What AI automation costs: the real numbers from my listing pipeline
What my eBay listing pipeline cost to build, what each run costs at API prices, and how many batches it takes to pay the build back.
Part one showed how my listing pipeline works, and part two covered what went wrong. This part answers the question I’d ask next if I were you: what does something like this cost?
There are two costs. You pay one once, to build it. You pay the other every time it runs. Here are both of mine, with the math, so you can swap in your own numbers.
What it cost to build
Version 1 took me about eight hours to build, test, dial in, and launch. I built it with Codex, the same AI tool that now runs it.
For this math I value an hour of my time at $150. That puts the build at $1,200.
That’s version 1 only. The changes I’ve made since aren’t in that number. And $150 is my figure, not a rule. Use what an hour of your own time is worth.
What the job costs by hand
Before the pipeline, every card took me 5 to 10 minutes: assign the SKU, upload the photos, identify the card, research what it sold for, and fill in the listing.
| Per card | Per batch of 50 | |
|---|---|---|
| My time by hand | 5 to 10 minutes | 4 to 8 hours |
| At $150 an hour | $12.50 to $25.00 | $625 to $1,250 |
Two things are left out because they’re the same either way. Photographing fifty cards takes me about ten minutes. And I sort every stack by condition myself before the camera comes out.
What a run costs
I pay for the AI with a flat monthly plan, so no run is billed to me by the token. That’s no help to you if you’re pricing your own project, so everything below uses the provider’s pay-as-you-go API list prices.
I added up 34 days of the pipeline’s own logs, from September 2 to October 5, 2026. A batch run is one run of the pipeline working through a batch of fifty cards:
| Runs | Median cost | Range | |
|---|---|---|---|
| Batch runs | 14 | $42.39 | $13 to $119 |
| Nights without a full batch | 28 | $1.29 | $0.41 to $21 |
The median is the middle run: half cost more, and half cost less. A median batch run works out to about $0.85 a card.
The second row is every other night. The worker wakes up nightly. When no full batch is waiting, it checks, finishes any leftover work, and stops.
The range is wide, and the biggest reason is the model. Here are the same 14 batch runs, token for token, priced on five of OpenAI’s models:
One batch run at each model’s prices
Most of those runs used a model that prices a median batch at about $47. On October 5 I moved the worker to one that prices it at about $15. Two batch runs had already used that model, at $17 and $38. I don’t have enough runs on it yet to say what the change does to accuracy, so I’m watching how many drafts I accept without corrections.
The math per card
Cost per card
The automation isn’t hands-off. Every draft still gets read and published by me, and that takes about 30 seconds a card. So my review stays in the math, at $1.25 of my time.
| Per card | Per batch of 50 | |
|---|---|---|
| By hand | $12.50 to $25.00 | $625 to $1,250 |
| AI, at list price | $0.85 | $42 |
| My review, about 30 seconds a card | $1.25 | $62.50 |
| Saved | $10.40 to $22.90 | $520 to $1,145 |
When the build pays for itself
Divide the build by what a batch saves:
| Saved per batch | Batches to pay back $1,200 | |
|---|---|---|
| If the cards were quick by hand (5 minutes each) | $520 | 2.3 |
| If the cards were slow by hand (10 minutes each) | $1,145 | Just over 1 |
Either way, the build is paid back by the third batch. I have thousands of cards to list.
If your review runs longer than mine, the answer moves, but not far. At three minutes a card, the same build pays back in 1.4 to 5.8 batches.
What this math leaves out
- Time spent improving it. The $1,200 is version 1. Later fixes cost hours too.
- The plan I actually pay. List prices make the numbers comparable. They aren’t my bill.
- Cards that need me. The pipeline stops and asks when it isn’t sure. Those cards take longer than the usual 30 seconds.
- Changing prices. Model prices move. The figures here are from October 5, 2026.
What I’d take from this for your business
- Time the job by hand first. Without that number, there’s nothing to compare the automation to.
- Keep your own review in the math. An automation that still needs a person isn’t free to run. It’s cheaper to run.
- Ask which model, and why. The same batch prices anywhere from about $2 to about $116, depending on the model.
- Ask for the cost of a run, not a month. A cost per run you can check against the logs. A monthly figure hides whatever is being wasted.
Why a run costs what it does
A median of $42 a batch made me ask what I was paying for. The answer isn’t thinking. It’s re-reading, and part four shows it call by call.
If you’d like this math done for one of your own jobs, that’s where AI consulting starts. Or tell me about the job you’d automate first.