A worked campaign example
A 45-minute webinar plus companion article for the good-fit sponsor, titled around the audience's decision rather than the technology: 'From demo to production: what a trustworthy AI shopping assistant actually requires.'
Run of show, built from Bailly's published evidence:
1. Open with the failure case. Bailly walks through the five-run Gemini test on the Sony headset question — three different prices, two different models — and the Product.ai study behind it: 220 buying questions tested five times each, 86% of answers contradicting each other, 54% wrong on product or price. This is Bailly's own material, reproduced with the original prompt, so the opening lands as evidence rather than a vendor warning.
2. Move to the build list. Using the Claude for Commerce analysis, Bailly sets out what an open-source reference implementation still leaves a retailer to build: connections to catalog, cart, orders and commercial policies; the conversation runtime; session memory; reporting and testing; model version management; proactive engagement formats; and connectors to tools such as Zendesk, Gorgias, Klaviyo and OneStock. The point is to turn a launch announcement into a work estimate.
3. Add the precedent. The parallel to in-house search engines in the 2010s that ended up at Algolia, Constructor or Klevu, and to recommendation engines between 2015 and 2020, gives buyers a reason to attend beyond the current news cycle.
4. Close on the contested question. Bailly's OpenAI timeline — checkout dropped in March, first US ads in February, self-service Ads Manager in May, product feeds at 2 million SKUs in June, in-chat ad formats in September — set against the 72% of US users who are neutral or annoyed by ads in chats, and Google's Universal Cart still limited to a small test group four months after a summer launch promise. End on an open question to give the audience something to debate and the sponsor a natural place for Q&A.
5. Hand over a decision checklist: what to connect, what to test, what to measure, and when to build rather than buy.
The article follows the same order, led by the Sony test, with the build list as a scannable table.
Sponsor role: supply the platform, a customer example if one exists, and the promotion. Bailly presents and writes. Confirm the fee and date, that they may appear with a sponsor that does not compete with iAdvize, and request a per-post engagement report from comparable posts to set expectations against more than follower count.
