Predictive retention for e-commerce brands

Know who's about to churn.
Before they're gone.

Growing brands lose repeat customers silently — no complaint, no unsubscribe, just a quiet drop in reorders. Berta scores every customer's churn risk daily and triggers the right save before the window closes.

Book a call See demo

30 minutes · No commitment

3–5×

cheaper to retain a customer than acquire a new one

68%

of churn happens silently — no complaint, no flag, just no reorder

24hr

risk scores refresh daily, not in a monthly report

What most CRMs miss

No risk score

You find out a customer's gone after they've gone

Static discount rules

Same 10% code sent to a healthy customer and a lost one

No reorder timing

Win-back emails land months after the window closes

Siloed data

Orders, tickets, and email tell three different stories

With Berta

Every customer scored daily, segmented by risk, with the right re-engagement triggered automatically — before silence turns into churn.

Your best customers are leaving quietly, and your tools can't see it.

Churn is silent

By the time a customer unsubscribes, they've already been gone for months.

Most retention tools react to a cancellation or an unsubscribe — signals that arrive long after the customer has actually checked out. Without a predictive layer on your customer data, retention is reactive at best, and the highest-value save window has already closed.

Discounts aren't a strategy

Blanket win-back campaigns waste margin on customers who were never leaving.

A 90-day "we miss you" email goes to everyone on the same clock, whether they're a healthy customer with a longer natural cadence or someone who's genuinely at risk. Undifferentiated discounting trains your best customers to wait for a code instead of actually saving the ones about to churn.

Data lives in five places

Orders, emails, tickets, and sessions never get combined into one signal.

Your store platform knows the orders. Your ESP knows the opens. Your helpdesk knows the complaints. None of them talk to each other, so no single tool sees the full pattern — declining engagement plus a slipping cadence plus an unresolved ticket — that actually predicts churn.

One engine. Continuous scoring, automatic action.

01

Connect

We connect to your store, ESP, helpdesk, and payments — Shopify, Klaviyo/Attentive, Gorgias, Stripe. No manual exports.

02

Unify

Every customer's orders, sessions, emails, and tickets are merged into one timeline instead of five disconnected tools.

03

Score

A model tuned to your category's real purchase cadence scores every customer's churn risk daily — not a generic days-since-order rule.

04

Segment

Customers are tiered into healthy, at-risk, silent-churned, and win-back — each with its own playbook.

05

Trigger

The right re-engagement fires automatically through your existing ESP/SMS tools — or gets flagged for a human on your highest-value accounts.

Output A · Risk dashboard

See exactly who's at risk, and why

Every customer's risk score, the signals driving it, and the dollar value at stake — updated daily, not buried in a monthly report.

Output B · Automated re-engagement

The save happens without anyone remembering to check a dashboard

Risk crossing a threshold triggers a workflow through your existing email/SMS tools — sized to the customer's value, not a blanket discount code.

Four systems. One score driving all of them.

You don't need four separate tools — connecting your data once produces the score, the segments, the triggers, and the reporting.

01Predictive

Churn Scoring

Predictive risk model

A churn-risk score for every customer, updated daily.

We train a model on your brand's actual purchase cadence — supplements reorder monthly, furniture doesn't — rather than applying a generic days-since-last-order rule. Signals include cadence drift, engagement decay, support sentiment, and discount sensitivity, combined into one score per customer.

Cadence-aware risk scoring Signals from orders, sessions, tickets, email Daily score refresh Revenue-at-risk in dollars
02Automated

Re-engagement

Automated win-back triggers

The right message, to the right customer, at the right time — without a human remembering to send it.

Once a customer's risk crosses a threshold, a workflow fires automatically through your existing ESP or SMS tool. Offers are sized to the customer's value instead of a flat discount code, and your highest-LTV at-risk accounts get flagged for a human instead of an automated email.

Threshold-based automated triggers Value-sized offers, not blanket discounts Runs through your existing ESP/SMS Human flag for high-LTV accounts
03Predictive

Reorder Prediction

Anticipate the window, don't chase it

Predict when a customer is likely to reorder — and reach them before, not after.

For consumable and subscription-adjacent products, we predict each customer's next likely reorder window and nudge just before it. This is retention by anticipation instead of retention by rescue, and it reduces how often you need to lean on discounting at all.

Per-customer reorder window Pre-emptive nudges Best for consumables & subscriptions Reduces reliance on discounting
04Reporting

Revenue-at-Risk

One number leadership actually watches

How much revenue is at risk this month, and how much you recovered.

Cohort retention curves and campaign opens are useful for a marketer; a dollar figure is what gets budget approved. We report revenue-at-risk and revenue-recovered alongside the underlying cohort and channel data, so the win-back program's ROI is never a guess.

Cohort retention curves Revenue-at-risk dollar figure Win-back campaign performance LTV by acquisition channel

Churn looks different depending on what you sell.

A generic RFM model treats a supplement brand and a furniture brand the same way. Our cadence models don't — we define what "normal" looks like for your category before the score runs.

🔄

Subscription & replenishables

Cadence is everything — a 3-day delay past the reorder window is a red flag, not noise.

Reorder window prediction
Cadence drift detection
Pause vs. cancel signal separation
Subscription-specific win-back flows
👕

Apparel & discretionary

No fixed cadence — risk shows up as engagement drop-off and browse-without-buy patterns.

Session & engagement decay scoring
Seasonal cadence normalization
Size/fit return correlation
Discount sensitivity modeling
🛋️

High-ticket & considered purchase

Long gaps between orders are normal — the real signal is engagement, not silence.

Long-cycle risk modeling
Post-purchase support signal tracking
Referral & repeat-category tracking
Warranty & service trigger points
💄

Beauty & fast-cycle consumables

Fast repeat cycles mean early warning is possible just days after a missed reorder.

Fast-cycle cadence tracking
Ingredient & variant preference signals
Bundle vs. single-SKU behavior
Early-warning trigger thresholds

We plug into the tools you already run.

No new tool for your team to check. Berta reads from and writes back into the platforms you already use.

Shopify

Store & order data

Klaviyo / Attentive

Email & SMS triggers

Gorgias / Zendesk

Support signal

Stripe

Payment & subscription data

"We don't send another discount email. We tell you who's actually about to leave, and why."

Earlier signal

Weeks of warning

Risk is flagged before a customer goes silent, not after — while there's still time to act.

Targeted spend

Margin protected

Win-back budget goes to customers who are actually at risk, not a blanket segment.

Tracked recovery

Dollars, not opens

Revenue recovered is measured directly — not inferred from email open rates.

Compounding

Sharper over time

Every order, ticket, and campaign response makes the next score more accurate.

Built by operators, not consultants.

Retention modeling

We build cadence models specific to your category, not a generic RFM score applied to every brand the same way.

Data pipelines & integrations

We build and maintain the sync layer connecting your store, ESP, helpdesk, and payments into one customer timeline.

Behavioral triggers

Re-engagement fires automatically through your existing tools — we're not asking your team to check another dashboard.

E-commerce operations background

We come from high-volume e-commerce, where a missed churn signal is quantifiable revenue, not a vague concern.

"We don't add another tool to your stack. We build the layer that tells you who's leaving before they're gone."

Built for brands where repeat revenue can't be left to chance.

Brands with a growing repeat base

You have thousands of repeat customers but no system that tells you which of them are slipping before they're gone.

Subscription & consumable brands

Reorder cadence should be predictable — most brands have the data to use it and aren't.

Brands sitting on CRM data

You have a CRM and a helpdesk, but neither one catches churn before it happens. We build the layer that acts on the data you already have.

Questions we get on every call.

How is this different from Klaviyo's win-back flows?

Klaviyo flows run on fixed segment rules — "no purchase in 60 days," applied to everyone the same way. Berta scores each customer's individual predicted risk daily and triggers through Klaviyo when that risk crosses a threshold, so the timing and offer are specific to that customer, not a fixed calendar.

Do you replace our ESP or CRM?

No. We connect to your existing ESP, helpdesk, and store platform and trigger through them. Berta is the scoring and decision layer sitting on top, not another tool your team has to learn.

What data do you need access to?

Order history and store data, email/SMS engagement, support tickets, and payment/subscription status where relevant. Session-level browsing data improves the model but isn't required to start.

How accurate is the score early on?

The model calibrates to your brand's actual cadence over the first 4–8 weeks as it accumulates order history. Early scores are directionally useful; accuracy compounds as more customer history feeds the model.

Do you handle subscription and one-off brands differently?

Yes. Cadence, signals, and triggers are tuned per business model — a subscription brand's risk signals look nothing like a high-ticket furniture brand's, so we don't run the same model on both.

What does the engagement look like?

We start with an audit of your current customer data and repeat-purchase patterns, then connect your stack and build the scoring model. Once live, we monitor and refine continuously — this is ongoing, not project-based.

Send us your customer data.
We'll show you who's about to leave.

We'll walk through your current repeat-purchase data, run a live risk audit against your category's cadence model, and show you exactly which customers we'd flag first — before you commit to anything.

Book a call See demo

30 minutes · No commitment