B2B Ecommerce World · Indianapolis · November 2, 2026

AI Year One: what we tried, what failed, and what’s next.

ULE Group’s first year putting AI to work on product data: 81,000+ data points added to their PIM, one invented-link scare, and the model I’d use to get your top 50 SKUs ready for buyers who shop through AI.

Download the slides (PDF) ↓View them here ↓

Title slide: AI Year One, Rudy Abitbol, ULE Group case study — 21 slides · ~25 minutes · Case study shared with ULE Group’s permission

The 60-second version. Product data is never done.

Give every product page a stage. Don’t try to lift your whole catalog at once: take your top 50 sellers to Stage 3 first and let them set the standard.

Stage 0

Identifiable

A SKU, an image, a price. Where ULE started.

Stage 1

Searchable

Core attributes in the PIM. Filters work.

Stage 2

Sellable

Full description, verified specs, real downloads.

Stage 3

Agent-ready

Grounded in source documents, cited, audited every month.

Five things to do before 2027.

Test your top 20 on Monday. Lift your top 50 this quarter.

01 · Audit

Ask ChatGPT or Claude about your top 20 SKUs, the way a buyer would.

02 · Prioritize

Pick your top 50 sellers as the north star.

03 · Define

One gold-standard product page per category. Show it to execs, procurement and manufacturers.

04 · Ground

AI fills gaps only from your own files (RAG). Never let it write the links.

05 · Repeat

Re-audit and enrich every month. Measure discoverability, not just conversion.

The slides.

Data is sourced on each slide (mostly Shopify’s 2026 platform data). Everything labeled “my view” is my opinion.

Slide 1: AI Year One — 01AI Year OneWho has a product where the spec sheet knows more than the product page?

Slide 2: Who I am — 02Who I amRudy Abitbol: B2B commerce AI, PIM and product content for manufacturers and distributors.

Slide 3: Three acts, one real story — 03Three acts, one real storyOne real story in three acts: what ULE tried, what failed, what's next.

Slide 4: The original product page was almost empty — 04The original product page was almost emptyWhere ULE started: an image, a price and almost nothing else.

Slide 5: A “match” could still leave the product empty — 05A “match” could still leave the product emptyA record can be identifiable without being sellable.

Slide 6: AI-referred buyers are better buyers — 06AI-referred buyers are better buyersShopify data, Q2 2026: AI-referred buyers convert 80% better, with 14% larger orders.

Slide 7: Specific beats popular — 07Specific beats popular75% of AI-referred sales went to specialized products. In B2B, every SKU is specialized.

Slide 8: AI could fill the gaps. Safety set the limits. — 08AI could fill the gaps. Safety set the limits.Electrical products: wrong data can mean misuse or a fire. Manufacturer content only.

Slide 9: An LLM helped build a Python workflow — 09An LLM helped build a Python workflowAn AI assistant coached a small Python workflow: 81,000+ data points into the PIM.

Slide 10: No platform works out of the box — 10No platform works out of the boxAbout 700 more attributes than the platform could store. Workarounds are the job.

Slide 11: You don’t need agents. You need to start. — 11You don’t need agents. You need to start.78% of merchants built their own store. Asking AI “how do I…?” is a strategy.

Slide 12: It looked finished. The links were invented. — 12It looked finished. The links were invented.The first automation looked finished. The PDF links were invented.

Slide 13: From a sparse page to a confident purchase — 13From a sparse page to a confident purchaseBefore and after: extract, structure, write, attach real files, publish.

Slide 14: ULE’s plot twist is the market pattern — 14ULE’s plot twist is the market patternThe market pattern, and what to put in place from day one: RAG, scoped permissions, an owner.

Slide 15: Enrichment takes more than one prompt — 15Enrichment takes more than one promptSource, structure, validate. Review your data architecture first.

Slide 16: Product data is an ongoing effort — 16Product data is an ongoing effortStage 0 to 3. Take your top 50 sellers to agent-ready first.

Slide 17: Earn the agent’s trust, then the buyer’s — 17Earn the agent’s trust, then the buyer’s2027: win the recommendation and the reorder before agentic checkout.

Slide 18: Five things to do before 2027 — 18Five things to do before 2027Audit, prioritize, define, ground, repeat.

Slide 19: Ask AI about your top 20 SKUs — 19Ask AI about your top 20 SKUsAsk AI about your top 20 SKUs. What it gets wrong is your 2027 roadmap.

Slide 20: Is AI telling your buyers about you? — 20Is AI telling your buyers about you?Free GEO audit: does AI name you when buyers ask?

Slide 21: Let’s keep the conversation going — 21Let’s keep the conversation goingPIM and product content that AI can read and buyers trust.

Download the slides (PDF) ↓Connect on LinkedIn ↗

Thank you to Denise Foley, EVP Ecommerce, and the ULE Group team for letting me share their story.

Run the Monday test in 90 seconds.

The free GEO audit asks Claude, ChatGPT and Gemini the questions your buyers ask, and shows whether they name you.

Run the free GEO audit →

Want your top 50 at Stage 3?

This is the work I do every day: Akeneo and Shopify Plus, supplier files turned into structured product data, and product content that AI can read and buyers trust.

Talk to Rudy →B2B Commerce AI