Marketing automation for ecommerce has one job in most brands: catch the abandoned cart. And your cart flow is probably the best-performing automation you own. Klaviyo’s benchmark study of more than 143,000 abandoned cart flows puts the average conversion rate at 3.33% and revenue per recipient at $3.65 — better than any other automated flow, and far better than a bulk campaign.
Congratulations. It also triggers on the last signal a buyer will ever send you.
Picture the person working through a $2,400 purchase. Nine product page views. Two comparison sessions. A spec sheet download. A return policy read at 11:40 on a Tuesday night. A second visit twelve days later off a retargeting ad. Then, finally, add to cart — followed by a flinch at the shipping number and a closed tab.
Your system woke up at step seven. Steps one through six went into a database and died there.
Marketing automation for ecommerce has a trigger problem, not a copy problem
Nearly every ecommerce stack is architected around the same handful of commerce events: viewed product, added to cart, started checkout, placed order, shipped. The money events. And the entire optimization industry built on top of them spends its energy on the wrong half of the problem — subject lines, send delays, discount ladders, dynamic product blocks.
All of that is copy. None of it is trigger.
Here is what the trigger is actually doing. Baymard Institute’s meta-analysis of 50 separate studies puts the average documented cart abandonment rate at 70.22% — a number that has moved roughly two points in twelve years despite a decade of checkout optimization spend. That is structural, not a copy problem.
Dig into the why and it gets worse for the cart-as-trigger model. In Baymard’s survey work, 42% of US shoppers said they abandoned because they were just browsing and not ready to buy, and among those who were serious, 39% left over extra costs like shipping, taxes and fees.
Read that as an operator. The largest single chunk of your “abandoners” were never buying that day. The largest chunk of the real ones left over a number your email cannot change. Your highest-performing flow is firing too late and taking credit for an unknown share of people who were coming back anyway.
The signal you want fires weeks before the cart does
Our thesis is the same one we apply to B2B pipelines: most companies do not need more traffic. They need to know who cares right now.
That matters more every quarter, because the cheap-acquisition era is over. US retail e-commerce hit $340.2 billion in the second quarter of 2026 — 17.1% of all retail sales — according to the Census Bureau’s quarterly e-commerce report. A mature channel means the next unit of growth comes from the people already on your site, not the next ad dollar.
And those people are broadcasting constantly. Every platform in your stack is already recording it. Klaviyo’s own developer documentation describes a Viewed Product metric on every product page hit, plus an Events API that will accept any custom metric you care to define — quote requested, financing calculator opened, spec sheet downloaded, dealer locator used.
The data is there. Most brands use it for exactly one thing: a browse abandonment email.
What we would build instead: a score, not another flow
Here is the architecture we would build. Labeled clearly: this is a proposed workflow, not a case study. It is the ecommerce translation of the Easy Button lead scoring system we run on the B2B side.
Site + Email + Ads + CRM events
→ one event stream (view, click, return, download, reply)
→ weighted score, decayed over time
→ threshold crossed
→ HOT BUYER ALERT
→ human call, SMS, or 1:1 email
Illustrative weights — these are a starting point to argue with, not gospel:
- Product page view: 3 points, capped at 15
- Return visit after 7+ days of silence: 10
- Pricing, financing or shipping page: 12
- Spec sheet, sizing guide or comparison download: 15
- Click on a product or offer email: 8
- Third session inside 14 days: 15
- Add to cart: 20
- Decay: minus 20% of current score per week of total silence
Two design decisions matter more than the numbers. First, decay. A score without decay is a leaderboard of people who were interested in March. Second, the threshold has to produce a list a human can actually work — if crossing 45 points surfaces 600 people a month and you have one person on the phone, your threshold is wrong, not your team.
Routing is where most builds die. The alert has to land in the place the human already lives, with the reason attached — which is a CRM automation problem, not an email problem. “Call Dana. Three sessions this week, opened the financing page twice, downloaded the spec sheet.” Not a dashboard she has to remember to check.
The economics of triggering earlier
Illustrative example — run your own numbers.
Take a brand doing 10,000 monthly sessions at a $2,400 AOV. Roughly 320 carts get created; at Baymard’s 70% rate, about 225 are abandoned. Run those through Klaviyo’s average 3.33% cart flow conversion and you recover around 7 orders — call it $18,000 a month. Good money for an automation that costs you nothing to keep running.
Now score the upstream traffic. Suppose 1.8% of sessions belong to repeat researchers who cross your threshold without ever adding to cart. That is roughly 180 people a month, or nine per business day — a real call list for one person working part-time. Convert 8% of them, which is a reasonable expectation for a live conversation with somebody who read your spec sheet twice, and you get 14 orders. About $34,560.
The cost side is roughly 20 hours a month of human time plus the build. The point is not that the cart flow is bad. Keep it — it is cheap and it works. The point is that the unscored segment sitting upstream of the cart is bigger than the cart segment, and almost nobody is working it. Your next order is more likely hiding there than in a rewrite of cart email number two.
Where this gets uncomfortable
Three honest problems with everything above.
One: if your AOV is under about $100 and the purchase is impulse, this article is wrong for you. Consumables, basics, supplements, low-ticket apparel. Nobody is calling anyone about a $34 order and the unit economics of human contact collapse immediately. The cart flow genuinely is the right architecture. Tune it and go spend your time on acquisition and retention instead.
Two: the measurement problem will probably be what kills your build. Back in 2019, Gartner predicted that 80% of marketers investing in personalization would abandon it by 2025 over lack of measurable ROI or the pain of managing customer data. That prediction is now gradeable, and the directionally honest read is that it landed. Scoring projects die because they get measured by attribution — which happily assigns credit to whatever touched the buyer last — rather than by incrementality.
Three: the fix for that is a holdout, and it costs real money. Randomly withhold the human touch from 10% of everyone who crosses the threshold, keep them there for 60 days, and compare. You will deliberately not call warm buyers. It stings, which is exactly why almost no ecommerce team does it, and why so few can defend their automation spend when someone asks. Run it anyway. It is the only number worth bringing to a budget meeting.
And a fourth, smaller one: if you build the score and it surfaces six people a month, you do not have a detection problem. You have a traffic problem, and you should go fix the top of the funnel first — through content and engagement that produces readable signal and disciplined retargeting that brings researchers back into a scored session rather than just firing impressions at a pixel audience.
The playbook: build version one in a week
Enough to do it yourself, in order:
- Pick one product line — your highest-AOV, longest-consideration SKU family. Do not score the whole catalog on attempt one.
- Inventory the events you already have. Open your email platform’s metrics list and your analytics event list side by side. Most brands find eight to twelve usable behavioral events already firing and unused.
- Add the three events you are missing. Usually: spec or sizing document download, financing or shipping page view, and second-session detection. These are custom events, and every major platform accepts them via API.
- Write the weights in a spreadsheet first. Pull 90 days of history, score your last 50 buyers retroactively, and see what score they hit before purchasing. That is your threshold. Not a number somebody blogged.
- Wire the alert, not a dashboard. Score crosses threshold, a record gets created or updated with the reason string, and a human gets pinged. One workflow. Make.com, n8n, or native automations all handle this.
- Set your holdout on day one. Ten percent, excluded automatically. You will never do it later.
If you also sell through dealers, installers or wholesale accounts, the same event stream feeds a second motion entirely — a scored account list for outbound, which is what our Prospect Pump TAM mining system is built to work.
Your prospects are already talking
Build something that listens. Connect the systems. Track the behavior. Score meaningful engagement instead of waiting on the one event that only fires after the decision is made. Compress thousands of activities into a list short enough that a human can work it before lunch.
Then put your people where they belong: talking to the humans most likely to care.
Want to build it yourself? Steal the framework above — it is the whole thing. Want us to bolt it onto the stack you are already paying for? See what this costs, or work with Jeremy directly for 90 days.