The DTC Retention & Growth Playbook

Welcome series — 5 emails over ~5 days, not 4 over two weeks

• Email 1 immediately on signup; email 2 the same day (+4h, a different angle — founder note or objection handling, never a duplicate); then daily. Urgency lands day 4–5, because conversion falls off a cliff after day 4.
• One selling point per email. Discount visible at the top of every email — "why are you sending welcome email #3 with no discount? That's just a campaign."
• No time-of-day gates: someone who signed up at 9:30am just proved they're on their phone buying at 9:30am.
• A "why didn't you buy?" survey email at the end gets ~5% click rates — free objection research.
• Send SMS simultaneously with email #1 (SMS half-life ~2h, email ~48h).

Abandonment — four distinct flows

• Site → browse → cart → checkout are different intents. Most brands' "cart" flow is actually a mislabeled checkout flow.
• Split every abandonment flow by buyer vs non-buyer at the trigger. Non-buyers: remind them of the welcome discount they already have in emails 1–2, escalate only at 3–4. Past buyers: no discount at first — try full price. Protect margin.
• The subject line that reportedly ~doubles opens on cart/checkout email #1: "Your order is ready to ship." And "we put a mystery discount in your cart" won by +50%–2× clicks in every A/B one agency ran.
• Dynamic product block as high as possible. Cart/checkout get no cross-recommendations (intent is shown); browse abandonment does.

Post-purchase — not a sales flow

• Founder thank-you immediately (not 30–60 minutes later), plain text, cross-sell only in the PS line — "everybody reads the PS," and it beat every dedicated cross-sell email.
• Days 2–3: consumption education — teach the invisible progress so people keep using the product before results show. Days 14–28: a human check-in ("it's been three weeks — you should be noticing X"), which beat every "running low?" email.
• Vary by order count; stop thank-yous after the ~5th order. The #1 reason subscribers cancel is too much product — emails that drive consumption are retention.

Winback — over-rated, rescued by two changes

• Compute the timing from your own data: fire at ~3× your median repeat-purchase interval (a 45-day cycle → day ~135). "9 out of 10 brands get winback timing wrong."
• Only offers reactivate. Lapsed customers already ignored months of campaigns; "we miss you" does nothing. The standout: the store-credit email — official-looking, an odd amount ("$17.63"), an expiry, a minimum spend. Loss aversion does the rest.

The flow portfolio

• Automations are ~2% of sends and ~37% of email orders. Most brands run ~2 flows; the recommendation is 8–12+, reviewed every 30 days. Back-in-stock is the sleeper: lowest volume, ~6% conversion.
• A real account audit: 45 automations built, 10 enabled, 4 core lifecycle flows missing entirely.

• Cadence floor: 3–4 campaigns/week to the engaged (90–120 day) segment. The common failure isn't over-sending — it's under-sending, and mailing the good stuff to 20k-person micro-segments while the 500k engaged list gets one email a week.
• A low unsubscribe rate is a warning sign. Target 0.5–0.7% on sale sends (vs the 0.3% "standard"); read unsubs jointly with revenue per recipient. The welcome series doubles as a deliberate list filter — harvest the "no"s early, on small volume, to protect reputation on the 30× larger campaign volume.
• Spam complaints are enforced on a ~7-day rolling window (0.3%) — sending more good volume dilutes the rate.
• Gmail ignores opens entirely. It scores clicks, replies, read time, scroll depth — plus your website and the full link/redirect chain. Per-user demotion is real: ignore a sender's next 7–10 emails and they slide to Promotions, then spam. Dormant addresses become spam traps on a clock (≈24 months Gmail, ≈12 Yahoo/Microsoft).
• Image-heavy is fine; heavy files are not. Statics under ~100KB, GIFs under 200KB, stay under the clipping threshold. Free machine-readability test: paste your email into a multimodal AI and ask it to summarize — if it can't recover the message, neither can Gmail's classifier or iOS 18's summaries.
• Treat the inbox like a social feed — same customer, same thumb, different app. One message per email; offer + CTA above the fold; subject lines get 6–7 words.
• From a 4,000-campaign creative study: the #1 revenue-correlated attribute is a strong action-verb CTA ("Shop the sale," not "Learn more"). Header nav menus in emails reduced revenue — delete them. Founder-voice emails appeared in 23% of top performers and 0% of the bottom. Social proof mattered far less than assumed.
• The personalization paradox: light personalization (first-name at big moments, send-time) returns the most per hour; deep 1:1 projects make "almost no money" and become maintenance liabilities. A well-tested plain template beat quiz-personalized welcome variants head-to-head.

• Framing > depth: BOGO reframes; "gift" language beat equal-value discounts 3–5× in one network's data; store credit with expiry; mystery discounts (judge on conversion, not opt-in).
• Display rule: show whichever number feels bigger — "$100 off" for high AOV, "20% off" for low.
• Biggest offer first, urgent and time-boxed. Escalating ladders (10% → 15% → 20%) just delay revenue.
• Consensus anti-patterns: never gift-with-purchase your slow-moving stock ("it never works, hurts deliverability, cuts LTV"); avoid tiered "more you buy" discounts outside BFCM; subscription discounts above ~30% correlate with churn spikes (ProfitWell, 2,000+ subscription businesses).
• Premium brands: build a value stack — bundle existing perks (consult calls, warranties, fast shipping) into an advertised offer — instead of ever discounting. One $2,000-chair brand "has been ripping" on exactly this.
• Purchase-entry giveaways are a full playbook: dollars-spent-as-entries toward an aspirational prize collapsed one brand's CPMs from ~$20 to ~$8 and took a dead store $0 → $2.6M in 30 days. Cash is the worst prize; the winner-reveal follow-up drove 45% repeat rates. (Legal caveats — free entry route required.)
• BFCM offers get a 6-month runway: split 10–30% of the audience in each summer/fall promo, read on profit, repeat across promos. "That offer will make or break your year, and you're deciding it in October."

• A multi-step quiz popup (question first, email second, SMS third) engages 40–50% at step one vs ~5% for a bare email ask. Typical result of a popup rebuild at one agency: 2.7× list growth. The question barely matters — it's commitment psychology.
• Header framing: "You've GOT 10% off" (already yours; leave and you lose it) beats "get 10% off."
• Benchmarks: with an offer, 10–20% opt-in of viewers; without, 5–10%. Delay ~10s on the homepage; quarter-scroll on landing pages — 60–70% of paid traffic never sees your homepage, so build funnel-specific popups.
• Test cashback vs discount: "25% cash back on your 2nd purchase" beat 10% off by up to +40% — and redemption forces a repeat purchase.
• The contrarian view worth testing: discount popups recruit discount-loyal customers. One $30M bootstrapped brand grew a 1M-person list selling access (scarce drops, harvest weeks) instead of coupons — with a 40–50% repeat rate.

• First-order take rate is the compounding lever: 70–87% of everyone who ever subscribes does it on their first order. Same traffic and churn, 60% vs 15% take rate → an ~8× larger subscriber base within a year. Benchmarks: >50% excellent · 30–40% solid · <20% urgent. Five plan-picker changes tripled one brand's take rate in three months.
• Churn is spiky: cancellations cluster at days 27/30 and 57/60 (first and second renewal) and at the order-3→4 transition. The default "your order is about to process" reminder causes cancellations — rebuild it with excitement, a renewal gift, and one-click skip/delay (skip-easier-than-cancel cut cancellations ~24%).
• Passive churn (failed payments) is 10–20% of total churn. Quiet retry windows, decline-code segmentation, and account updaters are existential.
• Flexibility is the product: 53% of 100M subscribers analyzed had altered their subscription. Pause/skip/frequency/flavor options beat retention theater; confusing subscription UX destroyed ~$1M at one brand via accidental duplicates.
• In-cycle upsell: hit subscribers within 24h of a delivery (the unboxing high) with a prepaid 3/6-month upgrade + gift — higher LTV and cash up front.
• Winback the file: brands typically hold 7–8× more lapsed subscribers than active — reactivation targeted by cancellation reason is the cheapest growth available.
• Define LTV honestly: windowed (12/24/36-month), on contribution profit (after discounts, returns, fees, COGS, shipping), split new vs returning, per channel. "Lifetime" without a window and "value" meaning revenue is how brands lie to themselves.

• The holdout method: hold 10% (up to 50% on small lists) out of any change — an extra weekly email, a new flow — and read order data for both groups a month later. Re-read the same cohort at 3–4 months for fatigue: over-sending damage shows up 6–12 months later, never week one.
• Incrementality coefficients (one 9-figure operator's working numbers): branded search ≈ 0.2 (an 8× platform ROAS is really ~2×) · Meta 1-day-click ≈ 3.0 · Amazon non-brand ≈ 1.5. Sanity check: Σ(spend × coefficient) should ≈ your P&L revenue.
• Cash beats ROAS: a 4× campaign with 30-day payback beats a 6× with 120-day payback "almost every time." Real budget capacity runs 20–30% below pure LTV math.
• Plan-vs-actual as an operating system: forecast every flow/campaign as an action-attached revenue commitment; read daily gaps; gap size dictates how extreme the correction must be. One team recovered a $289k-vs-$550k Black Friday day within 48 hours because the gap was visible immediately.
• Contribution framing: the median brand runs ~87% net revenue / 57% gross margin / 28% contribution margin / 7% EBITDA — at those margins, revenue dashboards actively mislead. Watch the active customer file (~100-day window) as the leading indicator; a blended P&L can hide dying acquisition behind a big file for years.
• Retention benchmarks (agency composites): 60-day incremental repeat revenue ideal ≈ +30% of first order (avg 21%); 1-year ideal ≈ +100% (avg 50%). One $100M apparel operator's ladder: 1st→2nd purchase 38%, 3rd→4th 60% — three purchases makes a lifer.

"You don't guess right — you make it right."

• Creative IS the targeting. Meta's algorithm reads every element — visuals, audio, copy, landing page — and picks the audience itself: "your content is the prompt." Casting a grandmother replaces building a 55+ ad set. One brand moved delivery from 22–34% female to 83% female over a weekend by changing copy alone.
• Don't prune to winners — feed diversity. Split tests became personalization inventory; cutting to the winner tanked campaigns. Landing pages now outweigh ad copy in Meta's indexing — run multiple page variants permanently.
• Evaluate angles, not ads. Ads die in ~2 months; a coded angle catalog — mined from customer language (reviews, ad comments, support tickets → pain points, benefits, objections, failed alternatives, emotional register) — compounds forever. Filter reviews to >300 characters before mining; short praise has no signal.
• Offer math gates everything: with a converting offer "almost all creatives work." One $1M+/day operator manages purely to per-offer 12-month lifetime contribution margin (100% return on CAC = gold standard).
• Organic is the paid filter: post at volume, watch over-indexers vs your own baseline, and put money only behind proven assets — "asinine to spend $1 of media on creative not validated by an algorithm." One TikTok-native brand runs 140K affiliates producing 10–15K videos/month and only Meta-boosts what already went viral (~$190M revenue).
• Secondary channels: judge on new-customer rate and cohort quality, never blended ROAS; port your proven Meta creative verbatim first. And in a deal-conditioned economy, winners run 4+ engineered demand peaks a year, using "excuse" moments (invented holidays, giveaways) as funnel pre-fill between real discounts.
• AI in creative: strong for research mining, briefs, statics volume, and intentionally unreal visuals — but even AI-forward agencies ship <20% AI-generated creative, and nobody reported a fully AI-made video ad winning at scale. Un-replicable formats — real founders, real experts, behind-the-scenes — are the hedge as feeds fill with synthetic content.

• Generic vs connected: unconnected chatbots give "absolutely logical but generic" advice. Connected to your store/ESP data (via MCP), they audit flows against benchmarks, build real segments, and draft campaigns in ~4 minutes instead of 30–60. Rules of the road: use reasoning models for connected work, start read-only, graduate to drafts-with-approval, never full auto — "a junior assistant: faster than any human, still makes dumb mistakes."
• Context engineering beats prompt engineering (~90% of the value): a condensed brand pack — voice, products, personas, your top-5 winning and bottom-2 losing sends — outperforms both bare prompts and 48-page brand-guideline dumps. Keep "don't do" lists short; models over-index on banned words.
• Org lessons from AI-native companies: handing out licenses does nothing — forced whole-company "AI days" drove adoption. Comprehensive agents get abandoned; descoped tactical slices stick (QBR prep: 30 → 5 minutes). Every agent needs a named human owner. AI support resolved 40% of tickets at human-parity CSAT — with two full-time supervisors.
• One 9-figure apparel brand runs global marketing with <10 people by collapsing handoffs — designers own campaign P&L with AI tooling. "Delete handoffs, not tasks."
• The 90% rule: 90% accuracy is fatal for customer-facing chat and transformative for internal tagging, sentiment, and trend-mining, where the baseline is zero.
• The durable human skill: asking the right questions, owning priorities and taste. "It's not humans versus AI; it's humans who employ AI versus humans who don't."

All talks are on the Commerce Roundtable YouTube channel. Starred () = start here.