Example report (sample data)
Sample burn audit · What the $150 hour hands back
A sample burn-audit report, start to finish
This is what you get after the burn-audit hour: a written map of every bot and scheduled routine, where the usage goes, what to cut, and how to keep critical workflows running when a seat runs dry.
Need the ledger and board letters without operator time? Burn Book is a $750 board packet, paid in full. The packet does not include someone making the cuts.
Read this first. The company, the bots and every number on this page are made-up example data, written to show the format. None of it is a client result, a benchmark or a promise of savings. Your report uses your own plan, your own routines and your own usage.
1. The team in this example
The problem as stated in the intake (example): "We keep buying top-offs, nobody knows which bot is spending it, and our Monday board report didn't run last week because the seat that owns it ran out."
2. The jobs and routines map
We list every bot and scheduled routine, who owns it, what wakes it up, how often it runs and roughly how much of the usage it takes. The sources are your plan's usage and billing pages, each bot's routine list and run history, and a short walk-through with the owners.
| Job | Bot / owner | Trigger | Runs / day | Avg min / run | Share of usage | Verdict | Why |
|---|---|---|---|---|---|---|---|
| Inbox triage | Ops bot | Polls mail every 10 min | 144 | 1.5 | 14% | Rebuild | Wake on new mail instead of polling. Cheap pre-check before the agent runs. |
| Competitor monitor | Research bot | Hourly | 24 | 6 | 18% | Cut to daily | Nobody reads it hourly. One morning digest, with an alert only for named competitors. |
| Social listening | Marketing bot | Every 15 min | 96 | 1 | 7% | Merge | Covers the same sources as the competitor monitor. Fold it into that digest. |
| Morning sales digest | 3 people, 3 copies | 8:00 AM daily, x3 | 3 | 9 | 9% | Merge | Three reps built the same digest. Keep one shared copy with an owner and send it to all three. |
| CRM hygiene | Ops bot | Every 30 min | 48 | 2 | 8% | Rebuild | Trigger on record change, and batch the cleanup once a night. |
| Exec research thread | Assistant bot | Ad hoc, one long thread | 12 | 14 | 22% | Rebuild | Each run re-reads a thread of 200+ messages. Start runs from a short state file and archive the history. |
| Support ticket summaries | Support bot | On each new ticket | 40 | 3 | 15% | Keep, trim | Already event-driven. Send only the ticket and the last 5 replies, not the whole account history. |
| Weekly board metrics | Finance bot | Mondays 7:00 AM | 0.14 | 25 | 2% | Keep | Low cost and high value. Add a fallback owner (see below). |
| Everything else (11 small routines) | Various | Mixed | n/a | n/a | 5% | Review | Fine for now. We look at these again in 30 days. |
Swipe the table sideways to see every column.
"Share of usage" is estimated from run counts, run length and how much context each run loads. Usage pages rarely break spend down per routine, so we label it as an estimate in your report too.
3. Where the usage goes
All shares are example data.
What the map shows (example)
- About two-thirds of usage goes to four jobs, and only one of them (support summaries) is something the team reads the moment it lands.
- Three jobs poll on a timer (every 10, 15 and 30 minutes), and most of those runs find nothing new. Polling pays for the "nothing" runs too.
- One long thread re-reads its whole history on every run. As the thread grows, each run costs more, even when the question is small.
- The same digest exists three times because three people built their own.
- The board report depends on one person's seat. When that seat runs out of usage, the report doesn't run.
4. The cut list, ranked
Ranked by estimated monthly effect against the build time it takes. Each line names the change, not just the problem.
| # | What's burning | The fix | Est. monthly effect | Build time |
|---|---|---|---|---|
| 1 | Hourly competitor monitor and 15-minute social listening | Merge them into one 7:00 AM digest, with an instant alert only for named competitors. | $520Example | 1.5 h |
| 2 | Exec research thread re-reading its full history | Start each run from a one-page state file (decisions, open questions, sources). Archive the old thread, keep the stable instructions first so caching can help, and cap the history window. | $330Example | 1.5 h |
| 3 | Inbox triage polling every 10 minutes | Switch to a new-mail trigger. Where that isn't possible, run a cheap check that wakes the agent only when something changed. | $260Example | 2 h |
| 4 | CRM hygiene every 30 minutes | Trigger on record change, and move bulk cleanup to one nightly batch. | $150Example | 1.5 h |
| 5 | Three copies of the same morning digest | One shared routine with a named owner. Deliver it to all three reps. | $140Example | 0.5 h |
| Total | Five changes, about 7 hours of build work | $1,400Example | 7 h |
Swipe the table sideways to see every column.
The estimated effect is example data, worked out as each job's share of the $3,500 example spend, minus what the rebuilt version still uses. In a real report it's an estimate you can check against next month's bill, not a promise.
What we deliberately don't cut
- Support ticket summaries. They're used right away. We trim what each run loads instead of cutting runs.
- The weekly board metrics. It's cheap and important, so it gets a fallback rather than a cut.
- Quality. Every change is checked against a few real past runs before it replaces the old routine, so a cheaper version that gives worse answers doesn't ship.
5. Fallbacks for scheduled workflows
When a scheduled workflow lives on one person's seat, it stops when that seat runs dry or that person leaves. The report gives each critical routine a written fallback:
| Routine | Owner | Backup | Budget alert | If it doesn't run | Low-budget mode |
|---|---|---|---|---|---|
| Weekly board metrics | CFO's seat | Ops lead's seat, same routine saved and paused | At 70% and 90% of the seat's monthly usage | A heartbeat check at 7:30 AM Monday alerts the ops lead | Numbers only, no written commentary |
| Support ticket summaries | Support lead | Second support seat | At 80% | Tickets still route normally, just without a summary | Summarize priority tickets only |
| Morning sales digest | Sales ops | Any rep's seat | At 80% | Skip that day and post a one-line notice | Headlines only |
Swipe the table sideways to see every column.
- Owner of record. Every routine has one named owner and one backup, written down.
- Budget alerts before top-offs. Alerts at set thresholds, plus a short rule on who may buy top-offs and when.
- A heartbeat for critical jobs. A small check that alerts someone when a routine that should have run didn't.
- A run log. One place showing what ran, how long it took and what it produced, reviewed weekly for the first month.
6. Next steps
Do it yourself
The report is written so your team can make every change on the list without us. Plenty of teams take it from here.
$1,250 operator pack
Ten hours of build and fix work. In this example: about 7 hours for the five cuts and 3 hours for fallbacks, alerts and the run log. Checkout is a $625 deposit.
AI Concierge, $2,000/mo
We keep it running: a monthly burn review, new routines checked before they ship, and two 45-minute sessions a month with async help in between.
How the hour works
- Pay the $150 and pick a time. Checkout is a secure Stripe page. We schedule the session after payment.
- Bring read access to the usage page and a list of your bots and routines. Screenshots are fine. We never need passwords, and we don't change anything during the audit.
- One working hour together. We walk the map with whoever owns the bots.
- You get the written report: the map, where the usage goes, the ranked cut list, fallbacks and next steps, in the format on this page.
Get this report for your own team
One hour, $150, and a written map of what's burning your team's AI agent spend and what to do about it. Want a rough number first? Use the button below to run the free burn check.
Questions first? Call the AI front desk demo at (320) 335-6186, or email daniel@agenthiveinc.com.
The practices behind the format
The cut list draws on published guidance for running AI agents efficiently, applied to your team's setup:
- Wake agents on events instead of polling on a timer, and batch work that doesn't need to happen right away.
- Keep stable instructions first and let prompt caching reuse them, and keep long histories from piling up by summarizing older turns and capping the window.
- Load only the context and tools a run needs, and send simple triage to a smaller model where your tools allow it.
- Measure cost per routine, not just the total bill, and check quality before swapping in a cheaper version.
Further reading: Anthropic's cost optimization cookbook and the AWS Well-Architected Agentic AI Lens on token cost and context window use.
More: First Deploy home, pricing, free burn check.
Related: what an AI agent usage audit covers, fallbacks for scheduled AI workflows, AI usage top-offs, context bloat, routine drift and loops.