How to Use AI for Email Marketing: 4 Levels of Autonomy

AI in email marketing works in four levels of autonomy. Level 1 drafts copy. Level 2 predicts send time and churn. Level 3 builds predictive segments and product recommendations. Level 4 decides the offer, timing, and channel for each shopper and executes without a human. Most ecommerce brands sit at Level 2 or Level 3, the levels mainstream platforms ship.

We build the Level 4 option, so read this as informed but interested. Everything at Levels 1 through 3 is written to be run this week on tools most brands already pay for, and the limits of each level are described as honestly as we can put them. The reframe at the end is ours.

Why levels, not features

Self-driving cars are graded by how much of the driving the car does, not by how many sensors are bolted to the roof. Email is easier to reason about the same way. The useful question is not which AI features a platform ships. It is how much of the decision-making has moved off the marketer’s desk. Each level below is a real place to be, with real work to do at it.

Level 1: AI writes the copy

Almost everyone starts here, and it works. A model drafts subject lines, preview text, body copy, and variants in seconds.

What separates useful Level 1 output from generic Level 1 output is not the prompt. It is the input. Before the next campaign, paste in your last fifty five-star reviews, a dozen real support tickets, and the copy from the three best-performing emails you have ever sent, then tell the model to write in that language. Reviews supply the words customers actually use for the product. Support tickets supply the objections. Keep a short voice file alongside it, ten phrases the brand always uses and ten it never uses, and paste that into every prompt.

What Level 1 does not do is know who is receiving the email. Our guide to AI email copy generation goes deeper on that job; the rest of this post is about the levels above it.

Level 2: AI predicts send time and churn

Prediction is already switched on in most accounts, which is why it is the easiest level to own without using. Three things are worth doing this week.

First, check what send-time optimization is optimizing for. Klaviyo’s Smart Send Time and Mailchimp’s send-time optimization both need a calibration send before they have anything to work with, and both lean heavily on open behavior. Opens got noisier once Apple Mail started pre-fetching images, so if the platform allows optimization on clicks or placed orders instead, choose that.

Second, go and find your predictions. Predicted lifetime value, churn risk, expected date of next order, average time between orders: these appear on customer profiles once a store has enough order history behind it. Newer stores and low-repeat categories will find them blank, which is worth knowing before building on them.

Third, do the part most accounts skip. A prediction is not a decision until something is wired to it. List every predictive segment in the account, then find the flow or campaign that actually references it. In a lot of accounts the honest answer is none, and the prediction has been decoration for a year. The cheapest fix that pays back: route high churn risk plus a purchase in the last few months into a win-back that fires before the shopper lapses, instead of the standard win-back that fires long after they are gone.

Level 3: AI segments and recommends products

At Level 3 the AI starts choosing content, not just timing. Two concrete moves, both under an hour.

Open the product recommendation block in your abandoned-cart and post-purchase emails and look at its source setting. Most blocks ship pointed at a store-wide bestseller list, which means every shopper in the flow sees the same products. Point the cart email at products related to the item left in the cart, and the post-purchase email at that shopper’s own browse and purchase history. It is a small change with a visible difference.

Then replace one demographic segment with a behavioral one. A segment of shoppers who viewed a product two or more times in the last seven days and have not purchased is a far better browse-abandon audience than any list attribute. While you are in there, add a frequency cap that excludes anyone who has already received three emails in the past seven days. Suppression is the least glamorous part of Level 3 and the part most likely to protect your deliverability.

Level 3 has a real limit, and it is structural rather than a flaw in the software. Segments and recommendations are computed on a refresh schedule, so a shopper whose intent changed yesterday can still be scored on last week’s behavior. And a segment is still a band. Tens of thousands of people, one decision. The tool is doing precisely what it was asked to do. It was asked to decide once, for a group.

Level 4: AI decides and executes per shopper

At the top level the decision itself moves. The loop is Observe → Decide → Act: the system reads live shopper signals, decides the offer, timing, and channel for that one person, and executes, with no flow built in advance and no human approving the send.

This is the category Relvino is in. Each decision takes about 80ms per shopper, and acts across email, SMS, and on-site. The marketer’s job changes from building flows to setting bounds: margin floors, discount ceilings, permitted channels, brand voice, frequency. Inside those guardrails, 100% of flows run without a human in the loop, because the decision replaces the flow rather than automating a flow someone still designed.

The engine is a Large Retail Model trained on 7M+ data points across 10K+ retailers and 1.78M shoppers, so it arrives with priors about what works in a category rather than learning only from one store’s own history. Brands that replace their flow stack see 2–6× ROI in 30 days and up to 10× revenue uplift year over year against their incumbent, with 80% less spam. Against Klaviyo specifically, the bill runs 3–7× cheaper.

Relvino is better than Klaviyo. Next gen is Relvino. - Joshua Rockoff, CMO, Omni Retail Enterprises, and a Relvino advisor

The mechanics are covered in more detail in our piece on autonomous email marketing.

The four levels at a glance

• Who writes the message · Levels 1–2: AI drafts, human approves · Level 3: AI drafts, AI picks the products · Relvino (Level 4): AI drafts and picks, no approval step
• Who chooses the recipient · Levels 1–2: Human, via a segment · Level 3: Predictive segment, refreshed on a schedule · Relvino (Level 4): Decided per shopper, on live signals
• Who chooses the timing · Levels 1–2: AI, inside a human-built flow · Level 3: AI, inside a human-built flow · Relvino (Level 4): Decided per shopper in ~80ms
• Who chooses the offer · Levels 1–2: Human · Level 3: Human, with AI product picks · Relvino (Level 4): AI, inside margin guardrails
• Who chooses the channel · Levels 1–2: Human · Level 3: Human · Relvino (Level 4): AI, per shopper
• Unit of decision · Levels 1–2: The campaign · Level 3: The segment · Relvino (Level 4): The individual shopper
• Human in the loop · Levels 1–2: Every send · Level 3: Every flow · Relvino (Level 4): Guardrails only

So which level should you choose?

Here is a decision rule that beats a feature comparison. Count the lifecycle decisions your team genuinely makes in a week: which segment gets which offer, at what discount, on which day. Then divide your active shopper count by that number. That is how many people share each decision.

If that number is small, and somebody on the team has operating lifecycle as their actual job, stay at Levels 1 to 3 and work the checklist above. AI features make a good operator meaningfully faster, and a good operator is a real asset. Our roundup of AI email marketing tools covers what is worth buying at those levels.

If that number runs to the tens of thousands, and results have been flat for two or three quarters while the team stayed lean, then more Level 1 to 3 features will make the same small set of decisions faster without making more of them. That is the case for moving up a level rather than adding another feature. See pricing for what Level 4 costs, or read our Klaviyo alternative comparison for what changes operationally on the day you switch.

Frequently asked questions

Which AI tool is best for email marketing?

It depends which level of autonomy is wanted. For drafting copy at Level 1, the standalone generators and the assistants built into Klaviyo, Mailchimp, and HubSpot all do the job well. For prediction and segmentation at Levels 2 and 3, Klaviyo has the deepest predictive feature set for Shopify brands, with Omnisend and Mailchimp close behind on the basics. For deciding and sending per shopper with no human in the loop at Level 4, that is a different category called autonomous decisioning, where Relvino decides each shopper’s offer, timing, and channel in about 80ms and executes on its own. Pick the level first, then the tool.

How can I use AI for emails?

Start at the level you can act on this week. At Level 1, feed a model your real reviews, support tickets, and best-performing emails, then have it draft in that language rather than from a blank brief. At Level 2, turn on send-time optimization and check that predictive fields such as churn risk and predicted lifetime value are actually wired to a flow instead of sitting unused. At Level 3, set product recommendation blocks to use browse and purchase history rather than store-wide bestsellers, and replace one demographic segment with a behavioral one. At Level 4, hand the offer, timing, and channel decision to an autonomous system and set guardrails instead of building flows.

What is the 80/20 rule in email marketing?

The 80/20 rule is a planning heuristic borrowed from the Pareto principle, and it gets used two ways in email. As a content rule it means roughly four emails in five should give the reader something useful and only one in five should sell directly. As an audience rule it means a small part of a list produces most of the revenue, so the highest-value work is usually identifying and looking after that group rather than mailing everyone more often. It is a rule of thumb for planning, not a measured benchmark, and it is worth checking against your own numbers before designing a calendar around it.

Is email marketing still worth it in 2026?

Yes. Email is still the highest-return owned channel in ecommerce, because the audience belongs to the brand and the cost per message is low. What has changed is where the remaining upside sits. It is no longer in sending more, and it is no longer in writing better copy, because generated copy is now cheap for everyone. The upside is in decision quality: which shopper gets which message, at which moment, with which offer. A brand still deciding once per segment is leaving most of that upside untouched.

How do I migrate from Klaviyo to Relvino?

The technical cutover takes about 30 minutes: authenticate the sending domain, install the Shopify app, and historical data ingests automatically. Existing flows are switched off rather than rebuilt, because at Level 4 there are no flows to recreate. From there it is a 14-day pilot. The standard advice is to keep the Klaviyo account on its current plan rather than cancel it, so both systems can be compared on revenue side by side before anything is turned off permanently.