You set the goal. The agent decides for each shopper, and proves the lift.
It chooses what each shopper sees in search, pages, chat and messages, and who joins your ad audiences.
- One goal behind every action
- Your caps, consent and rules apply
- Holdout read in your Shopify or GA4
- If the holdout shows no added orders in 30 days, you don't pay
Seven steps, running for every visitor
Capture
Every visit, server-side, including anonymous visitors and their ad clicks.
Remember
One profile per person across devices; segments refresh every 10 to 30 minutes.
Predict
Intent, the next purchase and 12-month value, per person.
Decide
Which products, message, channel and moment, per person. Or nothing.
Act
Search, product blocks, chat, messages, ad audiences.
Measure
Your own orders, against a holdout group.
Learn
What worked feeds back into each profile and placement.
Orders and customer value, not clicks
Each possible action gets one score: how much it is predicted to add to the customer's future value. A click that leads to no order is worth nothing. Two predictions feed the score: what each shopper is likely to buy next, and their likely spend over the next year, checked monthly against orders.
Winners are picked by results, not by hand
Each one is judged on orders.
Self-learning tests
They run on every page block, email and chat. Traffic moves to the variant that sells.
See it on the site →Next best action
Chooses each customer's next step: a message, an offer or a handoff to your team.
See it in messages →The cheapest channel that works
Push first for anonymous visitors. Paid SMS or Viber only when email got no response.
You choose the goal. You set the rules.
Inside those limits, it decides for each person.
Goals you can set
- Conversion
- Basket
- 12-month value
- Repeat rate
Decisions it makes
- Searchranks results per shopper
- Pageswhich products each block shows
- Chatwhat to recommend, add or hand off
- Messagescontent, channel and timing
- Adswho joins the audiences, the value sent per purchase
Rules it works within
- Frequency capsDaily limits per channel. At the cap, you choose: reschedule, skip or send.
- Stop on purchaseA shopper who buys leaves the flow.
- Your merchant rulesYour sales strategy, rules and product attributes.
- ConsentSet per flow. Every message follows them.
Check the lift in your own store data
Attributed revenue is an argument. A holdout is evidence. A share of visitors never sees the agent, and revenue per visitor is compared in your own data. The baseline stays clean because the agent experiments only on the group that gets it, and every attributed order is counted once and downloadable with its Google and Meta clicks.
The attribution window alone can change the number a lot: at one retailer in July 2026, a conservative window reported 31% less attributed revenue than a permissive one on the same orders (how much the window changes the number).
Know what drove each sale →Measured against a baseline, not just attributed
A controlled A/B test, a matched cohort and a test switch-off.
revenue per recipient against a copy of Klaviyo's logic; controlled A/B, week one.
Read the case study →Womenswear · Shopify · Meta+65%buyer rate for the value-based campaign against the old one; matched 30-day cohort.
Read the case study →Bedding and sleepwear · WooCommerce−10.2%items per order in the two weeks after personalization was switched off for a test.
Read the case study →Questions about how it decides
How does Releva decide what each shopper sees?
It scores each shopper and each option against the goal you set. A predicted-value score, self-learning tests on every placement and a next-best-action model pick the search order, products, banner, message and channel. All of it stays within your rules, frequency caps and consent settings.
Should recommendations optimize for clicks or for sales?
For the goal you choose. Engagement is not the goal: showing everyone the same thing can win clicks and still lose sales. Releva lets you choose conversions, bigger baskets or long-term customer value, then decides per person toward it.
Is it really autonomous?
Yes, within the goal and rules you set. Inside those, it decides per person without anyone writing a rule for each case.
What is a holdout group, and how big should it be?
A share of visitors who don't get the agent. Comparing revenue per visitor between the two groups shows what the agent added. Releva sets holdouts up automatically at the size you choose: a bigger holdout gives a faster, surer read but holds more visitors back.
How is Releva different from Dynamic Yield, Bloomreach or Insider One?
Dynamic Yield, Bloomreach and Insider One sell personalization and customer engagement platforms. Releva is one agent for mid-size B2C stores. You set the goal and your rules, and it decides per person without a rule for each case. A holdout group of any size is set up automatically.
Can our data team check the numbers?
Yes. Each attributed order is counted once and downloadable with its ad clicks. Default windows, which you can change: an open within 1 day or a click within 7 days for messages, 1 day for on-site clicks. Holdouts use stable test groups and significance tests.
Pick a goal. See what the agent would change.
Thirty minutes with your store open. We show the decisions it would make and how to read the holdout.
Book 30 minutes
You'll talk to one of us.