Measured, on a real report
The failure rate. An agent attempts discover → evaluate → compare → buy against your store and each stage is scored on what it actually managed.
AI commerce ROI calculator
An AI commerce ROI calculator estimates the revenue a store loses when AI shopping agents cannot complete a purchase on it. The loss is the store’s monthly revenue, multiplied by the share of it that is agent-mediated, multiplied by the share of agent attempts that fail. Only the last of those three is something anyone can measure for you.
We never see this. It stays in your browser.
Nobody can measure this yet, us included. It is a control, not a claim. Set it to what you believe and watch the estimate move. The default is deliberately low.
Estimated monthly revenue at risk
$550-1,100/mo
45% of agent-mediated attempts do not complete on a store like this. Modelled from the profile above.
Each stage on its own. These are not added up: an agent gets no further than your weakest stage, so the estimate uses the worst one.
| Discover | 25% would not get past this stage |
|---|---|
| Evaluate | 30% would not get past this stage |
| Compare | 20% would not get past this stage |
| Buy | 45% would not get past this stage (the weakest, used above) |
Get your real number. Run the scanThe scan replaces “a bit of each” with what an agent actually did on your store.
What this can and cannot tell you
Yours, always
Revenue. We do not have it, do not ask for it, and it never leaves your browser on this page.
Assumed, and visibly so
The agent-mediated share of your revenue. There is no trustworthy figure for this yet. When our benchmark dataset is large enough to publish one, it will arrive with a published methodology, not before.
Common questions
Where does the failure rate come from?
On this page you choose a profile, so it is modelled, not measured. The page says so beside every figure. In an Ottom report the same calculation runs on your store’s own stage scores, produced by an agent that actually attempted the purchase.
How much of my revenue is really agent-mediated?
Nobody can answer that reliably today, us included, which is why it is a slider rather than a number we assert. The default band is 1-3% and is deliberately conservative. Set it to what you believe; the estimate moves with it, and the spread you see is entirely that uncertainty.
Why is the result a range and not a number?
Because one of the three inputs is an assumption. A single figure would read as a quote. The range is the honest shape of an estimate with one unknown in it.
How is the loss calculated?
Each stage’s score becomes the share of agent attempts that get past it, and the headline is the worst single stage. An agent gets no further than your weakest one. We deliberately do not multiply the four stages together: a stage score is a quality rubric rather than a measured drop-off rate, and compounding it four times produces a figure far more alarming than we can defend. This understates the loss when several stages leak at once, which is the direction we would rather be wrong in. A stage we did not test is treated as passing, never as a failure.
Do you see my revenue figure?
No. The calculation runs in your browser and nothing is submitted.
Read next
- Why AI agents abandon carts
The failure modes behind the buy-stage number above, and how they differ from human cart abandonment.
- The agentic readiness checklist
What to fix first, in the order an agent hits it.
- What is agentic commerce?
The definition, and why a store can pass every SEO check and still fail this one.
- Agentic commerce: the reference page
The whole subject in one place: how an agent reads a storefront, what ACP, UCP, AP2 and MCP each cover, and what has actually shipped, with every claim sourced.
- Agent view: your product page as an agent reads it
The other free tool. Paste one product URL and see the sparse text an agent actually receives, with no revenue figure required.
- Agentic commerce benchmarks
What we can and cannot yet say about how a whole category performs, and the rule that decides when a category is published at all.
Replace the guess with a measurement
Every number above is your own estimate multiplied by an assumption. A scan is what your store actually does when an agent tries to buy from it. One free scan per store, per email address, every 30 days.