For online stores: every order judged by what it’s worth.
A store gets orders every day, so tests settle within weeks. What matters is whether each order was worth what you paid for it, so agents judge every order by its value to you.
Each agent sends a report every morning.
Two agents, two reports, both judged on the store’s own orders.
What the agents do.
The funnel, with each step’s worth
The New Customers agent’s funnel for a coffee roaster. Its target is first orders under $26.
Where the numbers come from
Orders come from your store or the events API, each with its order value. Each ad platform steers by its own count, and budget moves between platforms based on your orders.
- 1Added to cart$3
- 2Checkout started$8
- 3First orderits order value, about $47
- 4Subscription startedits first-year value, about $312
- 1
Move budget to what sells
With your stock and margin data connected, agents read it alongside the results. The New Customers agent proposed moving $60 a day from broad coffee campaigns to the Holiday Blend and gift sets. Any move above 50% of a campaign’s budget needs your approval.
- 2
Test creative that sells
When click prices on brewing gear rose, the agent asked the creative desk for new ads and, once you approved them, proposed a 14-day test: $40 a day, $560 in all.
- 3
Bring first-time buyers back
The Reorders agent writes a running-low note to people whose first bag is probably empty. Your contact rules narrowed 1,860 first-time buyers to 1,376, and 93 are held out so the effect can be measured.
- 4
Win back the near-misses
The Retargeting agent gave people who took the taste quiz but didn’t buy an audience and budget of their own. A first order from that group now costs $13, against $25 for other visitors.
- How they’re judged
- Your own conversion events, each valued by what it’s worth to you. A first order counts at its order value. Set it to the margin it earns instead, and agents judge every dollar against what you keep.
- How often they act
- Daily. With orders every day, a test can be called in two weeks.
- A good fit
- Stores that sell online and spend on search or social ads, from a few thousand dollars a month upward.
- Not the best fit
- Stores that need new short-form or creator-made video every week. The desks make still images, and your own footage runs as you made it.
The same store at $30,000 a month
The same kind of store at three times the spend: two platforms, three campaigns and video from your own footage. One week of it follows, covering catalogs, blended cost per order, incrementality and creative volume.
Raise the daily budget 30%, $1,000 to $1,300, all of it to Advantage+ sales
A 30% change exceeds the 20% limit per change. The agent proposed $1,000 to $1,150 instead, a 15% change, which ran and told you.
Win-back email to lapsed customers
- Sent · 15,960
- 9.0%
- 1,436 second orders
- Held out · 840 (5%)
- 6.2%
- 52 second orders
About 448 of the 1,436 second orders were caused by the email. The rest would have come anyway, and the report says so.
The creative desk makes still images. Video comes from footage you upload: approved once, kept in the library, and used by any agent.
What changes at larger spend
Catalogs and automated sales campaigns
Agents manage them as campaigns and leave your feed alone.
How it works
To an agent, a catalog campaign or a platform’s automated sales campaign is one campaign. It can propose a new budget, pause it or resume it, under the same limits as any other.
Your product feed stays where it is, in your store and the platform’s catalog. Agents don’t edit feeds or product sets.
Blended cost per first order
Judged on your orders across channels.
How it works
Cost per first order is the agents’ ad spend across platforms divided by the first orders they brought in, each order counted once, from your own events. For the roaster, that’s $4,027 ÷ 195 = $21.
Each order counts for the worth you set: a first order at its order value, a subscription at its first-year value. Budget moves between platforms on that worth, whatever each platform credits itself.
Incrementality
Email sends keep a holdout by default. With enough volume, paid tests leave a region out.
How it works
By default, each email send holds back 5% of its audience, so its effect is measured against people who didn’t get it.
For paid ads, when volume allows, agents propose tests that leave out a region or audience. When it doesn’t, the report says so instead of guessing.
Creative volume
The desks make stills and copy. Your footage covers video.
How it works
Agents brief the creative and copy desks when they need new ads. The desks make still images and copy, and the brand review agent checks every ad before it runs.
Your own footage goes in the library. Once you approve it, any agent can run it in ads alongside the desk’s images.
A week with the agents.
| When | The agents | You |
|---|---|---|
| 6:02 a.m. | The New Customers agent files its report: first orders rose to 71 from 60, and cost per first order fell from $23 to $20. | Read it over coffee. |
| Same morning | The Retargeting agent reports on its taste-quiz audience: a first order from a quiz taker cost $13, against $25 for other visitors. It had already shifted 15 percent of its budget to them, within your limits, and said so in its report. | Nothing needed. |
| Mid-morning | The creative desk delivers three feed ads for the pour-over kit. They stay drafts until you approve them. | Approve the images, or send them back with a note. |
| Afternoon | With the images approved, the agent proposes the launch: +$40 a day for 14 days, $560 in all. | Approve or decline. |
| Afternoon | The Reorders agent proposes the running-low note to 1,376 of 1,860 first-time buyers, with 93 held out. | Read the email, then approve or decline. |
| Ongoing | The roast-date test for gift sets is on day 9 of 14, and 87% confident. The agent calls it when the test ends, and only at 95% confidence. | Nothing needed. |
| Every day | Small changes you’ve allowed, such as lowering a bid on broad site visitors, are made and logged. | Skim them when you like. |
Show the week, day by day
| When | The agents | You |
|---|---|---|
| 6:02 a.m. | The New Customers agent files its report: first orders rose to 71 from 60, and cost per first order fell from $23 to $20. | Read it over coffee. |
| Same morning | The Retargeting agent reports on its taste-quiz audience: a first order from a quiz taker cost $13, against $25 for other visitors. It had already shifted 15 percent of its budget to them, within your limits, and said so in its report. | Nothing needed. |
| Mid-morning | The creative desk delivers three feed ads for the pour-over kit. They stay drafts until you approve them. | Approve the images, or send them back with a note. |
| Afternoon | With the images approved, the agent proposes the launch: +$40 a day for 14 days, $560 in all. | Approve or decline. |
| Afternoon | The Reorders agent proposes the running-low note to 1,376 of 1,860 first-time buyers, with 93 held out. | Read the email, then approve or decline. |
| Ongoing | The roast-date test for gift sets is on day 9 of 14, and 87% confident. The agent calls it when the test ends, and only at 95% confidence. | Nothing needed. |
| Every day | Small changes you’ve allowed, such as lowering a bid on broad site visitors, are made and logged. | Skim them when you like. |
Questions
What happens when a product runs low?
Your brief can say never to advertise anything with less than three weeks of stock. With your stock data connected, the agent reads it before it plans and puts budget behind what you have plenty of.
What happens when a platform reports more sales than your store?
Each platform’s own count steers that platform. Decisions across platforms use your own conversion events, so no platform wins budget by counting generously.
What happens on a big sale day?
Your daily and monthly caps still apply. If a day needs more, you raise the cap. Agents can’t. No budget change can move an agent’s total by more than your per-change limit, and new campaigns stay under your daily cap.
What happens when a test is still unclear at the end?
It’s called “no difference” and the current version stays. The agent never picks a winner the numbers don’t support.
What spend is it built for?
The roaster’s New Customers agent has a $230 daily budget under a $275 cap, and all its agents together stay under a $400 daily cap and a $10,000 monthly cap. Those are this company’s settings; yours can be higher or lower. The section above shows a store at $30,000 a month.
Does it know my margins?
It knows what you tell it. Each funnel step’s worth is yours to set, either as a fixed amount or from a value on the event, such as order value or margin. Agents judge spend against that worth.
Can it make video ads?
It runs video you supply. The creative desk makes still images and copy. Upload your footage once, approve it, and any agent can use it.
How do you know the ads caused the sales?
Agents don’t take a platform’s credit at face value. They’re judged on your orders, against a baseline from your own history. Where your volume allows, they propose tests that leave a region or audience out, and email sends keep a holdout by default.
In the demo
Open the inbox in the demo (opens in a new tab)Put every order to work.
Join the beta, or try the demo first.
