TL;DR
- Don’t ask AI why a shopper left — it can’t know. Build a 3-email sequence where each email answers one of the real, measured abandonment causes: shipping/extra cost (39% of abandons per Baymard 2026), forced account creation (24%), slow delivery (21%), card-trust doubt (19%).
- Klaviyo’s own benchmark for abandoned-cart flows (as of June 2026): ~50.5% open, ~6.25% click, ~3.33% placed-order rate, and ~$3.07–$3.65 revenue per recipient. Top performers hit ~7.69% placed-order and ~$28.89 per recipient.
- Three emails beat one by a wide margin: a Klaviyo cohort study attributed $24.9M to 3-email flows vs $3.8M to single-email flows (~6.5x).
- Send cadence that still holds up in 2026: Email 1 at ~1 hour, Email 2 at ~24 hours, Email 3 at ~48–72 hours. Keep discounts out of Emails 1 and 2.
- Use GPT-5.5, Claude Sonnet 4.6, or Gemini 3.1 Pro to draft 5–8 subject variants per slot and match your brand voice from 2–3 sample emails. Always exclude purchasers at the flow level, not per email.
The task
It is Tuesday afternoon and your Klaviyo dashboard says 72% of carts abandoned last week — right on the Baymard Institute’s June 2026 average of 70.22% (mobile runs higher, 73–75%). Recovery is sitting at 1.8%, well under Klaviyo’s ~3.33% placed-order benchmark. The current flow is the default Shopify template — Email 1 has the cart image, Email 2 has 10% off, Email 3 has 15% off. Open rates are fine, click rates are not, and your CMO wants to know why a customer who put a $90 sweater in the cart needs to be discounted to come back. You want a 3-email sequence that earns the return click — without training every shopper to abandon on purpose for a discount.
Where AI helps — and where it does not
A current model — GPT-5.5, Claude Sonnet 4.6, or Gemini 3.1 Pro — is good at three things in this flow: pacing the cadence so each email has a different job, writing 5–8 subject-line variants per slot so you can A/B without writing them all yourself, and matching tone to your brand voice once you paste 2–3 real past emails as a sample. All three models have a 1M-token context window as of June 2026, so you can drop in your full brand guide plus past flows without truncation.
What AI cannot do: tell you why a specific customer abandoned. It cannot read minds or session replays. The fix is to stop guessing and instead use the measured reasons. Baymard’s 2026 survey ranks the real causes: extra costs like shipping (39%), forced account creation (24%), slow delivery (21%), didn’t trust the site with card info (19%), checkout too complicated (18%), no upfront total (17%), poor return policy (15%). Build a sequence where each email answers one of these in parallel — Email 1 the “I’ll think about it” doubt, Email 2 the trust and shipping concern, Email 3 the price ceiling.
A common failure mode: the model defaults to a discount ladder (no offer → 10% → 15%). That ladder is the laziest possible sequence and it trains repeat buyers to abandon on purpose. Tell the model explicitly: a discount appears only in the last email, and only if your margin allows it.
What to feed the AI
- Product category and typical AOV — a $20 candle and a $400 jacket need very different sequences
- The 2-3 abandonment reasons you actually hear in support tickets (not the ones you assume)
- Brand voice anchor — paste 2-3 real past emails you liked, so tone is not generic
- Incentives you are willing to offer (free shipping at $X, 10% off, free gift, no discount at all)
- Whether the cart contained a single SKU or multiple — single-SKU sequences should name the product, multi-SKU should reference category
- Customer status — first-time vs returning; returning buyers should not get the same trust-builder copy as cold prospects
- Hard exclusions — anything you do not want said (no fake urgency, no “only 2 left” unless true, no “we miss you”)
- Whatever your strongest non-discount lever is — a review snippet, a 30-day return policy, a styling guide, a sizing tool
What “good” looks like — June 2026 benchmarks
Set your targets against Klaviyo’s published flow benchmarks before you blame the copy. If your open rate is near 50% but placed-order rate is under 2%, the problem is the offer and the job of each email, not the subject line.
| Metric (abandoned-cart flow) | Klaviyo average | Top performers |
|---|---|---|
| Open rate | ~50.5% | 55%+ |
| Click rate | ~6.25% | 9%+ |
| Placed-order rate | ~3.33% | ~7.69% |
| Revenue per recipient | ~$3.07–$3.65 | ~$28.89 |
Source: Klaviyo abandoned-cart benchmarks and Baymard cart-abandonment statistics, both June 2026. The single biggest structural lever: a Klaviyo cohort study attributed $24.9M in recovered revenue to 3-email flows versus $3.8M to single-email flows — roughly 6.5x. That gap is why the prompt below builds three distinct emails, not one repeated three times.
Copy-ready prompt
Write a 3-email abandoned cart sequence for an ecommerce brand.
Product: {category + AOV + single-SKU or multi-SKU}
Brand voice: {paste 2-3 past emails or describe in 1 sentence}
Real abandonment reasons (from support tickets): {top 2-3}
Available incentives: {list, with rules}
Strongest non-discount lever: {review, return policy, sizing, etc.}
Customer status filter: {first-time / returning}
Hard exclusions: {fake urgency, "we miss you" language, etc.}
Return:
Email 1 (sent at 1 hour) — no incentive, no discount. Address the "I will think about it" doubt by offering one specific useful thing (sizing guide, fit comparison, review of the exact item). Subject must not contain the word "cart" or "forgot."
Email 2 (sent at 24 hours) — address the strongest non-price concern (trust, shipping speed, return policy). Include one social proof element. Offer free shipping only if margin allows.
Email 3 (sent at 48 hours) — last call. If allowed, the largest incentive goes here. Lead with the strongest review or use-case, not the discount.
For each email return:
- 5 subject line variants under 50 chars (skip the obvious ones)
- 1 preview text under 90 chars that does not repeat the subject
- 60-90 word body
- Single CTA button text (3-5 words)
- One short P.S. line that adds value, not pressure
Tone: helpful, specific, not desperate. Replace any phrase you would not say out loud.Shorter variant — single replacement email
Rewrite our current abandoned cart Email {1, 2, or 3}. Here is the current copy: {paste}.
The current email's problem: {what underperforms — open rate, click rate, unsubscribe spike}.
Constraint: keep the send time slot but change the job. The new email must do {specific job}.
Return 3 versions: one playful, one direct, one expert-led. 60-word body each.Sample output
A good Email 1 subject that beats “You forgot something”: “Quick question about the navy sweater” — frames it as a question the brand wants to help with, not a sale to close. Body opens with “Saw you were looking at the navy crewneck — the most common question we get on this one is fit. It runs true to size in the chest, half a size big in the sleeves. Here is what 4 different heights look like in it: [link].” No CTA to “complete your purchase.” The CTA is “See the fit photos.”
A good Email 2 body opening (trust concern): “Wanted to flag — every order ships from our Brooklyn warehouse, and returns are 30 days, no questions, including final sale. The most cautious shoppers I talk to say the return policy is what makes them finally click. So in case that was the hesitation: it is genuinely no risk.” Closes with one verified review of the exact SKU.
A good Email 3 P.S.: “P.S. If the price is the part holding you up, reply to this email and we will see what we can do. We mean that.” — more powerful than a 10% off code because it routes high-intent abandons to a human and protects margin on low-intent ones.
How to refine
- If Email 1 feels promotional: “Email 1 must not contain a CTA to buy. The CTA links to information (sizing, reviews, comparison). The purchase is the customer’s idea, not ours.”
- If subject lines feel templated: “Write 5 subject lines that sound like a friend texted you, not a brand. Do not use the words ‘forgot,’ ‘still,’ ‘waiting,’ or any emoji.”
- If the discount escalation feels predictable: “Replace the discount ladder. Email 2 offers free shipping, not money off. Email 3 offers a styling consult or a curated alternative, not a bigger discount.”
- If the sequence reads pushy: “Cut every sentence that would not survive being read aloud to a friend. If the email pretends to be more urgent than the situation, rewrite.”
- If multi-SKU emails feel generic: “Reference the category and price range, not ‘your cart.’ E.g., ‘About those two crewnecks’ beats ‘about your cart.’”
Common mistakes
- Discount in Email 1: teaches your repeat buyers to abandon on purpose; you are now paying margin for behavior you trained.
- Same hero image in all three emails: inbox previews look identical, the second send gets archived in 1 second.
- No actual reason to come back beyond “still interested?”: that is not a reason; the email is a notification, not a pitch.
- Fake urgency: “only 2 left!” when there are 200 destroys trust the one time it matters; smart shoppers know.
- Ignoring product type: a $20 candle needs 1 short email; a $400 jacket can carry a 4-email considered-purchase sequence with a styling email in slot 3.
- Not excluding purchasers: Email 3 hitting someone who bought 12 hours ago tanks NPS more than any single bad subject line.
- Treating returning buyers like cold prospects: they already trust you; the trust-building copy reads condescending and they unsubscribe.
- No mobile preview check: 70%+ of opens are mobile; a 70-char subject that reads great on desktop gets truncated to “Quick question about the navy…” which is actually fine, but “We saw you were looking at…” is not.
FAQ
- When should the sequence stop firing if they buy?: Immediately. Use Klaviyo or Shopify’s “placed order” exclusion at the flow level, not just per email. Feed the “Checkout Completed” event server-side within ~5 minutes so the flow can exit before Email 1 fires. Sending Email 3 after a purchase costs more in unsubscribes than any recovery is worth.
- How much can a recovery flow realistically add?: As of June 2026, Klaviyo’s average abandoned-cart flow places an order on ~3.33% of recipients at ~$3.07–$3.65 revenue per recipient; top performers reach ~7.69% and ~$28.89. If you see a placed-order rate above ~10%, double-check your “abandoned” event definition — it may be counting stale carts or browse-abandons rather than true checkout starts.
- One email vs three?: Klaviyo’s cohort data shows 3-email flows generated ~6.5x the recovered revenue of single-email flows ($24.9M vs $3.8M). Klaviyo’s own guidance is 2–3 messages for optimal performance. Three is the right answer for AOV > $100 and considered purchases; for AOV < $30, two emails is plenty and three reads desperate.
- Should I include a discount code?: Default to no, especially Emails 1 and 2 — extra cost (39%) is the top abandonment driver per Baymard, so removing shipping cost often beats cutting price. If margins allow, Email 3 can carry a discount. Better lever for premium brands: free shipping at a threshold the cart already exceeds, which gives a real reason to return without resetting the price anchor.
- What about SMS cart recovery?: SMS converts harder than email (commonly cited at 15–40% vs email’s 5–15%) but legally needs prior express written consent and 10DLC registration in the US, so coverage is limited to opted-in shoppers. Typical 2026 cadence: 1 SMS at ~30–60 min plus email at ~1h / 24h. Do not mirror the email body in SMS; SMS gets the question, email gets the context.
- Which model should I use to draft these?: Any of GPT-5.5 ($20/mo ChatGPT Plus), Claude Sonnet 4.6 ($20/mo Claude Pro), or Gemini 3.1 Pro ($19.99/mo Google AI Pro) handles this well as of June 2026; all three carry a 1M-token context window, so you can paste a full brand guide plus past flows. Claude tends to hold a consistent brand voice across the three emails with the least prompting.