Salesforce’s newest commerce search story sounds deceptively simple: turn on an AI-native search engine that understands what shoppers mean instead of forcing them to speak in keywords.
For a B2C Commerce team, that is not the interesting part.
The interesting part is what happens to years of search tuning after the switch.
When Agentic Commerce Search is enabled, several familiar native B2C search controls no longer determine search and browse results. Synonym lists, stop words, stemming exceptions, Einstein Search Dictionaries, Einstein Predictive Sort, native merchandising rules and existing sort rules do not simply transfer into the new engine. Salesforce documents that explicitly.
That makes Agentic Commerce Search less like adding an AI feature and more like changing the operating model of product discovery.
Cimulate is now part of the Salesforce search stack
Salesforce completed its acquisition of Cimulate in March 2026 and has since brought the technology into Agentforce Commerce as Agentic Commerce Search.
The product became generally available in July. Salesforce describes it as an AI-native discovery engine spanning search, browse and recommendations, with a custom small language model created for each retailer using catalogue, behavioural and synthetic shopping data. It also powers guided discovery inside Shopper Agent.
The October B2C release makes the timing more interesting.
Agentic Commerce Search can now understand shopper queries across locales, and Salesforce has added “Shop the Store” support so natural-language searches can take local store availability into account. A shopper asking for “rain jackets in stock near me” can therefore combine intent with store-level inventory instead of receiving a generic product list.
That is useful progress.
But the first implementation question should not be, “How good is the AI?”
It should be:
What currently makes our search experience work, and which of those controls survive the move?
The migration surface is easy to underestimate
Most established B2C Commerce sites do not have “a search configuration.”
They have years of accumulated search behaviour.
A synonym list fixes vocabulary the catalogue never handled properly.
A redirect sends a common query to a campaign landing page.
A merchandising rule keeps a commercial priority visible.
A custom sort option supports a category strategy.
Predictive Sort influences relevance.
A search index query helps an operations team understand what is actually in the index.
These things become institutional knowledge. They often stop looking like architecture because the business has lived with them for years.
Agentic Commerce Search changes that boundary.
Salesforce says native page and content search continues to use B2C Commerce. Redirect rules remain. Existing facet definitions continue to support category pages.
But several search-result controls move elsewhere or cease to apply.
Native synonym and linguistic configuration is replaced by the model’s interpretation.
Einstein Search Dictionaries and Einstein Predictive Sort no longer drive the result set.
Native merchandising rules do not carry over; teams recreate the required commercial logic in the Agentic Commerce Search Rules dashboard.
Existing Business Manager sort rules do not control ACS search and browse results.
Some native capabilities, including slicing groups, the Search Index Query Tool and Einstein personalized suggestions, are not supported in the same way.
That is not necessarily a reason not to migrate.
It is a reason to treat migration as a search-product project rather than a feature toggle.
AI does not eliminate merchandising. It changes where merchandising happens.
The vendor narrative around AI-native discovery often emphasizes reduced rule maintenance.
That is attractive. Search teams can spend an unreasonable amount of time maintaining synonyms, boosts, buried products and one-off exceptions.
But “fewer rules” should not become “less commercial control.”
A fashion retailer may need to suppress a product for a campaign.
A grocery retailer may need availability to dominate relevance.
A marketplace may need seller eligibility to constrain the result set.
A brand may have contractual or inventory reasons for prioritising one product family over another.
Those are not search-engine accidents. They are business decisions.
Salesforce retains merchandising controls in Agentic Commerce Search, but the control surface changes. That means the migration backlog should include every rule that exists because the business deliberately wanted a particular outcome, not merely every rule that happens to exist today.
That distinction can save a lot of bad migration work.
A rule created three years ago to compensate for poor synonym handling may no longer be necessary.
A rule that protects margin, inventory exposure or a contractual obligation probably still is.
Do not compare engines with five curated demo queries
Commerce teams should not decide the migration from “black dress,” “running shoes” and a handful of impressive natural-language prompts.
Build an evaluation set from real behaviour.
Include high-volume queries, zero-result queries, misspellings, brand terms, SKU searches, category-like terms, ambiguous queries, long natural-language requests, seasonal language and terms where merchandising teams currently intervene manually.
Then measure business outcomes, not just semantic elegance.
Did the shopper click?
Did they add to cart?
Did the result respect inventory?
Did the correct variant appear?
Did a commercially protected product disappear?
Did a query that used to redirect still behave appropriately?
Did a customer-service or merchandising team lose a diagnostic tool it depends on?
That is a much stronger migration test than asking whether the new results “look smarter.”
Search analytics becomes an architecture question too
One easily missed detail in Salesforce’s documentation is that Agentic Commerce Search does not introduce a separate analytics dashboard for B2C storefront search performance. Salesforce directs teams to B2C Commerce Reports and Dashboards for search and browse measurement.
That matters when a business currently relies on a third-party search platform with its own operational analytics.
Before migration, map who uses search data and why.
Merchandising may care about zero-result terms.
Product teams may care about search-to-PDP conversion.
Engineering may care about latency and failures.
Commercial teams may care about category exposure and revenue.
Customer experience teams may care about the language shoppers use when catalogue terminology does not match customer terminology.
A new search engine should not make those questions harder to answer.
Global and local search make the source-of-truth problem more visible
The October release adds two capabilities that are strategically more important than they first appear.
Multilingual intent means a shopper no longer needs to express a request in the catalogue team’s preferred language.
Local-store discovery means search relevance can depend on what is actually available nearby.
Both increase the value of search.
Both also increase its dependency on trustworthy commerce data.
If store inventory is stale, “in stock near me” becomes a broken promise.
If product relationships are inconsistent across locales, a model can understand the sentence perfectly and still recommend the wrong item.
If the catalogue has the same commercial concept represented differently across brands or regions, AI can expose that inconsistency faster than keyword search did.
Intent-aware search does not remove the need for product and inventory governance.
It makes bad underlying truth easier for the shopper to interrogate.
A practical migration gate
Before enabling Agentic Commerce Search broadly, we would expect a B2C team to answer seven questions.
Which native search controls are active today?
Which of them represent genuine business policy versus historical search-engine workarounds?
Which capabilities do not have a one-for-one equivalent in ACS?
How will merchandising rules be recreated and owned?
How will semantic and native search be compared with real traffic?
Which metrics define a successful migration?
And what is the rollback path if relevance, commercial control or operations degrade?
If those answers are documented, the technology choice becomes much easier to evaluate.
If they are not, “turn it on and let the model learn” is not an enterprise search strategy.
The important change is not that search understands sentences
Natural-language search is visible. It demos well. It is easy to explain.
The more consequential change is behind it.
Salesforce is moving B2C Commerce product discovery away from a stack of lexical configuration, Einstein search features and native merchandising controls toward a different search engine with a different relevance model and a different control surface.
That can be a meaningful improvement.
But the right migration question is not:
Can Agentic Commerce Search understand our shoppers?
It is:
Can we move to it without losing the commercial decisions, operational visibility and fallback controls that make our current search experience dependable?
That is where the implementation starts.
