Shopify AI — shipped, not demo'd.

It depends on the job. OpenAI for generative content and structured output (most product description tasks). Anthropic Claude for long-context tasks like catalogue-wide reasoning and complex customer-service triage. Google Gemini for multimodal (image + text) and where pricing matters. We benchmark on your actual data before committing.

Token-cost projection is part of every quote. In production we cap per-request token budgets, cache aggressively, and use prompt distillation (smaller models for simpler tasks). Most clients run AI at under £500/month total spend; high-volume clients with conversational commerce run £2,000–£8,000/month with clear ROI tracking.

For stores under ~500 SKUs with simple content needs — Shopify Magic is usually enough. Our Magic Config tier (£999) is a one-week setup that gets you brand-aligned product descriptions and email subject lines. Above 1,000 SKUs, custom integrations earn their cost via measurable PDP conversion lift, AI search lift, or service-cost reduction.

PII is redacted before any external LLM call by default. Where personal context is needed (CS triage, conversational commerce), we use the provider's zero-data-retention enterprise tier or run inference locally on your infrastructure. Every AI integration ships with a DPA and a data-flow diagram.

Yes — that is the Custom AI App tier (£14,999+). We handle Polaris admin UI, embedded chat, OpenAI or Claude backend, Shopify-hosted billing, App Store review compliance, and submission. Typical build is 6–10 weeks plus 2–6 weeks of Shopify review.

Sidekick is built into Shopify admin and works out of the box for most stores. Explicit configuration earns its keep on Plus B2B stores with custom catalogues, multi-store organisations, and stores with complex Functions — we tune Sidekick prompts and document workflows for your team during a 1-week sprint.

Yes. Three approaches: Algolia AI (drop-in, fastest, expensive — best for £500k+/month stores), Klevu AI (mid-tier pricing, strong for apparel and home), or custom OpenAI embeddings + pgvector (most flexible, lowest run-rate cost, requires our middleware). Most stores see 15–30% lift in search-to-conversion.

Yes — conversational commerce with AI triage on Shopify Inbox handles 40–60% of low-value tickets without human handoff (order status, shipping ETAs, returns, sizing). Trained on your help docs and order data, with human escalation for anything sensitive. We do not deploy AI chat on regulated categories (supplements, wellness with claims).

Yes — typically using LLM-driven recommendations layered on top of rules-based recs (LimeSpot, Rebuy, Nosto). LLM recommendations handle cold-start (new products with no purchase history), semantic matching ("looks like this but in green"), and explanation copy. Average uplift on click-through is 18%.

Yes. Model Context Protocol (MCP) is the open standard most major LLM platforms now consume. We build MCP servers that expose your Shopify catalogue, inventory, pricing, and policies (returns, shipping, sizing) so external agents — ChatGPT, Claude, Perplexity, Shop's own agentic shopping surface — can answer questions about your products and place orders on a customer's behalf. Built on the Custom AI App tier. Pairs with the LLM-discoverability work in the AI SEO service.

Three layers. (1) llms.txt at the root listing your canonical product, policy, and content URLs in machine-readable form. (2) Structured product data — Schema.org Product, Offer, and Brand markup on every PDP, kept in lockstep with your Shopify catalogue. (3) An MCP server (or equivalent agent-callable API) for stores that want to be transactable, not just discoverable, by agents. We deliver layers 1 and 2 in the AI SEO service; layer 3 sits inside the AI Integration Custom App tier.

Yes. Beyond the default Sidekick configuration we tune for Plus B2B and multi-store stores, we build custom Sidekick skills that wrap your store-specific data and operations — "show me orders flagged for fraud review", "draft a re-stock notification for returning customers of SKU X", "generate German PDP copy from the English variant". Skills are grounded in your Shopify data via Admin API, and can trigger Functions or Flow from a natural-language prompt.

For consumer-brand DMs, increasingly yes. The Shop app surfaces brands to existing Shop Pay shoppers, and Shop Minis lets you build small in-app brand experiences (clienteling, AR try-on, personalised stylist). Shop's agentic shopping surface (early 2026) will pull from the same data layer as MCP-discoverable stores. We scope Shop App / Minis presence as a small add-on inside the Custom AI App tier when it fits the brand.