Indexwright Commerce Agent Lab
Field Notes
Findings from the lab: how AI shopping agents are built, tested and priced, and what the search underneath them really costs. Evidence first, no hype.
How to Test an AI Shopping Assistant Before Launch
A 30-question starter test set, hard-fail vs scored checks, go/no-go thresholds, and a worked failed run: how to find out what your assistant does on questions you did not pick.
Read the field noteIs Your Product Data Ready for AI Shopping Agents? A 12-Point Audit
Agents cannot read your marketing copy. They read your data, and gaps in it quietly remove products from consideration.
Read the field noteACP, UCP, AP2 and MCP: Agentic Commerce Protocols Explained
Four acronyms, four different jobs. Once you know which job each does, the to-do list for your store gets short.
Read the field noteWhat Should an AI Agent Handle in E-commerce Returns? A Risk-Based Guide
Order status is easy to automate. A refund is a promise of money. Route each request by risk, and widen the agent’s authority only as it earns it.
Read the field noteHow Much Does an AI Shopping Agent Cost Per Conversation?
The pricing page gives you one number. Your bill has six layers, and the biggest one may be people.
Read the field noteWhy Your AI Shopping Assistant Is Only as Good as Your Search
The model writes the answer, but search chooses what it writes about. Fix retrieval first and most of the other problems shrink.
Read the field noteHow to Migrate from Algolia to Typesense Without Downtime: A Step-by-Step Guide
A seven-step migration plan — parallel setup, side-by-side testing, feature-flag cutover, and a rollback window — built so search never goes down and nothing gets lost.
Read the field note7 Ways to Cut Your E-commerce Search Costs in 2026
From debouncing keystrokes to right-sizing your plan tier: seven concrete levers to pull before assuming a full engine switch is the only option.
Read the field noteAlgolia Alternatives 2026: Typesense vs Meilisearch vs OpenSearch Compared
Typesense, Meilisearch, and OpenSearch all promise cheaper search than Algolia. Here's what actually differs between them on cost, setup effort, and relevance quality.
Read the field note9 Signs Your Algolia Setup Is Costing You Customers (Not Just Money)
A high bill is the easy signal to notice. The harder ones — throttled features, unexplained ranking rules, roadmap decisions shaped by pricing — usually show up first.
Read the field noteWhy Is Algolia So Expensive? The Real Cost Breakdown
Records, request volume, bundled features, overage rates — a breakdown of the four line items that turn a reasonable-looking Algolia plan into a 3–5x bill.
Read the field note