching outcome telemetry

FND-0607

Sustained spend growth on project client-birch

Classification
Observed spend
Amount
$644.20 weekly increase
Detector
sustained_growth 1.0.0
Status
Open
  1. Detected by scheduled audit26 Oct

    client-birch weekly spend up from $802.10 to $1,446.30.
  2. Note from the account lead26 Oct

    Client added two new content feeds in week 2. Outside the retainer scope.
  3. Export attached to the scope review26 Oct

    CSV of client-birch spend by day and model, 27 Sep – 26 Oct.

Synthetic example: generated for this page, not a customer result. Anomalous and observed amounts are not savings, potential savings are not validated, and nothing becomes a verified saving without an approved experiment.

What changes for each role. Same data, three different questions answered.

Finance

billing_admin

Today

Splits one provider invoice across clients by estimate, then defends it.

With Audon

  • Spend per client project, reconciled
  • CSV exports for rebilling
  • Provisional days marked until final

Account lead

read_only

Today

Finds out a retainer lost money at the monthly review.

With Audon

  • Growth on their client flagged within weeks
  • Alerts before the retainer budget is gone
  • Evidence to bring to a scope conversation

Delivery and engineering

analyst

Today

Runs prototypes on the agency key with no view of what they cost.

With Audon

  • Spikes on internal projects caught next day
  • Per-model spend to choose defaults
  • Fixes proven by experiment on paid plans

Your first audit takes about fifteen minutes. Thirty days of real spend, read-only.

Free, with no card. The report stays in the app and exports to CSV or JSON.
  1. About 2 minutes

    Connect a key (read access only)

    Paste an OpenAI organization admin key, an Anthropic Admin API key or an xAI management key. We validate it against the cost and usage endpoints before saving it, encrypted.

  2. Usually under 10 minutes

    Import and reconcile 30 days

    Daily cost records by project, workspace and model, reconciled against what is already stored. Recent days stay provisional until the provider finalises them.

  3. Detectors run in seconds

    Read the report

    Six versioned detectors run on a hashed snapshot. You get an executive summary, coverage, spend profile, findings with evidence, next steps and the method.

Client work stays client work. We see what it cost, not what it said.

  • No client content

    Connectors read cost and usage reports only. Prompts, outputs and client documents never reach Audon.

  • Evidence that reconciles

    Amounts come from the provider’s organization cost API and are reconciled daily. Revisions on recent days update in place rather than double counting.

  • Finance-only roles

    billing_admin handles plans and invoices; read_only lets an account lead see spend without touching connections or alerts.

  • A log of who did what

    Notes, alerts, proposals and approvals go into a hash-chained activity log you can export alongside the data.

Whether and how you can rebill AI costs depends on your client contracts. Audon gives you the evidence, not the contractual right. Nothing on this page is legal advice. Read the security overview and privacy policy for the details.

Questions agencies and consultancies ask before connecting.

Can we split spend by client?
Yes, when each client maps to its own OpenAI project or Anthropic workspace. If several clients share one project, the provider data can’t separate them, so we show that spend as one line.
Can a client log in and see only their own spend?
Not today. Access is per organization, so a client user would see every project. Share an export instead.
Will the numbers match the provider invoice?
Observed spend comes from the provider’s organization cost API for complete UTC days. Recent days are provisional and are re-reconciled on every sync. Taxes, credits and invoice-level adjustments can still differ.
Can we use it across several provider organizations?
Yes, within your plan’s connection limit. The free audit includes up to two connections.
Is a spike on a client project billable?
That’s your call and your contract’s. We label it anomalous spend with its evidence; we don’t judge whether it was waste or demand.
What does the free audit include?
One 30-day audit with the full report and export. Monitoring, alerts beyond one rule, and the verification workflow are on paid plans.

Know your margin per client. Before the next invoice goes out.

Free 30-day audit. Read access only, no card, revoke any time.