What the review responder costs

The fixed cost is essentially zero

There is no per-listing licence, no minimum monthly fee, no “starter plan.” If your listings have a quiet week, the AWS bill matches. Lambda, DynamoDB pay-per-request, S3, CloudWatch, EventBridge, and SNS all sit in or near the always-free tier at the volumes a small business sees. Even running the polling job for Yelp every hour, twenty-four hours a day, all month, costs a vanishingly small amount — EventBridge scheduled invocations are essentially free at this volume, and each poll is a single Lambda call that reads from one platform endpoint.

The biggest single line item in tier one is usually CloudWatch log storage, and you control that by retaining seven days instead of forever. After day eight, the logs are gone and the bill stops compounding.

The variable cost is per-review pennies

Three things scale with review volume:

• Bedrock Haiku tokens for the extractors. Each review pays for three small model calls (rating-with-sentiment, themes, specifics) running against the review text. At small-model rates and the short input lengths reviews tend to be, the all-in cost per review for extraction lands at fractions of a cent.
• Bedrock Haiku tokens for the composer. Auto-replies and drafts each pay for one composer call. The input is the structured review plus the relevant policies excerpt; the output is a short reply. Even on the most expensive moves, this is well under a cent per review. The composer doesn’t fire on Ignore, so junk reviews cost only the screening tier.
• Titan Embeddings for the policies and voice files. One-off per file change; if you edit your policies file twice a month, that’s two embeddings calls totalling tiny fractions of a cent. The embeddings power the citation gate — finding the right policy passage to ground a claim.

Add it up at typical small-business review volumes (twenty to a hundred new reviews a month across all platforms): tier two is in the under-three-dollars range, tier one floors are under fifty cents, and the all-in monthly bill lands at a coffee a month. At higher volumes — a chain with several hundred reviews a month — tier two is the headline number, and even there it stays in the “phone bill, not Netflix subscription” range.

Three traps you’re avoiding

• Per-listing reputation tools. Most off-the-shelf reputation managers charge $40–$200 per listing per month, regardless of volume. You’re trading a flat per-listing bill for pay-per-use that mostly comes in pennies, with the model running on the same shape regardless of how many listings you connect.
• Always-on infrastructure for polling. A naive setup might run a small server that polls Yelp once a minute. That’s a $30+/month idle bill before it processes a single review. EventBridge schedules trigger Lambda — pay only when the schedule fires, scale to zero between fires.
• Compose-on-every-review at long context. A larger model would spend ten to twenty times more per review for what is, at this scale, a short-input short-output task. Haiku-class models are correctly sized for review composition; reaching for a bigger one because it sounds smarter is the most common way to triple the bill.

When this stops being cheap

The math changes if you connect dozens of listings (a multi-location franchise) and the review traffic goes into the thousands per month. At that point the tier-two number becomes the headline, and there are levers: caching policy embeddings (already cached after the first lookup), batching the extractors into a single multi-output call, or moving to a smaller dedicated extraction model for the rating-with-sentiment step. Most small businesses, including small chains, never need any of those levers.

For everyone below that — and that’s most small businesses — a $10 monthly AWS Budgets alarm catches anything strange before it becomes a surprise.

In plain words

The fixed bill is nearly zero. The variable bill is pennies per review, dominated by tokens. A typical setup runs at coffee-money for the whole month. Set a budget alarm that fits your expected volume and the bill can’t surprise you.

All posts