AI Biscuit — Self-Learning Shopify Product Agent

// What it really is

A SEARCH, DISCOVERY
& FILTERING ENGINE.
DRESSED AS AN AI APP.

AI Biscuit exists for one reason: to make every product on your store filterable, discoverable, and findable. Everything else — the classification, the metafield filling, the taxonomy — is in service of that single goal.

01 // Search & Discovery

FILTERS THAT
ACTUALLY WORK.

AI Biscuit fills your product metafields using the exact field keys defined in Shopify Custom Data. Those fields work immediately as filters in Shopify Search & Discovery — no theme changes, no third-party apps.

Consistent values are everything. AI Biscuit learns your exact formatting from existing products — your filters stay clean.

02 // Basic data in

YOU LIST THE
BASICS. AI DOES
THE REST.

Import a product with just a title, description, and price. Leave product type blank. Skip the metafields entirely. AI Biscuit reads each product, assigns the correct type, and fills every required field — formatted exactly as your store expects.

No more mapping supplier category codes. No more manual metafield checklists. Just the raw product — AI does the structure.

03 // Speed to ads

IN YOUR NEW
ARRIVALS ADS
SAME-DAY.

Products usually sit in a classification backlog for 2–4 days before they're live and ad-ready. With AI Biscuit, new imports can be properly tagged, filtered, and feed-ready within hours. That means your new arrivals campaigns capture revenue sooner.

Faster time-to-market = faster ad spend on fresh inventory = faster sell-through on the products with the highest margin.

// Filter values for 'Speed' — same store, before vs after

Before AI Biscuit

700c47 products

700C23 products

70012 products

700c wheel5 products

// Customer clicks one option · misses 40% of relevant products

After AI Biscuit

700c90 products

All 90 products visible under one filter value.

Customers see the full inventory · click-through and conversion rise · no products hidden behind format inconsistencies.

How it works

THREE STEPS,
ZERO MANUAL
EFFORT.

01 // Setup

BUILD YOUR
TAXONOMY.

AI Biscuit scans your existing products and learns your product types, metafield structure, and value formats. It builds a knowledge base from your real store data — not generic assumptions. The more products you have, the smarter it gets.

02 // Classify

CLASSIFY NEW
ARRIVALS.

Import products from any supplier with just a title, description, and price. Leave the Product Type blank. AI Biscuit reads each product and assigns the correct type, then fills every required metafield — formatted exactly as your store expects.

03 // Apply

REVIEW &
APPLY.

Results split into Ready and Needs Review. Preview exactly what will change before touching Shopify. One click applies everything. The taxonomy self-updates after every run, improving accuracy over time.

// Cost savings analysis

PAYS FOR ITSELF
IN THE FIRST WEEK.

Most Shopify merchants underestimate how much time they spend classifying products and filling metafields. Across every supplier drop and seasonal restock, those minutes add up to days — and the cost is real.

£3,900+

Annual saving
typical store

95%

Less time on
data entry

<1 wk

Payback period
typical store

100%

Consistent
metafield values

// Time cost by store size

Products / month	Avg. time each	Hours / month	Days / year	Cost @ £15/hr
50 products	4 min	3.3 hrs	5 days	£594/yr
100 products	4 min	6.7 hrs	10 days	£1,188/yr
200 products	4 min	13.3 hrs	20 days	£2,376/yr
500 products	4 min	33.3 hrs	50 days	£5,940/yr
1,000 products	4 min	66.7 hrs	100 days	£11,880/yr

// Based on UK warehouse/e-commerce admin rate of £15/hr · Time estimates from merchant interviews

// Growth plan ROI — 200 products/month

£4,172
net annual return

Staff time saved, catalogue cleanup, improved filter accuracy with a conservative 0.5% conversion lift on £300k revenue, and reduced error correction.

Industry data suggests filter usability improvements of this type typically drive 0.3–1.2% conversion lift.

Per year breakdown

Staff time saved+ £2,400

Catalogue cleanup (one-off)+ £500

Filter conversion lift (0.5%)+ £1,500

Reduced error correction+ £360

AI Biscuit Growth Plan− £588

Net return£4,172

// Payback period by plan

Starter · $19/mo

~1 day

Saves 13 hrs/month

Growth · $49/mo

< 1 day

Saves 67 hrs/month

Scale · $99/mo

< 4 hrs

Saves 100+ hrs/month

// Worked example

CYCLING RETAILER —
BEFORE & AFTER.

Store profile: South East England · 4,200 SKUs · 3 major suppliers · 300 new products/month. Two part-time staff spent a combined 25 hours per month classifying imports and filling metafields. Inconsistency had built up over 3 years — 17 different variants of '700c' existed in the database.

Before AI Biscuit

×25 hrs/month spent on classification

×17 variants of '700c' in the database

×3–5 day publishing delay on new imports

×Filters showing fragmented values · customer complaints

££375/month in staff time on classification

After AI Biscuit

→~90 minutes/month on review and apply

→1 consistent value across all 4,200 products

→Same-day publishing after supplier delivery

→Clean filters · measurable improvement in time-on-site

££49/month plan · net saving £326/month

// What's inside

EVERY TOOL YOU NEED TO
OWN YOUR CATALOGUE.

Nine tightly integrated tools, all running on the same self-learning taxonomy. Each one solves a specific problem — together they replace hours of manual data work every week.

01 // Core engine

AI Classification

Claude AI reads each product's title, description, vendor and SKU, then assigns the correct product type with a confidence score. Haiku model by default for speed and low cost, Sonnet on retry for difficult or ambiguous products. Pause and resume runs any time.

02 // Intelligence

Self-Learning Taxonomy

Builds automatically from your existing products. Learns your exact product types, metafield keys, and value formats. Rebuilds itself after every apply run — the longer you use it, the more accurately it represents your specific catalogue. Day-one accuracy ~90%, typically reaches 95%+ after a month.

03 // Catalogue health

Fix Gaps

Scans every product that has a type set but is missing required metafields. Products imported before metafields existed, or products that need new fields added — Fix Gaps backfills the entire catalogue in one run. No product-by-product manual work.

04 // Pre-flight

Data Quality Scanner

Pre-classification scan flags products with missing descriptions, short titles, SKU-style names, or missing vendors. Rates each as Ready, Will Try, or Cannot Classify. Exports a CSV of issues so you can fix them in Excel before wasting API credits on products that will fail.

05 // Catalogue audit

Duplicate Type Detector

Finds inconsistent product types fragmenting your catalogue — "Road Tyre" vs "Road Tyres" vs "road tyre" — and suggests which one to keep. Detects capitalisation differences, singular/plural conflicts, and spacing variants. Critical for clean Search & Discovery filters.

06 // Offline workflow

Classify from CSV

Upload a Matrixify export directly. AI Biscuit classifies every product and downloads an enriched CSV with types and metafields filled in. Nothing touches Shopify until you import it — perfect for QA, supplier file processing, or staging large catalogue uploads.

07 // Safety net

Undo Apply

Before writing to Shopify, AI Biscuit captures a full snapshot of every product's original state. If anything looks wrong after applying, one click restores everything — type, metafields, all of it. The last 3 apply snapshots are kept per store.

08 // Targeting

Smart Product Filtering

Choose exactly which products AI Biscuit classifies — Blank Type (default), All Products, or a custom label like "NEW PRODUCT" or "IMPORT". Optionally include blank-type products alongside. Tag new arrivals from your supplier feed and let AI Biscuit find them automatically.

09 // Audit trail

Import Log & Audit Trail

Every apply run produces a downloadable timestamped log: which products were classified, the assigned type, every metafield set, the AI's confidence and reasoning. Useful for QA, supplier audits, and compliance — full visibility of exactly what changed and when.

The problem

EVERY NEW PRODUCT
NEEDS SOMEONE TO
FILL IT IN.

Brand. Colour. Speed. Size. Material. Wheel size. For every product, every time. On a store importing 50 products a week that's 2–4 hours of repetitive data entry. And when it's done by hand, it's inconsistent.

One person writes "700c". Another writes "700C". A third writes "700". Now your Search and Discovery filters show three versions of the same thing. Customers can't filter properly. Sales suffer.

Manual entry doesn't scale

As your catalogue grows, so does the backlog. New supplier imports sit unclassified for days because there aren't enough hours.

Inconsistent data breaks filters

Different people use different formats. Search and Discovery shows duplicate values. Customers see "700c", "700C", and "700" as three separate sizes.

Existing catalogues have gaps

Products imported before metafields existed. New fields added after thousands were live. Gaps that would take months to backfill manually.

// Beyond cost savings

FIVE BENEFITS THAT
DON'T SHOW UP ON THE
P&L.

01 // Speed to publish + ad

New products go live within hours of import — not days. That means they show up in your New Arrivals ad campaigns, Performance Max feeds and email send-outs same-day. Faster time-to-market becomes faster time-to-revenue on every fresh line.

02 // Staff focus

Removing repetitive data entry frees staff for higher-value tasks — merchandising, customer service, buying. Reducing mundane work also improves job satisfaction and reduces turnover risk.

03 // Catalogue intelligence

The Duplicate Type Detector identifies inconsistent product types across your catalogue — Road Tyre vs Road Tyres vs road tyre — giving you a clear picture of data quality issues before they affect customers.

04 // Supplier independence

Any supplier file can be classified regardless of how it's structured. No more mapping supplier category codes to your taxonomy. Import the bare minimum, let AI Biscuit do the rest.

05 // Self-improving accuracy

Every classification run teaches the system more about your catalogue. Accuracy improves over time. What hits 90% accuracy on day one typically reaches 95%+ after a month of regular use.

06 // Audit trail

Every apply run produces a downloadable import log with timestamps, product names, assigned types, and metafields set. Useful for QA, compliance, and understanding exactly what changed and when.

Who it's for

MADE FOR HIGH-VOLUME
SHOPIFY STORES.

→

Regular Supplier Imports

Importing 50–500 new products every week. Manual classification is a bottleneck — AI Biscuit makes it one click.

→

Metafield-Driven Filtering

Using Shopify Search and Discovery or a filter app — and need clean, consistent metafield data to power them.

→

Catalogue Cleanup Projects

Thousands of products with incomplete data. Fix Gaps backfills the entire catalogue in one run — no manual work.

→

Sports, Outdoor, Hardware

High-SKU categories with consistent attributes — speed, size, colour, material. AI Biscuit was built and tested in exactly these environments.

→

Fashion & Apparel

Colour, size, material, gender, fit — filled automatically from titles and descriptions. Size guides and filter panels stay clean.

→

Multi-Brand Retailers

Stocking hundreds of brands with different naming conventions. AI Biscuit normalises everything to your taxonomy.

// FAQ

COMMON QUESTIONS.

Do I need an Anthropic API key? +

Yes — AI Biscuit uses your own Anthropic API key so you have full control over your AI spend and rate limits. Get a key from console.anthropic.com (5-minute signup), paste it into the Settings tab, and you're set. Most stores spend $5–$20/month on API usage on top of the AI Biscuit subscription — far less than a single hour of manual classification work.

How accurate is the classification? +

On products with good titles and descriptions, accuracy is consistently above 90% from day one — typically reaching 95%+ after a month of use as the self-learning taxonomy refines itself on your real store data. Products with very short or vague descriptions go to the Review list rather than being auto-applied, so you stay in control. The Data Quality Scanner identifies these before you run a classification, so credits aren't wasted on products that will fail.

Will it work with my existing metafield structure? +

Yes. AI Biscuit builds its taxonomy from your existing products and learns your exact metafield namespaces, keys, and value formats. It doesn't change your structure — it fills in what you already have defined. If a metafield appears on 3 or more products of a given type, AI Biscuit treats it as part of that type's taxonomy and ensures every new product of that type has it filled.

What if it classifies something wrong? +

Three safety layers: (1) Before applying, you can preview every change in a per-product modal — type, confidence, metafields, and the AI's reasoning. (2) The full results table splits into Ready and Needs Review so you can choose what gets applied. (3) After applying, the Undo button restores the last 3 apply runs to original state — type and all metafields. Mistakes are reversible at every stage.

Does it work with any product category? +

Any category where attributes can be determined from the title and description. AI Biscuit works best in high-SKU categories with consistent attributes — sports and outdoors, cycling, hardware, automotive, fashion, electronics, beauty, tools. Lower performance on niche or one-off products where the catalogue isn't structured enough to build a meaningful taxonomy from. The 14-day free trial lets you test it on your specific catalogue before committing.

How does billing work? +

Billing is handled entirely through Shopify — the AI Biscuit charge appears on your regular Shopify invoice alongside any other apps. The 14-day free trial classifies up to 100 products with no credit card required. Cancel any time directly from your Shopify admin — no separate billing portal, no account to delete elsewhere. Your Anthropic API usage is billed separately by Anthropic to your own account.

Can I classify products without importing them to Shopify? +

Yes — the Classify from CSV feature lets you upload a Matrixify export, classify every product against your taxonomy, and download the enriched file with types and metafields filled in. Nothing touches Shopify until you choose to import it. Use it for QA on staging files, processing supplier CSVs before they go live, or backfilling a catalogue export from another platform.

How does the self-learning taxonomy work? +

When you first build the taxonomy, AI Biscuit scans every product in your store, groups by product type, and identifies metafields that appear on 3 or more products of each type. It learns your value formats (e.g. "700c" not "700C"), detects fields with many unique values as freeform, and stores everything as your store's taxonomy. After every apply run, it rebuilds — picking up new types, new metafields, and refined value formats. The system gets smarter the more you use it.

Is my store data secure? +

Yes. AI Biscuit only requests read_products and write_products Shopify scopes — nothing more. No customer data is ever read or stored. Your taxonomy and settings are stored encrypted, scoped to your store. GDPR-compliant: data deletion is handled via Shopify's shop/redact webhook on uninstall. Product data sent to Claude AI for classification is processed via the Anthropic API — no training, no retention beyond the immediate request.

// Still got a question?

We answer support emails personally — usually within one working day. Built and supported by the same people who use it on real client stores.

Email us

AI BISCUIT.
SELF-LEARNING
PRODUCT AGENT.

A SEARCH, DISCOVERY
& FILTERING ENGINE.
DRESSED AS AN AI APP.

FILTERS THAT
ACTUALLY WORK.

YOU LIST THE
BASICS. AI DOES
THE REST.

IN YOUR NEW
ARRIVALS ADS
SAME-DAY.

THREE STEPS,
ZERO MANUAL
EFFORT.

BUILD YOUR
TAXONOMY.

CLASSIFY NEW
ARRIVALS.

REVIEW &
APPLY.

FROM HOURS
TO MINUTES.

PAYS FOR ITSELF
IN THE FIRST WEEK.

£4,172
net annual return

CYCLING RETAILER —
BEFORE & AFTER.

EVERY TOOL YOU NEED TO
OWN YOUR CATALOGUE.

AI Classification

Self-Learning Taxonomy

Fix Gaps

Data Quality Scanner

Duplicate Type Detector

Classify from CSV

Undo Apply

Smart Product Filtering

Import Log & Audit Trail

EVERY NEW PRODUCT
NEEDS SOMEONE TO
FILL IT IN.

FIVE BENEFITS THAT
DON'T SHOW UP ON THE
P&L.

MADE FOR HIGH-VOLUME
SHOPIFY STORES.

Regular Supplier Imports

Metafield-Driven Filtering

Catalogue Cleanup Projects

Sports, Outdoor, Hardware

Fashion & Apparel

Multi-Brand Retailers

SIMPLE, TRANSPARENT.
BILLED THROUGH SHOPIFY.

COMMON QUESTIONS.

WE BUILT IT FOR
OURSELVES FIRST.

STOP FILLING IN
METAFIELDS BY HAND.

AI BISCUIT.SELF-LEARNINGPRODUCT AGENT.

A SEARCH, DISCOVERY& FILTERING ENGINE.DRESSED AS AN AI APP.

FILTERS THATACTUALLY WORK.

YOU LIST THEBASICS. AI DOESTHE REST.

IN YOUR NEWARRIVALS ADSSAME-DAY.

THREE STEPS,ZERO MANUALEFFORT.

BUILD YOURTAXONOMY.

CLASSIFY NEWARRIVALS.

REVIEW &APPLY.

FROM HOURSTO MINUTES.

PAYS FOR ITSELFIN THE FIRST WEEK.

£4,172net annual return

CYCLING RETAILER —BEFORE & AFTER.

EVERY TOOL YOU NEED TOOWN YOUR CATALOGUE.

AI Classification

Self-Learning Taxonomy

Fix Gaps

Data Quality Scanner

Duplicate Type Detector

Classify from CSV

Undo Apply

Smart Product Filtering

Import Log & Audit Trail

EVERY NEW PRODUCTNEEDS SOMEONE TOFILL IT IN.

FIVE BENEFITS THATDON'T SHOW UP ON THEP&L.

MADE FOR HIGH-VOLUMESHOPIFY STORES.

Regular Supplier Imports

Metafield-Driven Filtering

Catalogue Cleanup Projects

Sports, Outdoor, Hardware

Fashion & Apparel

Multi-Brand Retailers

SIMPLE, TRANSPARENT.BILLED THROUGH SHOPIFY.

COMMON QUESTIONS.

WE BUILT IT FOROURSELVES FIRST.

STOP FILLING INMETAFIELDS BY HAND.

AI BISCUIT.
SELF-LEARNING
PRODUCT AGENT.

A SEARCH, DISCOVERY
& FILTERING ENGINE.
DRESSED AS AN AI APP.

FILTERS THAT
ACTUALLY WORK.

YOU LIST THE
BASICS. AI DOES
THE REST.

IN YOUR NEW
ARRIVALS ADS
SAME-DAY.

THREE STEPS,
ZERO MANUAL
EFFORT.

BUILD YOUR
TAXONOMY.

CLASSIFY NEW
ARRIVALS.

REVIEW &
APPLY.

FROM HOURS
TO MINUTES.

PAYS FOR ITSELF
IN THE FIRST WEEK.

£4,172
net annual return

CYCLING RETAILER —
BEFORE & AFTER.

EVERY TOOL YOU NEED TO
OWN YOUR CATALOGUE.

EVERY NEW PRODUCT
NEEDS SOMEONE TO
FILL IT IN.

FIVE BENEFITS THAT
DON'T SHOW UP ON THE
P&L.

MADE FOR HIGH-VOLUME
SHOPIFY STORES.

SIMPLE, TRANSPARENT.
BILLED THROUGH SHOPIFY.

WE BUILT IT FOR
OURSELVES FIRST.

STOP FILLING IN
METAFIELDS BY HAND.