16 Use Cases. One Predictive Database.
The shop that learns — built on Aito's _predict, _relate, _recommend, _estimate, _evaluate, _search
Smart search, recommendations, co-purchase intelligence, demand forecast, inventory, markdown, churn, win-back, catalog enrichment, evaluation. Every view in this open-source reference is built on Aito's predictive operators. PetNord — a Nordic pet store with 700 SKUs, 3,000 customers, 12,000 orders, 6,000 reviews — drives the demo.
16 E-commerce Features Grouped by Function
Consumer-facing assistance, retention intelligence, merchandiser operations, analysis and enrichment. Pick a feature to read its implementation guide on GitHub.
Discover & Assist
Consumer-facing intelligence — search re-ranks per persona, recommendations learn from every click, cross-sell surfaces with calibrated lift.
Dashboard
KPI grid + top patterns (_relate × 6) + segment cards + recent orders — one Aito DB
See in demo →Source →
Smart Search
Side-by-side standard _search vs predictive _recommend — right column flips per persona pill click
See in demo →Source →
For You
Personalised tile grid re-ranks in <300 ms on persona switch — same _recommend body, different goal
See in demo →Source →
Bought Together
Anchor + 4 cross-sell tiles with live lift — dog dry-food → dental treats at ~2.7× baseline
See in demo →Source →
Cart Completion
Top add-on per cart scenario via _relate — confidence + expected uplift in €
See in demo →Source →Understand & Retain
Retention intelligence — review triage, churn risk on a customer-month panel, win-back with empirical revenue impact.
Feedback
Multi-field _predict over review text — category, sentiment, assigned-to agent, churn risk — one round-trip
See in demo →Source →
Churn
100 active customers scored by P(churn in 3 mo) with drivers via _relate × 5 and held-out _evaluate
See in demo →Source →
Win-back
€1,354 recoverable from 20 churned customers at €30 send cost — 45× ROI via _recommend + _estimate
See in demo →Source →Operate the Catalog
The merchandiser's workbench — demand forecast, reorder queue ranked by revenue at risk, markdown that picks the discount that clears, price intelligence with fair bands.
Demand Forecast
Per-SKU next-month units via 25 parallel _predict over monthly_sales — seasonality drivers via 4 _relate
See in demo →Source →
Inventory Intelligence
Reorder queue ranked by revenue at risk in € — per-row $why on the demand forecast — overstock with tied capital
See in demo →Source →
Price Intelligence
Fair-band per SKU (mean ± 1.5σ, outliers) + sweet-spot _relate over discount bands ↔ category
See in demo →Source →
Markdown
For each overstock SKU, _estimate units_sold at 5 markdown levels — picks the one that clears in 3 mo at highest margin
See in demo →Source →Analyze & Enrich
The numbers behind the predictions, the full lift band, catalog enrichment as a query, and the held-out evaluation that lets honest failures fail loudly.
Purchase Analytics
Monthly orders + revenue (24 mo), top-10 SKUs, per-segment KPIs and category mix
See in demo →Source →
Pattern Explorer
Full lift band — positive, neutral, protective (anti-recommendation). Same _relate body as Bought Together, no filter
See in demo →Source →
Product Filling
5 _predict in parallel — pet_type, category, weight_kg, dietary, tax_class from name + brand, ~480 ms
See in demo →Source →
Evaluation
Four _evaluate, three pass — one honest failure (Return Risk +0.0 pp gain). Honest pass/fail by design
See in demo →Source →Implementation Details
Explore the source, the demo moments, and the architecture decisions
{
"from": "orders",
"where": { "line_categories": { "$match": "dog_dryfood" } },
"relate": "line_categories",
"limit": 5
}
Why This Pattern Works for E-commerce SaaS
One predictive database, sixteen views, every override learned from instantly
SQL-Like Predictions
- _predict, _relate, _recommend, _estimate, _evaluate, _search
- No feature engineering required
- Confidence + $why on every prediction
Consumer + Operational, One Substrate
- Search, recs, cross-sell, cart completion
- Demand, inventory, markdown, price
- Churn, win-back, feedback triage
Production-Ready Architecture
- Next.js 16 App Router + FastAPI
- 2-layer cache, schema link traversal
- PUBLIC_DEMO mode, two-tier rate limit
Address
Episto Oy
Putouskuja 6 a 2
01600 Vantaa
Finland
VAT ID FI34337429
