Build or Buy Predictive Churn Scoring on Shopify?
Predictive churn scoring is a BUILD, because no app verified for this decision states a per-customer risk score. Peel reports retention, LTV and churn rate after the fact at $499 to $899/month (verified Sep 2026), and Shopify's native segmentation predicts a spend tier with nothing churn-specific. A model scored off your own order history runs $20,000 to $60,000 (Deploi estimate, illustrative).
Your profile — see how the verdict shifts
- Confidence
- High — Checked Peel's App Store listing and Shopify's customer segmentation documentation on 2026-09-03. Peel is real and strongly reviewed at 5.0 stars across 35 reviews, and its listing describes customer lifetime value, retention and subscription growth insights — descriptive metrics, with no mention of a predictive churn-risk score or an at-risk-customer flag. Shopify's segmentation examples page references a predicted spend tier filter and says nothing about churn risk, at-risk customers or an attrition indicator. A second candidate named in research could not be located under any current App Store slug and was dropped rather than cited. So retention reporting is buyable and per-customer prediction is not.
- Reference scenario
- $20M–$100M GMV · roughly 60,000 repeat customers · Klaviyo flows firing on a fixed 90-day inactivity window · no data scientist on staff
- As of
- September 2026
Decision at a Glance
| Your profile | Verdict | Why |
|---|---|---|
| Under 20,000 orders a year | WAIT | A model needs repeat depth to beat a rule, and at this volume it will not. Segment on each customer's own purchase gap instead and revisit when repeat revenue gets big enough to argue about. |
| 20,000–200,000 orders a year, single storefront | BUILD | Enough history to fit something useful, and a fixed 90-day window is demonstrably wasting both save offers and salvageable customers. This is the band the build was designed for. |
| Subscription or replenishment mix with predictable cadence | BUILD | A real skip, pause or cancel event gives the model a clean label to learn from, and the save action is concrete rather than a discount guess. The signal is strongest exactly where the intervention is easiest. |
| Retention already reported by a paid analytics subscription | CUSTOMIZE | Keep the dashboards your finance team reads and build the per-customer score on top. The reporting layer and the prediction layer are different products, and only one of them is for sale. |
What Predictive churn scoring for repeat customers Actually Drives
| Outcome | Impact | How it works |
|---|---|---|
| Retention & LTV | High | A score timed to each customer's own cadence moves the win-back from post-mortem to intervention, reaching a monthly buyer weeks before a fixed 90-day rule would. |
| Revenue — indirect | High | Save offers land on customers who were actually going to lapse rather than on everyone who happened to cross a date threshold, which changes both the response rate and the discount bill. |
| Data & insight | High | The feature set behind a churn score — cadence, category mix, discount dependence, support history — feeds LTV prediction, paid audiences and buying decisions from the same pipeline. |
| Operational efficiency | Medium | A per-customer score gives retention and CX a ranked list to work each week instead of a cohort chart nobody can act on. |
Spend ceiling: Size the spend against retained margin, not against the subscription price. On 60,000 repeat customers, pulling one percent back from lapse at a $120 average order is roughly $72,000 a year (illustrative), which is the number the build has to beat and the reason a score with no save action attached is worth nothing.
What buying enables (top apps)
- + Cohort retention, repurchase rate and churn-rate reporting without a data team, on a 7-day trial before committing
- + A free tier for Smartrr users covering up to 16,000 orders a month across 3 stores (verified Sep 2026)
- + Lifetime-value and subscription growth views that finance and marketing can read from the same screen
- + Shopify's native predicted spend tier as a segmentation filter, included on every plan
What building additionally unlocks
- + A per-customer risk score written back as a metafield, so segments and email flows target individuals rather than a date rule
- + Churn defined against each customer's own purchase cadence, which is the difference between reaching a monthly buyer and missing them
- + The reason behind each score, so the save offer matches the cause instead of defaulting to a discount
- + A reusable feature pipeline feeding LTV prediction, paid audience exclusions and demand planning from the same data
Find Your Verdict in 3 Questions
Do repeat customers carry a meaningful share of revenue, with at least two years of order history behind them?
Yes: Go to question 2.
No: Your verdict: WAIT. A model has nothing to learn from yet; segment on recency and revisit once repeat revenue matters.
Is there a save action ready to fire — an offer, a call, a swap — the moment a customer gets flagged?
Yes: Go to question 3.
No: Your verdict: WAIT. A score with nothing behind it is a dashboard, so design the save motion before the model.
Do you already pay for a retention analytics subscription that reports churn after the fact?
Yes: Your verdict: CUSTOMIZE. Keep the reporting layer and build the per-customer score on top at $20,000–$60,000 (Deploi estimate, illustrative).
No: Your verdict: BUILD. Nothing verified for this decision ships a risk score, so the definition, the model and its activation are yours.
The TCC Scorecard — 12 Dimensions
TCC — Total Cost of Capability: what it actually costs to have this capability over three years, whichever way you get it. Each dimension is scored 0–5 for both paths. How we score →
| Dimension | Buy | Build | Why |
|---|---|---|---|
| Cost | |||
| Acquisition & implementation | Peel installs and backfills in days on a 7-day trial, while a scored model and its write-back take an estimated 6–12 weeks (Deploi estimate, illustrative). | ||
| Recurring fees | Analytics subscriptions run $499–$899/month at Peel's paid tiers (verified Sep 2026) and never stop, while a model costs compute plus periodic retraining. | ||
| Maintenance & upgrades | A vendor absorbs its own upkeep, while a model drifts as your customer mix, cadence and discount behavior change and needs retraining on a schedule. | ||
| Switching & exit | Dashboards and computed cohorts live in the vendor's warehouse, while scores written back as customer metafields stay on your own records. | ||
| Risk | |||
| Vendor risk | Peel is well reviewed at 35 reviews, though this category is thin: a second widely-cited retention app could not be found under any current App Store slug. | ||
| Security & compliance surface | Either lane processes customer purchase history, and a build keeps that history inside infrastructure you already assess rather than in another vendor's. | ||
| Platform-deprecation exposure | Orders, customers and metafields are stable Admin API resources, and a score written to a metafield survives theme and app changes entirely. | ||
| Value | |||
| Fit to requirement | Retention reporting answers how many customers lapsed last quarter, and the requirement here is which customer is lapsing this week. | ||
| Time to market | Dashboards arrive in a week; a scored, activated model that marketing trusts takes a quarter including the argument about thresholds. | ||
| Performance & scale | Scoring 60,000 customers nightly is a small job, and the built lane scales with customers rather than with an orders-per-month plan cap. | ||
| Data ownership & AI-readiness | The decisive dimension: the feature set behind a churn score — cadence, category mix, discount dependence, support history — is reusable for LTV, paid audiences and buying. | ||
| Focus & opportunity cost | Nobody should build retention dashboards, and the score is only worth building where a save action already exists to fire it into. | ||
The App Landscape
| App | Status | Pricing | Best for |
|---|---|---|---|
| Peel: Retention Analytics | Live — 5.0★, 35 reviews. Reports customer lifetime value, retention and subscription growth, including repurchase, retention and churn-rate metrics. The listing describes descriptive analytics: it does not state a predictive churn-risk score or an at-risk-customer flag ahead of time. Read it as the measurement layer rather than the prediction layer. | Free for Smartrr users up to 16,000 orders/month across 3 stores; Essentials $499/month or $5,389.20/year; Accelerate $899/month or $9,709.20/year; 7-day trial on both paid plans (verified Sep 2026) | Cohort retention and LTV reporting that finance and marketing can both read without a data team |
| Shopify customer segmentation | Native — First-party Shopify surface. Segmentation supports a predicted spend tier filter for marketing, which is a spend prediction rather than a churn prediction. Churn risk, at-risk customers and attrition indicators appear nowhere in the segmentation documentation. Recency filters exist, so a fixed-window win-back is native; a per-customer risk score is not. | Included on every Shopify plan (verified Sep 2026) | Fixed-window recency segments and predicted spend tiers, which is where most win-back programs currently stop |
| Retention and LTV analytics apps | Category — The nearest App Store category, and the reason this decision looks solved when it is not. These apps report repurchase rate, cohort retention and churn rate after the fact, which is measurement rather than prediction. Before assuming a listing predicts anything, search its own page for the words risk score or at-risk; the listing verified here uses neither. | Mid-tier analytics subscriptions; confirm plan pricing on the current listing (illustrative) | Understanding what already happened to your cohorts, which is a genuinely different job from flagging who is going next |
| Churn-risk model on your own order history (custom) | Build lane — The piece nobody sells: features built from each customer's own purchase cadence, category mix, discount dependence and support contacts, a fitted model, and the score written back as a customer metafield so segments and email flows can act on it. The write-back is the part that turns a model into a program. | $20,000–$60,000 one-time plus upkeep (Deploi estimate, illustrative) | Any store whose win-back fires on the same 90-day window for a weekly buyer and an annual one |
The Build Path
- Define churn before modeling anything: A store without subscriptions has no cancel event, so churn has to be defined. The usable definition is per-customer: the gap since a customer's last order measured against their own median gap, not one 90-day window applied to a weekly buyer and an annual one alike. Write this definition down and test it against last year's data before anyone opens a modeling notebook. It costs nothing and it settles most of the argument.
- The cadence baseline, which is most of the value: Score every repeat customer on their own cadence, recency and order count, with no machine learning at all. This baseline routinely beats a fixed window by a wide margin and takes days rather than months. Ship it, put it into flows, and measure the lift. Any model you build afterward has to beat this number, and having it stops a data-science project from being judged against nothing.
- The model, and the write-back that makes it real: Add category mix, discount dependence, support contacts, return history and subscription state as features, fit a model, and then write the score and its top reason back to a customer metafield. Segments and email flows read metafields directly, so activation needs no extra tooling. Retrain on a schedule and watch for drift, because a customer mix that shifts makes yesterday's model quietly wrong.
- Effort band
- $20,000–$60,000 for the definition, cadence baseline, model and metafield write-back — Deploi estimate (illustrative); the cadence baseline alone lands in the $10–25K contact-form band, the full model in $25–75K
- Typical timeline
- 2–3 weeks to the cadence baseline in production, 6–12 weeks to a retrained model with reasons attached (Deploi estimate, illustrative)
- Maintenance, honestly
- ~15–20% of build cost per year (Deploi estimate): roughly $3,000–$12,000/yr (Deploi estimate, illustrative), mostly scheduled retraining, drift monitoring and adding features as new channels and product lines appear.
- What you own — and what you take on
- You own: the churn definition, the feature set, the score and its reasons on your own customer records, and every downstream use of those features. You take on: retraining, drift monitoring, and the discipline of measuring the model against the cadence baseline rather than against nothing.
3-Year Total Cost of Capability
| Buy (app path) | Build (custom path) | |
|---|---|---|
| Year 0 (setup) | $0 (7-day trial, then subscription) | $20,000–$60,000 |
| Years 1–3 (recurring) | $17,964–$32,364 (analytics subscription) | $9,000–$36,000 (retraining and upkeep) |
| 3-year total | ≈$17,964–$32,364 | ≈$29,000–$96,000 |
- † All figures illustrative samples for the reference scenario — not quotes, not verified pricing.
- † Buy column uses Peel's Essentials tier as a realistic mid-market retention-reporting subscription, which measures churn rather than predicting it.
- † Build column covers the churn definition, cadence baseline, fitted model and metafield write-back, plus annual retraining; three-year horizon.
What the Sticker Price Hides
On the buy path
- — Retention dashboards report churn that already happened, which is a measurement, not an early warning
- — Analytics plans are capped by orders per month, so a good year moves you up a tier
- — The free tier is tied to being a Smartrr user, which is a second vendor relationship rather than a discount
- — A dashboard with no per-customer output cannot feed an email flow, so activation stays manual
On the build path
- — A score with no save action behind it is an expensive dashboard, and designing the intervention is the harder half
- — Customer mix drifts, so a model trained on last year's buyers gets quietly wrong without scheduled retraining
- — Discount dependence is a powerful feature and a dangerous one: a model can learn to flag people who only ever buy on promotion
- — ~$3,000–$12,000/yr in retraining and monitoring (Deploi estimate, illustrative)
What Merchants Say
The pattern retention teams keep describing: the win-back email fires on day 90 for everyone, which is far too late for a monthly buyer and meaningless for an annual one.
A recurring complaint about retention dashboards: cohort curves look great in a board deck and nobody can turn them into a list of customers to contact this week.
If You Change Your Mind Later
If you bought and outgrow it
Retention analytics computes cohorts inside the vendor's warehouse, so leaving means losing the computed history rather than the raw orders. Check what exports at your plan tier before you sign, and keep your own copy of the order and customer extract feeding it.
If you built and want out
Nothing strands. Scores and reasons live as metafields on your own customer records, the feature pipeline is code in your repository, and both keep working if you change email platforms, analytics vendors or storefronts entirely.
When This Answer Changes
We're watching for:
- ▸ Shopify adding a churn-risk or at-risk predictive attribute to customer segmentation, alongside today's predicted spend tier
- ▸ Any retention app stating a per-customer risk score on its listing rather than cohort reporting
- ▸ Your own cadence baseline beating the model on lift for two quarters, which means the model is not earning its retraining cost
Verdict change log:
No changes since first publication (September 2026).
Common Questions
Is there a Shopify app that scores churn risk per customer?
No app verified for this decision states a per-customer churn-risk score. Peel: Retention Analytics is real and strongly reviewed at 5.0 stars across 35 reviews, and its listing describes lifetime value, retention and churn-rate reporting rather than prediction (verified Sep 2026). Shopify's native segmentation offers a predicted spend tier and nothing churn-specific.
What does retention analytics cost on Shopify?
Peel charges $499/month on Essentials and $899/month on Accelerate, or $5,389.20 and $9,709.20 paid yearly at a 10% discount, with a 7-day trial on both. Smartrr users get a free tier covering up to 16,000 orders a month across 3 stores (verified Sep 2026). Shopify's own customer segmentation is included on every plan.
How do you define churn for a store with no subscriptions?
Churn without a cancel event gets defined per customer: the gap since their last order measured against their own median gap, rather than one 90-day window for everyone. That definition is the first half of the work and costs nothing to test against last year's data. The fitted model and its write-back run $20,000 to $60,000 (Deploi estimate, illustrative).
Your Next Steps
If you're going with BUILD(matches your selected profile)
- Write the churn definition down first: each customer's gap against their own median, not a single 90-day window
- Ship the cadence baseline into your email platform within a month and measure lift against the current fixed-window flow
- Design the save action before the model, since a flagged customer with no offer behind them is a wasted signal
- Write the score and its top reason to a customer metafield, so segments and flows read it without extra tooling
- Set a retraining schedule and hold every model version to beating the cadence baseline on real lift
If you're going with CUSTOMIZE
- Run the 7-day trial on a retention analytics plan and confirm what exports at your tier before committing
- Search any candidate listing for the words risk score or at-risk before assuming prediction is included
- Keep your own order and customer extract running alongside the subscription, since computed cohorts do not travel
- Layer the per-customer score on top of the dashboards rather than waiting for a vendor to ship one
Official Docs & Sources
- Customer segmentation examples — Shopify Help Center
- About web pixels — shopify.dev
- Analytics — Shopify Help Center
Official documentation linked for verification — our verdicts and estimates are our own.
Related Decisions
Build or Buy Event Tracking-Plan Governance on Shopify?
No Shopify app owns your event taxonomy: the connectors ship a schema of their own, and the tools that enforce one live off the App Store.
Should You Build or Buy GA4 Correctness on Shopify?
GA4 correctness on Shopify is a customize call: audit once, rebuild the tagging, own the layer.
Should You Build or Buy Server-Side Tracking on Shopify?
Server-side tracking on Shopify splits by data ambition: buy for maintained speed, build to own the event stream.
Should You Build or Buy an Attribution Platform on Shopify?
Attribution platforms are a buy for multi-channel Shopify stores: every model is an opinion, so buy with eyes open and triangulate.
Should You Build or Buy Checkout Tracking & Pixels on Shopify?
Customize wins for checkout tracking on Shopify: an Elevar-class app for destinations plus an owned audit and server-side glue layer.
Ready to flag customers before they go quiet?
Start with the definition and the cadence baseline: churn measured against each customer's own rhythm, scored nightly, written back as a metafield your Klaviyo segments can read. That version ships in weeks and gives any later model something real to beat.
Contact us todayVerdict scored for the reference scenario above. Estimates are not quotes; app pricing carries its verification date and gets re-verified quarterly. Full scoring anchors: see the TCC methodology.
Read how we score these decisions (the TCC Framework). No affiliate links, no paid placement — no app vendor pays to appear here.