Getting a purchase after an email doesn't prove the email sold. Here's how I measure it

Monday comes and someone asks in the meeting: "How much did Friday's email sell?". The tool says lots of people opened it and several bought afterward. Sounds good, but nobody can answer the real question: did they buy because of the email, or were they going to buy anyway?

For a Chilean eCommerce with two stores, I built its own mailing platform. You choose who you write to, artificial intelligence drafts the email in the store's brand voice, you review it, schedule it, and then you see opens, clicks, unsubscribes and how much it really sold. Order emails keep going out from the store, and every campaign is scheduled by a person.

---

The dashboard: can I send?

In this app the scarce resource isn't contacts: it's the emails the server lets you send, a limit shared with purchase confirmations.

• Hourly and daily quota, in rolling windows, with the time it frees up.
• Installation status: mail server, domain, AI and syncs.
• Sales attributed to emails at 7 or 30 days, against the store's total.
• Campaigns in progress and sent, with alerts when unsubscribes or bounces climb higher than is healthy.

---

Contacts, lists and segments

Email is the identity: if a person arrives through the web store and through the ERP, they merge into a single record. Contacts arrive on their own every early morning, or from a CSV or by hand, always recording where the consent came from.

What changed the way we work were the automatic lists, rebuilt from scratch every day:

• Behavior: bought recently, frequent, at risk, inactive, best customers.
• Arrival: which channel and page each customer came from on their first purchase.
• No second purchase, in stages for the first reminder, the second nudge and the last call.
• Repurchase based on each customer's rhythm: who's due to come back and who has drifted away.
• Category, age, birthday and region.

When the list doesn't exist, you build a segment: rules like "spent over a certain amount and haven't bought in 90 days," with a live count, recalculated every time they're used.

---

A campaign in four steps

1. Audience

• Lists and segments are added together and each person receives the email only once.
• If it doesn't fit within a day's quota, the app tells you how many parts it will go out in.
• Send order: the most active or best customers first, so the last day isn't theirs.
• Control group: a random percentage that doesn't receive the email.

2. AI content

You write a brief with what's being communicated and the real conditions; the AI is forbidden from inventing prices, discounts or dates. In a minute or two the email arrives with the brand, three subject lines and the preheader (the line the inbox shows after the subject). You refine it by chatting, every change is saved as a version, and a test email goes out before you move on.

3. Review

Some errors block sending, like a sender from another domain or too many links, and warnings let you send anyway, like a subject line that gets cut off on a phone. A spam score sums up the risk.

4. Schedule

It goes out now or on a set date. A long send can be split into campaigns, each with its own day and subject line. Once scheduled, the email and the audience are frozen.

---

Send without burning the domain

• Steady pace: the hourly cap is spread across the whole hour, with no bursts.
• Warm-up: the daily volume starts low and rises every day.
• Automatic pauses if the server rejects the key or several emails fail in a row.
• Authenticated domain: the app checks the records that prove to Gmail the email comes from the store.

---

How much it really sold

The platform matches the store's and the ERP's orders against what each person did. A confirmed sale is one from someone who came in through a link in the email; an attributedsale is one from someone who opened or clicked and bought in the following days.

Each order carries an attribution percentage: an AI judges how much the email influenced it based on the interaction, the timing and the customer's history, within fixed margins. An order is never counted twice.

The control group

Even so, attributed is not the same as caused. The honest way to measure is to compare against those who didn't receive the email.

The incremental sales are the difference in sales per person between the two groups, multiplied by the number sent. It can come out negative, and the app warns you when the sample is small.

---

Unsubscribes, bounces and suppression

Unsubscribing asks for confirmation, because some antivirus programs open every link in an email. Bounces and spam complaints are detected automatically, and unsubscribes from the ERP arrive too. Everything ends up in the suppression list, which overrides any other list, and only an admin can reverse it, with a log.

---

What the business gains

Audiences that build themselves from real purchase data, emails in minutes with no invented figures, a domain that doesn't end up in spam, and an honest answer to Monday's question: how much the email sold, and how much of that wouldn't have happened without it.