If you run a Shopify store and you ever clicked into Shopify Analytics, you have seen the same dashboards everyone else has. Total revenue. Conversion rate. Average order value. Maybe a top-products list. Useful information. None of it tells you what to do tomorrow morning.
The hard problem in commerce in 2026 is not measurement. It is action. With customer acquisition costs at all-time highs and acquisition channels (Meta, Google, TikTok) compressing margins year-over-year, every brand we work with has come to the same conclusion: the next dollar of revenue comes from a customer you already have. Not a new one.
That conclusion is uncontroversial. The hard part is operationalising it. Who, exactly, should you email this week? Which customer is most likely to churn in the next 30 days? Which one is silently about to become your top spender? Shopify Analytics will not tell you. Klaviyo will not tell you (it sends, it does not decide). Spreadsheets full of order exports get you 40% of the way there if you have a data analyst on staff.
That gap is exactly what Miko AI exists to close.
What Miko AI does, in one sentence
Miko AI is a native Shopify app that reads your customer purchase history, segments customers using machine learning, predicts churn and lifetime value, and surfaces specific actions for you to take to grow revenue.
That is the whole product in a sentence. The interesting part is how it does it, and why we believe it works better than the alternatives.
The five pieces, in order
1. Automatic RFM segmentation
RFM – Recency, Frequency, Monetary – is the oldest, best-understood segmentation framework in retail. It works. It has been a quiet workhorse of CRM strategy for thirty years. Miko AI computes RFM scores for every customer the moment you install, then keeps them fresh nightly. Every customer lands in one of the eleven canonical segments: Champions, Loyal Customers, Potential Loyalists, Recent Customers, Promising, Customers Needing Attention, About to Sleep, At Risk, Cannot Lose Them, Hibernating, and Lost.
You have probably read about these on a marketing blog. The difference is that until now, computing them required a data team. Miko does it automatically.
2. Churn prediction
A churn-probability score (0–100) per customer. Logistic-regression model trained on common churn-signal features: recency relative to the customer's own historical cadence, decreasing order size, decreasing frequency, single-product dependency, and a handful of others. We update scores nightly so the moment a Champion starts looking like At Risk, you see it.
The point of the score is not to be perfect (no churn model is). It is to rank. If you have 5,000 customers and you can only realistically reach out to 100 this week, you want the right 100. Churn ranking turns that from guesswork into a list.
3. Customer Lifetime Value (CLV) forecasting
Predicted 12-month spend for every customer based on their order pattern, recency, product mix, and similarity to other customers. A new customer with two orders, both high-margin, low-discount, premium SKUs is treated very differently from a new customer with two orders, both deeply discounted, low-margin loss-leaders – even though they look identical on a basic dashboard.
CLV forecasts let you do things like find customers in the top 10% predicted spend and route them to a VIP onboarding sequence, automatically.
4. Behavioural clusters (the unsupervised piece)
RFM is rule-based. Clustering is the opposite: we let the data tell us what segments exist. Miko AI runs K-Means clustering on your order history to surface natural customer groupings that no rule would have caught. Examples we have already seen in test data:
- "Bulk reorder accounts": customers who order the same SKU at 3-4x typical volume on a quarterly cadence. Often hidden B2B accounts buying through your retail storefront. Tag them and route to a wholesale conversation.
- "Discount hunters": customers whose entire order history sits behind promo codes. Different lifecycle from full-price customers; do not waste a re-engagement campaign on them.
- "Single-product loyalists": customers who reorder the same one item for years and never explore the catalogue. Different cross-sell strategy entirely.
These show up automatically. You did not have to think of the segment first.
5. Claude-powered explanations
Here is the part where we sound different from every other "AI" customer app on the App Store. We do not just hand you a dashboard of numbers and let you figure out what to do. We use the Anthropic Claude API to translate every segment, every cluster, every prediction into plain English – with a recommended action.
So instead of seeing:
Segment: cluster_7 (n=237, recency_mean=72d, frequency_mean=3.2, monetary_sum=$18,400)
You see:
"237 customers used to order every 30 days. They have all gone quiet in the last 60. Average past order $77.65. Recommended action: send a win-back email with a small incentive (free shipping or 10% off). Projected recovery value: ~$5,400 on a typical 20% conversion."
That is the difference between insight and action. The first is a status report. The second is an instruction.
What Miko AI does not do
One thing we want to be explicit about: Miko AI is not an email tool. It does not send messages, it does not run campaigns, it does not replace Klaviyo or Shopify Email or whatever you use.
Miko AI is the brain. Your existing email tool is the sender. Miko writes its segments back into Shopify customer tags (with a miko- prefix so we never collide with your existing tags), and Klaviyo / Shopify Email / Mailchimp pick those up automatically. There is no integration to set up.
This was a deliberate design choice. There are already excellent senders. There are very few good thinkers. We wanted to ship a thinker that plugs into the senders you already use, not a half-built sender that competes with them.
Why we built it inside the Miko ecosystem
Miko AI is the third app in a deliberate stack:
- Miko Loyalty & Rewards: turns customer behaviour into retention via points, tiers, and rewards. Live on Shopify App Store.
- Miko B2B Wholesale Hub: gives wholesale buyers a native experience with grouped pricing, NET terms, and merchant-side PO management. Private beta.
- Miko AI: identifies which customers to focus on and what to do about them. In development.
Each app is useful on its own. The interesting math is what happens when you install all three. Miko AI predicts that a wholesale buyer is at high churn risk – Miko B2B surfaces that account in your dashboard with a flag. Miko AI identifies a retail customer hitting Champion status – Miko Loyalty bumps them into the next VIP tier with a thank-you reward. One brain, three apps, compounding effect.
This is not a marketing slide. It is the actual architecture. The apps share the same customer-tag namespace and the same Tripster engineering standards. Install them together and they work together with no integration code from you.
The stack, for the technically curious
Miko AI is built on:
- Remix + Shopify Polaris for the embedded app UI. Looks and feels like part of the Shopify admin.
- PostgreSQL via Prisma for the application database.
- Python FastAPI + scikit-learn for the ML microservice. RFM scoring, K-Means clustering, churn classification, CLV regression.
- BullMQ + Redis for background jobs – ML runs do not block the merchant UI.
- Anthropic Claude API for plain-English segment explanations and action recommendations.
- Recharts for the in-app visualisations.
- Resend for the weekly merchant insight email.
Everything runs on Railway. Everything is observable. Everything degrades gracefully – if the ML service is down, Miko falls back to rule-based RFM scoring so you never see a blank dashboard.
When can I install it?
Soon. The core app, ML service, Claude integration, and Shopify auth flow are all built and working in development. We are polishing the onboarding experience, hardening the data pipeline, and writing the documentation before we open a closed beta to a small group of Shopify merchants.
If you want to be on that list, join the waitlist. We will reach out the moment closed beta opens. Beta merchants get extended trial credits and locked-in launch pricing.
In the meantime, if you are running Miko Loyalty already, you are part of the data that will make Miko AI better at launch – the model is being trained on the same kinds of stores you run. The ecosystem compounds in both directions.
The next post in this series will go deeper on the architecture: how we serve ML predictions at scale through the Python microservice, how we cache results to keep the merchant UI instant, and how the Claude integration handles segments with thousands of customers without melting our context window. Stay tuned.