From Batch-and-Blast to Database Marketing: RFM and LTV in Klaviyo

RFM segmentation groups your customers by three behaviors: how recently they bought (Recency), how often they buy (Frequency), and how much they spend (Monetary). Score every customer on those three and you can tell a first-time buyer from a loyal repeat from someone about to lapse, then send each of them a different email. That is the whole idea behind database marketing: you stop broadcasting to one big list and start running it.

Most Klaviyo users never get there. They can send a campaign and they have a welcome flow, but every send still goes to more or less the whole file. This guide is for that person. By the end you will know how to build VIP, at-risk, and win-back segments in Klaviyo, what to send each one, how to layer lifetime value (LTV) and Klaviyo's predictive scores on top of RFM, and how to trigger flows when a customer moves from one segment to another. We run this playbook across our clients' retention programs, so the examples are from real accounts, anonymized.

What RFM segmentation actually is

RFM is a scoring system. You take three facts every store already has, rank your customers on each, and combine the ranks into a segment.

The classic version scores each dimension 1 to 5, which gives you a grid of up to 125 cells. You do not need 125 segments. In practice you collapse the grid into a handful of groups people can actually act on: champions, loyal, at-risk, can't-lose, new, and lapsed. The scoring is just the machinery underneath. What matters is that you stop treating a customer who bought ten times as if she is the same person who bought once and ghosted.

Why database marketing beats batch-and-blast

The best proof for this approach does not come from ecommerce. It comes from a casino.

In 1998, Gary Loveman took over operations at Harrah's, and one number bothered him: for customers who visited Harrah's at least once a year, the company was capturing only 36 cents of every dollar those customers spent on gambling (Harrah's Entertainment case, Harvard Business School, 2001). The other 64 cents went to competitors. His customers were loyal, just not to him.

What Harrah's did next became the founding case study for database-driven marketing. Instead of rewarding customers based on what they had already spent, they built models to predict each customer's future worth and marketed to that. Richard Mirman, who ran the program, called it the secret recipe. Loveman put the principle plainly: they began "building relationships with customers based on their future worth, rather than on their past behavior." They found that 26% of players drove 82% of revenue, so they stopped spending the same on everyone and concentrated on the customers who mattered. Stock price and profit doubled in a year.

The lesson for your email list is direct. Your best customers are hiding in the same file as your worst ones, and if you blast everyone the same offer, you overspend on people who will never come back and underinvest in the ones who would buy again tomorrow.

The DTC numbers say the same thing the casino did. According to Smile.io's dataset of more than a billion shoppers, the top 8% of customers generate 41% of a store's revenue, and the top 10% spend twice as much per order as everyone else (Smile.io). The probability of selling to an existing customer is 60% to 70%, versus 5% to 20% for a new prospect (Marketing Metrics, via Rejoiner). And the retention math compounds: Bain research published in Harvard Business Review found that a 5% increase in retention lifts profit by 25% to 95%, while acquiring a new customer costs five to 25 times more than keeping one (HBR).

How RFM, LTV, and predictive scores fit together

People treat these three as competing methods. They actually stack: RFM first, then LTV, then predictive scores on top.

RFM is the backbone. It is transparent, it runs inside Klaviyo, and anyone on your team can explain why a customer landed in a segment.

Lifetime value is what you are actually optimizing for. RFM tells you who a customer is today. LTV tells you what the whole relationship is worth over time, which is how you decide how much to spend keeping it. A simple constant-retention model makes the point: LTV rises far faster from small, real improvements in retention than from margin changes, because the multiplier explodes as your repeat rate climbs, which is why retention is the number worth pushing on.

Predictive scores are the accelerant. Klaviyo already calculates predicted CLV, churn risk, and expected next order date for accounts with enough order history. You do not have to build a data science model. You have to use the ones sitting in your account.

We run every lifecycle program off the same handful of ideas. RFM decides who gets what. LTV decides how hard you fight to keep them. And the flow-versus-campaign revenue split tells you whether the system is actually built yet. A healthy retail program earns 25% to 40% of email revenue from flows. If you are sitting at 5%, you have not built one.

You do not need a customer data platform or a warehouse to start. You need RFM plus one or two predictive signals inside Klaviyo, and that gets you most of the value. Graduate to a warehouse when you are stitching data across channels or your logic outgrows what Klaviyo segments can express, not before.

How to build RFM segments in Klaviyo

Klaviyo will not label a customer "champion" for you, but it gives you every input: last order date, order count, total and average spend, and predictive scores. You build the segments from definitions. Here is a practical starting set and what to send each one.

Two rules make this work. First, never mail the whole file at once. Segment on behavior and recency, not on "has been sent to." Second, what you send has to match the segment's intent. A champion does not need 20% off; a can't-lose customer might. When we audited a large home and furniture retailer, they were sitting on a four-message win-back flow, never once turned on. It had been in Draft since 2022. The reactivation asset already existed, nobody had ever flipped the switch.

Segment migration models almost nobody builds

RFM segments are not permanent. Customers move. A champion goes quiet and becomes at-risk. A new buyer takes a second order and becomes loyal. The whole point of database marketing is to notice the move and respond to it.

This is where most Klaviyo programs stop. It is also the biggest win you can add. Klaviyo exposes two properties for tracking migration: the customer's current RFM group and their previous one. When someone crosses from Champion to At-Risk, that transition is a trigger. It means a proven, high-value buyer just broke their pattern, and you have a narrow window to pull them back before the habit dies.

Build flows around the transitions, not just the states:

• Champion to At-Risk: the most valuable alert in your account. A best customer is cooling off. Reach out fast, personal, no heavy discount yet.

• New to Loyal: celebrate the second order and move them toward a third. Second and third purchases are where retention takes hold. After one purchase a customer has roughly a 27% chance of buying again; after a second and third, that jumps to 49% and 62% (Smile.io).

• Loyal to At-Risk: a habitual buyer missed their usual reorder window. If you sell consumables, this is your replenishment reminder doing its job.

Harrah's built its entire program around this idea decades ago: a "New Business" phase to earn the second and third visit, a "Loyalty" phase to grow share of wallet, and a "Retention" phase that fired specifically when a customer "broke their historical visitation pattern" (Harvard Business School, 2001). That is a migration model. You can build the same thing in Klaviyo this quarter.

What database marketing looks like week to week

Everything above is the strategy. This is the operating rhythm, the actual sequence of moves we run on a retention account, using a subscription appliance brand as the example (a device with a replaceable consumable customers can put on subscription).

The order we turn flows on

When we inherit a neglected or paused account, we do not switch on every flow at once, and we do not start with the discount-heavy ones. We build the flows for repeat behavior first, before the ones for harvesting demand you already have.

1. Abandoned Checkout first. Highest intent, low volume, and safe to fire while a sending reputation is still warming. On the subscription brand this was the deliberate first flow we turned on.

2. Welcome, buyer and non-buyer versions. The welcome series is close to free forever revenue. We did not retroactively blast it to people who signed up during the warm-up blackout, since an upcoming sale would touch them anyway.

3. The rest of the abandonment set (cart, browse, site), in intent order.

4. Post-purchase education. What the product does, how to get value from it, why the next step matters. This is the on-ramp to subscription.

5. Replenishment. For a consumable this is where revenue compounds. It triggers on time since purchase, time since the last consumable reorder, and device-usage signals, and it carries a convert-to-subscription goal, not just a reorder nudge.

6. Winback, keyed to a missed reorder, including a dedicated flow for cancelled subscribers.

7. Sunset. The safety rail, keeping the dead file from dragging down everyone else.

The weekly operating rhythm

Once the flows are live, the account is a weekly job. Every week we are looking at:

• Deliverability by mailbox provider, not a blended average. We split inbox performance across Gmail, Outlook, Apple, and Yahoo, because problems hide inside one provider. On this account Apple sat around 71% while Gmail and Outlook were the ones to watch.

• List growth. We caught a 16% drop in subscribers over 30 days and negative list growth for the first time, traced it to a traffic-source change, and flagged it. Negative list growth is a leading indicator of falling revenue.

• Flow share versus campaign share of revenue, the master gauge. Campaigns amplify demand; flows create it. If campaigns carry 85% of revenue on a repeat-purchase brand, the lifecycle system is not built yet.

• Send frequency, especially for the best cohort. Email stays on a sensible weekly cadence and SMS is capped at one to two sends a week. As the subscriber base grew, we got more protective, not less, about how often we message it.

• What we are testing. We schedule creative and subject-line tests into softer, non-promo weeks so we are not gambling with revenue during a sale.

A subscription brand, move by move

This is the actual sequence, in order.

We rebuilt the sending reputation before touching the full list. The program had been paused, so instead of blasting the cold file we warmed the IP by mailing the most-engaged customers first and ramping volume gradually, watching bounce and complaint rates each week. It reached full volume in time for the Labor Day sale with no perceived risk.

We tested storytelling against templates, and storytelling won. On product-launch emails we ran a short standard template (version A) against a long-form, story-driven version (version B). Version B significantly beat A on both tests, so storytelling became the default first draft for launches. A separate health education email hit around an 85% open rate, which told us to send education broad rather than narrow-target it.

We found the subscription take rate, and it reframed everything. Adding the consumable subscription to the product pages produced a take rate higher than expected, around 40%. That moved the whole retention thesis. The consumable subscription is where lifetime value actually compounds for this brand, and it beats chasing one-off reorders.

We wired subscription conversion into the flows. The replenishment flow got a second job, converting one-time consumable buyers onto subscription, with triggers at the six-month device mark, time since the last consumable reorder, and a 15%-off-into-subscription path. We pushed the same messaging into post-purchase and winback.

We protected the cohort we worked hardest to earn. As the subscription base grew, the standing rule was to stay more careful about frequency and relevance to that group, because over-messaging your best customers is how you churn them.

We planned the holiday cohort by cohort. Going into Black Friday we pulled last year's numbers and used October to test retention and acquisition offers across cohorts, so the holiday offers were validated before they went out.

When deliverability got hard during a launch (spam flags, provider-specific weakness on Gmail and Outlook), the fixes were seed lists to monitor inbox placement, engagement-based quarantining to stop mailing dead addresses, and registering BIMI and Apple Business Connect. The warm-up recovered and hit full volume for the sale.

Where database marketing quietly breaks

RFM and LTV only work if the data underneath is honest. These are the traps we find most often.

Send-based segmentation instead of behavior-based. If your segments are built on how often you have emailed someone rather than whether they engage or buy, you are measuring your own activity, not theirs. That is how brands end up mailing 300,000 people, including tens of thousands of dead addresses.

Counting recapture as marketing revenue. Some of the "revenue" your ESP attributes to email is money you would have earned anyway. On one audit, a transactional email list showed a click rate above 100%, which is not superhuman engagement. It is machines: link scanners, image proxies, and security bots re-crediting an already-completed sale. Order confirmations are not a marketing channel. If your reporting treats them as one, your numbers are inflated. Treat your ESP's revenue figure as the ceiling and your GA4 or Shopify number as the floor.

Mailing your worst segments to death. In one 18-day window, more than half of a retailer's email sends went to lapsed and mass-blast tiers and produced barely 2% of the revenue. Those sends do not just underperform. They drag your sender reputation down and put your deliverability to good customers at risk.

Ignoring deliverability until it is on fire. A DTC superfoods brand imported a stale list and watched its Klaviyo deliverability score fall to 28 out of 100 in a single month, with bounce rates blowing past 14% against a 1% ceiling. The tell was hidden in plain sight: open rate looked fine around 52%, but clicks sat near 0.57%, about half of healthy. High opens with near-zero clicks and rising bounces is the signature of a list padded with unengaged and invalid contacts, and Apple's Mail Privacy Protection inflates those opens further. This is why a sunset policy matters. Cutting dead contacts feels like shrinking your list, but it protects the revenue-producing part of your list.

Frequently asked questions about RFM segmentation

Where to start

Pick one transition and build for it this week. The Champion-to-At-Risk flow is the highest-value place to begin, because it protects the customers you already earned. Then turn on the win-back you probably already have sitting in Draft, set a sunset policy so you stop mailing the dead file, and start watching the share of email revenue coming from flows versus campaigns. When that number climbs, you have stopped broadcasting and started running a database.