What the invoice dispute triager costs
Dispute volume is low even in businesses that feel overwhelmed by them — fifteen a month is a normal figure for a firm raising four hundred invoices. The bill reflects that: this is one of the cheapest systems in the series to run, and the value is entirely in the hours it gives back rather than in anything it costs. Here is where each cent goes.
Key takeaways
- About $2 a month at 40 disputes. Roughly $3 at 150 disputes.
- One Bedrock read per dispute is the only line that scales. Everything else is rounding.
- Nothing is always-on, so a quiet month genuinely costs almost nothing.
- The duplicate test runs before the read, so resends are free.
- The three real risks: a retry loop, storage nobody expires, and a bigger model than the job needs.
The bill at three volumes
These are US East prices at the time of writing, at three volumes that bracket most small businesses. Find the bar closest to your own and read across.
Line by line
| Line | At 40 disputes | How it scales |
|---|---|---|
| Bedrock read | $0.29 | Linear. One call per dispute, roughly 1,800 in and 200 out tokens. |
| SES | $0.03 | Linear. About 2.2 messages per dispute. |
| DynamoDB + S3 | $0.29 | Storage grows with what you retain, not with throughput. |
| Lambda + SQS | $0.12 | Linear, and effectively free at this scale. |
| CloudWatch | $0.16 | Flat, if you set retention. Unbounded if you do not. |
| Secrets Manager | $0.40 | Flat. One secret, $0.40 a month. |
| AWS Budgets | $0.46 | Flat. Two actions, so you find out before the bill does. |
The read cost per dispute is higher than in most systems here because the prompt carries the invoice lines as well as the customer message. That is the right trade: it is what lets the classifier identify which line is being disputed, which is what makes the gathering narrow.
The three ways this bill surprises you
Every one of these has happened to somebody, and all three are cheap to prevent.
- Classifying every inbox message. If the cheap structural filters are skipped and every message that arrives gets a model call, a single customer’s out-of-office loop can cost more in a week than a year of real disputes. Auto-reply and remittance detection are free and must come first.
- A retry loop on a malformed attachment. A dispute email carrying a corrupted PDF makes the function throw, and without a dead-letter queue the queue redelivers indefinitely. Maximum receive count of three.
- Log retention left at never. At this volume the logs will out-cost everything else within months. Thirty days of retention is the whole fix.
What it costs when nothing happens
This matters more than the headline number for a seasonal business. In a month with nothing to process the bill is the fixed band: Secrets Manager at forty cents, AWS Budgets at forty-six, and a few cents of storage. Call it a dollar. There is no instance to stop and nothing to remember to turn off.
- Management
- Analytics
- Front-end & mobile
Set these on day one
- A dead-letter queue on every SQS queue, with a maximum receive count of three.
- Thirty-day retention on every CloudWatch log group. There is no default that is safe.
- An S3 lifecycle rule on the object prefix, tiering at 90 days and expiring at your actual record-keeping horizon.
- Two AWS Budgets actions — one that emails at half your expected spend, one at double it. The second is how you find out about a loop in an hour instead of a month.
- Provisioned concurrency: none. Nothing here is latency-sensitive enough to justify paying for a warm function.
Next: the same system drawn for engineers — service names, resource identifiers, IAM scopes, table schemas and the model id.
All posts