Planned module

Tests that end with an answer.

Two subject lines, with or without a discount, Tuesday or Thursday. The agent sets up the test, decides how many customers are needed, and tells you when there’s a real winner, measured in orders.

Planned: 30% off for the first 100 stores. One email when your invite is ready, no newsletter.

  • Works with the email tool you have
  • Never priced per contact
  • You approve every message
Design · not built yet: Late regulars: a reminder without discount brought 31 orders, with 10% off 33. The difference is too small to justify the discount. Keep it off.

Most email tests are too small to mean anything.

Picking a winner after 40 opens.

Measuring opens instead of orders.

Tests that are started and never looked at again.

Oftenflow keeps them in the loop

Inside the app

What you’ll see, step by step

  1. Step 1 / 3

    Ask for a test

    “Test the discount for late regulars.” The agent proposes the variants.

  2. Step 2 / 3

    Right size, right split

    It calculates how many customers are needed and splits them at random.

  3. Step 3 / 3

    Honest result

    A winner only when the difference in orders is clear; otherwise it says so.

Planned module

Want A/B testing in your store?

Join the waitlist and tell us which modules you need first. Founding stores help decide the order.

  • Works with Shopify Messaging, Klaviyo or Brevo
  • Never priced per contact
  • Export all your data, anytime

Planned: 30% off for the first 100 stores. One email when your invite is ready, no newsletter.

A closer look

A closer look

Sizedbefore it starts

Sized before it starts

Before a test runs, the agent works out how many customers it needs to see a real difference in orders. If your store is too small, it says so instead of running a test that can’t answer.

Ordersnot opens

Measured in orders

Opens are unreliable because mail apps preload images. Tests are judged on reorders within each customer’s usual interval.

Learnedby the next proposal

Results that change what happens next

A finished test updates the numbers the decision engine uses, so the next proposal is based on what worked in your store.

A/B testing

Orders, not opens

Measured on reorders within the customer’s usual interval.

Built on the holdout

The same comparison method as the weekly digest.

Learns automatically

Results feed the decision engine, so the next proposal uses what worked.

Stops itself

A test ends when it has an answer or clearly won’t get one.

Example result
Late regulars: a reminder without discount brought 31 orders, with 10% off 33. The difference is too small to justify the discount. Keep it off.
In practice

Discount or no discount for late regulars

  1. Ask

    “Does the discount help for late regulars?”

  2. Size

    About 400 customers needed; your store gets there in roughly 5 weeks.

  3. Run

    Late regulars are split at random: reminder with or without 10% off.

  4. Result

    31 vs 33 orders. Not enough difference to pay for the discount.

  5. Learn

    Late regulars get reminders without a discount from now on.

Planned flow; numbers are illustrative.

Without Oftenflow / With Oftenflow

Without OftenflowWith Oftenflow
Winner picked after a handful of opensWinner only when the difference in orders is clear
Tests that never endStops when there’s an answer, or won’t be one
Results that stay in a reportResults that change the next proposal
Guessing the right sample sizeCalculated up front

What you decide

  • What gets tested and for which group
  • Maximum share of customers in a test
  • How long a test may run
  • Stopping a test at any time
Questions

Questions

Is this available now?

No. It’s a planned module; the holdout measurement it builds on is part of the core.

How small can my store be?

Small stores need longer tests. The agent tells you up front if a test is realistic for your volume.

What can be tested?

The plan: subject lines, discount or not, timing, channel and recommendations.

Is a winner always declared?

No. If there’s no clear difference, the result is “no difference”, which is also useful.

Does testing use credits?

Module pricing isn’t final. Drafts for test variants are expected to use credits like other drafts.

Waitlist · in development

Get an invite when we open the doors.

Planned: 30% off for the first 100 stores. One email when your invite is ready, no newsletter.

  • Plans from €29 a month, never per contact
  • Works with the email tool you already use
  • No card, no commitment
Example from the Oftenflow app: overnight the agent read new orders, updated buying rhythms and wrote drafts.
Join the waitlist