AI Product Recommendation Agents: Turn Product Discovery Into Guided Selling (2026)

Monday morning on the floor: a shopper types “waterproof trail shoe under $140, wide, in stock, works with my gaiters.” Your related-products rail answers with something that was frequently bought together last month. It does not know width. It does not know stock.

Baymard still puts average cart abandonment at 70.22% across 50 studies (updated Sep 22, 2025). Wrong recommendations do not fix that number. They add another untrue claim before checkout.

This is guided selling: what they need in, structured catalog out, ranked. It is not shopping-agent readiness. Readiness asks whether APIs tell the truth. This post asks how you rank once they do.

This is post 13 in the 15-automations map. We have no published store results that say click-through or average order value moved. The matrix you can fill in is the useful part.

The job. Rank products from intent, catalog fields, stock, and compatibility — not because they were frequently bought together.

This week. One category with a real fits graph. Lookups only: search, product, inventory, compatibility. No price change.

A person still signs. Price changes, discounts, published collections, and “will this fit my weird case.”

Skip it when attributes live only in HTML, stock is a nightly dump, or leadership wants “just use frequently bought together.”

Our take: fewer honest recs beat a related-rail that sells inventory you cannot fulfill. Trading volume of suggestions for carts that do not bounce is the deal.

Copy the matrix — Open recommendation-agent-decision-matrix.md. Fill Owner and Approval. Score frequently bought together as secondary. Ship gates: monday-checklist.md.

Why the related rail fails

Shoppers state constraints in language. Facets can encode some of that if they click. Most of them do not. Merchandisers then fill homepage slots and “complete the look” widgets from rules or batch models that never saw this session.

A rec that ignores size, voltage, or fit becomes a return, a where-is-my-order ticket (often shortened to WISMO), or a chargeback. Catalog quality work in AI product catalog management is the input. Ranking is the output. Who breaks without structured attributes: anyone selling accessories, replacements, or sized goods.

How stores handle this today

Keep search and merchandising rules. They own deterministic ranking and promotions. The agent is a conversational ranker over the same facts — hybrid, same thesis as agent vs workflow.

What runs this week (lookups and drafts)

• Map messy intent to a structured query (attributes + constraints).
• Call catalog, inventory, compatibility tools.
• Return a ranked list with evidence fields (sku, attrs used, stock as-of).
• Compare two SKUs on the same schema.
• Propose a bundle candidate when items actually fit.

It does not publish collections. That is post 17. It does not mint a discount to make the rec convert. That is personalized offers. Upsell / cross-sell is the cart-aware slice of this same ranking discipline.

What a person still owns

• Inventory truth, price, tax, restricted categories.
• Promo eligibility in the promo engine — not in the prompt.
• Compatibility as data, not as a vibe (“this bag goes with that camera”).
• Hard filters: out of stock (unless backorder policy), restricted, over stated budget.

Merchandisers own relationship graphs and which collections exist. CX owns “will this fit my weird case.” Legal owns restricted. Nobody should need a human to answer “is size 11 wide in stock” if the tools work.

Escalate when attrs are missing, stock is unknown past SLA, or the shopper asks for a deal the promo engine does not have.

flowchart TD
  intent[Shopper intent]
  agent[Recommendation agent]
  tools[Catalog inventory compatibility]
  rank[Rank and compare]
  present[Present SKUs]
  escalate[Merch or CX]
  intent --> agent
  agent --> tools
  tools --> rank
  rank -->|evidence complete| present
  rank -->|missing attrs restricted unknown stock| escalate

Lookups you attach this week

Week-one allow-list:

Writes in week one: none. Later cart-add is a signed-in tool with a shopper confirm — still not a price write. Knowledge Bases hold size charts and restricted policy. Memory holds this session’s constraints. Do not dump the catalog into Memory.

There is no native Shopify connector. Shopify Admin GraphQL, a headless catalog, or Elasticsearch sit behind your signed-in host.

Ship one category first

One category with a real compatibility graph (accessories, replacements, sized footwear). Ten goldens: fit, stock, budget, restricted, “just give me what people buy.” Fail the last one if frequently bought together wins without attrs.

Do not start with homepage personalization. That is merchandising (post 17) and a bigger blast radius.

For your technical lead

On June 17, 2026, Amazon Bedrock AgentCore Harness reached general availability (What’s New). Agents Classic is in maintenance for new customers after July 30, 2026. A rec bot should not start on Classic InvokeAgent.

Bedrock is the model layer (Converse / Guardrails), not the agent host. Harness (GA June 17, 2026) or Runtime hosts the loop. Gateway + Cedar on any write. Browser and Code Interpreter off unless a named supplier portal has no API — recs should never need a browser. Payment capture stays out of tools. Next.js (or the storefront) is the human-in-the-loop / presentment surface, not the runtime. Strands can orchestrate specialists; it does not give you Gateway, Identity, Policy, or microVMs.

Harness is enough for ≤5 read tools and one team. Graduate to Runtime when a supervisor must stop the rec specialist from calling issueDiscount (Harness + Strands ship map). The store-agents sample is the topology, not a mandate to copy four specialists.

First-party signals we reuse (not eCommerce client outcomes) — Gateway server-side tools cut median tool round-trip ~180 ms → ~95 ms on a B2B CRM assistant (12 tools, ~8k turns/day) — Gateway post. Platform TCO silhouette: support-style AgentCore at 50K sessions/mo ~$791/mo platform + model (decision guide). Model your mix on the AgentCore pricing calculator.

Gateway ~95 ms median from the CRM canary is a platform floor. Your search cluster will dominate. Measure that; do not quote our canary as product-page latency.

What broke — Week-two rec harness ranked from a frequently-bought-together table because “the related-products widget already existed.” A stove accessory that was frequently bought with brand A was recommended for brand B; voltage did not match. Detection: golden eval “fits stove X” returned the co-occurrence SKU; Gateway traces showed getFrequentlyBoughtTogether and no getCompatibility. Recovery: co-occurrence demoted to tie-break; compatibility required for accessory intents; fail the eval if co-occurrence is the only tool. Lesson: co-occurrence is not fit.

That failure is an engineering counter-case, not a client conversion number.

What to do this week

1. Score shopping-agent readiness for one category. Below 8: stop. Fix the PIM.
2. Copy recommendation-agent-decision-matrix.md. Mark frequently bought together secondary. Name Owner and Approval.
3. Expose searchProducts, getProduct, inventory, compatibility. No setPrice.
4. Stand up Harness; Gateway reads; Policy LOG_ONLY; default-deny writes. Browser off.
5. Ten goldens + three must-escalate (restricted, unknown stock, discount demand).
6. Human queue for merch when attrs are missing. Session id + tool trace.
7. Price sessions on the AgentCore pricing calculator. Run monday-checklist.md.

Need Gateway on catalog search before a rec widget hits production? Contact us. Related: Amazon Bedrock consulting, Generative AI on AWS, AWS for retail / eCommerce.

What this post doesn’t cover

• Shopping-agent readiness (data contract) — that post.
• Cart-aware upsell / cross-sell — next in this cluster.
• Personalized discount issuance — offers post.
• Homepage / collection publishes — dynamic merchandising.
• Measured rec click-through, average order value, or conversion from a FactualMinds engagement — we are not inventing them.
• A native Shopify AgentCore connector (there isn’t one).
• Vector DB bake-offs as a substitute for structured attributes.

FAQ

When should we NOT deploy an AI product recommendation agent?

Skip it when size and fit live only in product-page HTML, variants have no parent and child ids, stock is a nightly dump, or leadership wants the widget to “just use frequently bought together.” That is a merchandising rule, not an agent. Fix catalog quality first (shopping-agent readiness). Also skip the older Agents Classic product for new work after July 30, 2026.

What could go wrong if we recommend only because items are frequently bought together?

You attach the wrong accessory, an out-of-stock SKU, or a restricted add-on because co-occurrence does not know fit, voltage, or policy. Frequently bought together is a tie-break after attributes, stock, and compatibility pass. If it disagrees with the catalog, it loses.

What could go wrong if the agent writes prices or invents a discount to close the rec?

Margin leaks and ungoverned codes. Recommendation tools are reads only: search, product, inventory, compatibility. Price changes and promo minting stay in merchandising, with a person approving (human-in-the-loop). Instructions in the prompt do not authorize a write.

How is this different from shopping-agent readiness?

Readiness is the data contract — unique SKUs, structured attributes, current stock, reliable price, APIs. This post is decisioning: given those facts plus intent, what should rank, compare, or bundle. Do not staff a ranking agent on a catalog that still fails the readiness checklist.

Is there a native Shopify connector for AgentCore recommendations?

No. Do not wait for one. Attach the product information management system (PIM), catalog search, inventory, and compatibility as signed-in lookups. Shopify, Magento, and a custom stack look the same: reads first, a person still approving any later write.

Harness or Runtime for a recommendation agent?

Harness (AgentCore) fits a short read-tool list — search, getProduct, inventory, compatibility — and one team. Use Runtime plus Strands when recs sit under a supervisor with hop caps or share a queue with offers. Strands does not provide Gateway, Identity, Policy, or microVMs — AgentCore does. Bedrock is the model layer, not the host.