Resolution agents shouldn’t cost more than the ticket. Find the loops and handoffs that do.
Audon reads your OpenAI, Anthropic and xAI (Grok) cost data by project, workspace and model, flags the days and queues where spend broke pattern, and connects cost to resolved tickets when you send outcome telemetry. A fix counts as a saving only after a measured experiment.
Free 30-day audit. Read access only, no card.
Observed spend
$21,627.76
30 of 30 days
Findings
4
1 anomaly to review
Largest project
64.8%
resolution-agent
Verified savings
$1,539.90
1 approved experiment
- resolution-agent64.8%$14,018.94
- escalation-summary16.3%$3,540.27
- triage-classifier14.8%$3,206.40
- qa-sampling3.9%$862.15
Mapped as one project per agent in the ticket flow.
Support agents fail quietly. The ticket closes; the bill doesn’t.
Retry loops on resolution agents
A tool call to the order system times out, the agent retries, the classifier step retries again. The helpdesk shows a resolved ticket. The provider bill shows five attempts.
The spike detector flags days above 2.5× the trailing 14-day median for each project and model. With outcome telemetry, the failed-run detector estimates the cost of runs whose failure was verified, with average retries per run, as a potential saving, not a confirmed one.
- spend_spike
- failed_run_spend
- Outcome telemetry
Escalation handoffs that re-run the whole context
When a conversation moves from the bot to tier 2, the summariser often re-reads the full thread, attachments included. Long tickets pay for the same context twice.
Give the escalation step its own project and you see what handoffs cost per day and per model. Sustained growth flags it when that cost rises over 20% a week for three weeks. A fix goes through a proposal, approval by a second person and a production experiment.
- Breakdown by project, workspace and model
- sustained_growth
- Verification workflow
Ticket spikes after releases
A release, an outage or a billing change sends volume up for two days. Some of that spend is real demand; some is the agent struggling with a question type it hasn’t seen.
Spikes are labelled anomalous spend, never waste, because billing data alone can’t tell demand from a loop. Your team notes the release on the finding, and the activity log keeps who said what.
- spend_spike
- Evidence class on every amount
- Hash-chained activity log
Cost per resolution nobody can quote
Leadership asks what an automated resolution costs compared with a human one. An honest answer needs the bill and the outcome together.
Send outcome telemetry (run id, status, verification, cost estimate; no transcripts) and Audon reports cost per verified successful task. Until then, the outcome data gap finding says plainly that the number is unknown.
- Outcome telemetry
- outcome_data_gap
A month of a support stack, audited. Synthetic, but shaped like the real thing.
- FND-0421High
Spend spike on project resolution-agent
Anomalous spend$1,729.24 above median12–13 Oct at 3.4× and 2.9× the trailing 14-day median · 412 cost records
- FND-0422Medium
Spend on failed runs in workflow refund-request
Potential saving$1,184.30 estimated214 of 1,032 verified runs failed (20.7%) · 2.8 retries per run on average
Uses outcome telemetry (paid plans)
- FND-0423Medium
Sustained spend growth on project escalation-summary
Observed spend$493.22 weekly increaseWeekly spend $612.40 → $748.90 → $912.35 → $1,105.62
- FND-0424Info
64.8% of attributed spend is on project resolution-agent
Observed spend$14,018.94 project totalShare of attributed spend over 30 days · 4 projects with attribution
FND-0421
Spend spike on project resolution-agent
- Classification
- Anomalous spend
- Amount
- $1,729.24 above median
- Detector
- spend_spike 1.0.0
- Status
- Awaiting approval
Detected by audit26 Oct
Excess over median $1,729.24 on resolution-agent.Note from the support ops lead26 Oct
Release 4.12 changed the order-lookup tool. Timeouts triggered up to 5 retries per ticket.Proposal: cap retries at 226 Oct
Rollback: restore max_retries=5. Awaiting approval by a second person.
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.
Support operations
analystToday
Sees ticket volume and handle time, but not what each automated resolution costs.
With Audon
- Cost per agent in the ticket flow, daily
- Spikes tied to releases with notes
- Cost per verified resolution with telemetry
Engineering lead
adminToday
Hears about retry storms from the invoice, weeks after the release.
With Audon
- Spike findings the day after they happen
- Failed-run cost by workflow
- Fixes proven by experiment, with rollback noted
Finance
billing_adminToday
Gets one AI line item with no link to tickets handled.
With Audon
- Spend per support agent and model
- Monthly spend alerts under budget
- Savings reported only once verified
Your first audit takes about fifteen minutes. Thirty days of real spend, read-only.
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.
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.
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.
Built to stay out of your customers’ conversations. Cost data in, transcripts never.
No transcripts, ever
Connectors read cost and usage reports. Telemetry carries run ids, statuses and cost estimates, and rejects metadata keys that look like messages, content, emails or phone numbers.
A second person approves changes
Whoever proposes a fix to a resolution agent can’t approve it. Every note, approval and result is in a hash-chained activity log.
Experiments with quality guards
A change only becomes a verified saving if the success rate drops by no more than 2 points and p95 latency rises by no more than 20%.
Read-only access to the provider
The admin key reads billing data only. It is encrypted with AES-256-GCM, never shown again, and you can revoke it at the provider any time.
If your support data is subject to privacy law, check your own obligations: Audon helps you keep transcripts out of cost analysis but doesn’t make you compliant with any regulation. Nothing on this page is legal advice. Read the security overview and privacy policy for the details.
Questions support teams ask before connecting.
Do you read our tickets or chat transcripts?
How do we get cost per resolution?
A spike was a real ticket surge. Will it be counted as waste?
Can it stop a retry loop in real time?
Which helpdesks do you integrate with?
What’s free?
Find out what a resolution really costs. Start with last month’s spend.
Free 30-day audit. Read access only, no transcripts, revoke any time.