Cimulate and Agentforce Commerce search

A merchandiser exports a week of zero-result searches and asks for a commerce rebuild. The export is a list of words customers typed. The rebuild would be a new storefront. Those are different projects. Cimulate is AI product discovery for commerce. The announcement said it would strengthen Agentforce Commerce search. Search sits on a catalogue you already have. It does not become a mandate to throw the shop away. If you run B2C Commerce, that product remains the storefront it already is. This quote is about finding a product, not about replacing the shop that shows it.

The line you can date is this one. "Salesforce completed its acquisition of Cimulate on March 3, 2026." Strengthening search is a scoped change to how queries resolve. It is not a cutover of the site, and it is not a till. Price the query. Leave the rest of the commerce estate on its own pages.

#A failed search is one of three things

Take ten queries from the export. For each, a merchandiser who knows the catalogue can usually say which of three boxes it belongs in. A gap: the product exists in the business and is missing or invisible in the catalogue. A synonym: the product is there, and the customer used a word you did not map. A real absence: you do not sell it, and the honest result is to say so or to show the nearest thing you are willing to show.

Those three boxes are the scope. A partner who prices "better search" has not asked which box dominates your export. The work is different. A gap is a catalogue job, with an owner who can add the product or the attribute. A synonym is a dictionary job, with an owner who can approve the word. A real absence is a design job: the page a customer sees when you mean no, including whether you offer a nearest product or a quiet empty state. Mixing the boxes into one relevance project hides the catalogue work inside a model.

Write the catalogue fields search may use. Title, the attributes you trust, the category, the locale. Fields you do not trust, such as a marketing paragraph that changes weekly, should be listed as weak or excluded. A model that leans on an untrusted paragraph will surface whatever was last pasted there. The merchandiser already knows which field is the messy one. Ask them. Put that field on the deny side of the ranking notes.

Locales are a split, not a footnote. A synonym in one language is not a synonym in another. A quote that says "the catalogue" when you trade in several languages has collapsed the dictionary. Name the locales in the pilot. Leave the others until the first dictionary has an owner.

#What the quote has to name

The export you can actually produce: the query, the locale, the result count, the date. If you cannot export zero-result searches, the first line of the quote is to make that export possible. You cannot steer a dictionary from memory.

Who approves a synonym. One merchandiser, or a small group, with a place the approval is stored. A synonym added in a workshop and lost in a slide will be added again, differently, next month. The dictionary is a record. Treat it as one.

Ranking rules you already wrote. Boosts for stock, for margin, for a campaign, if you truly run them. The pilot should respect the rules you mean to keep. A new model that ignores stock will recommend what you cannot ship. Write whether availability is in or out. If it is in, name the inventory signal and how fresh it is. A daily feed is a different promise from a live one. Do not let the quote assume live.

What Agentforce Commerce means on your contract. The announcement ties the work to that search. This page will not assume the product is on your order form. Check. If it is absent, the quote may say what is required. It may not mark it as included.

The storefront's boundaries. Pages, checkout, promotions, and the till are out unless a single crossing is explicit. A search result that must show a price from a system you name is a crossing. "Replatform while we are here" is not. The till is a different neighbour again. A missed query is not fixed by a new point-of-sale project.

#What cost means before anyone prices it

Cost is the export, the three-way sort of a sample, the dictionary, the catalogue gaps you choose to fill, the ranking rules you keep, the locales in the pilot, the licences you confirm, and the merchandiser hours. This page has no prices. It also has no lift in conversion, because a lift would be your measurement, on your traffic, after a change you can isolate. A proposal that opens with a revenue figure has skipped the ten queries.

Partner time implements the dictionary, connects the fields you allowed, respects the stock rule, and sits with the merchandiser on the sample. Internal time is larger. Someone must say a synonym is safe. Someone must fix a gap in the catalogue, which may mean a product load that has nothing to do with a model. Someone must decide the empty state. If those people are missing, the model will guess, and the guess will be visible on the site.

Out of scope: a new storefront, a new catalogue from scratch, a point-of-sale rollout, a campaign calendar, a promise that every zero-result query becomes a sale. Some queries should stay empty. Forcing a product onto them is how you teach the customer not to trust the page.

Acceptance is the ten queries, marked again after the change. Each one should land in the box you intended: the right product, the right synonym, or an honest absence. A vague sense that search feels better is not an acceptance. Feelings do not survive the next export.

#Migration services, for words and rows

If you are leaving another search tool, migrate the dictionary and the rules you still mean, not every automatic synonym the old tool invented. Review them. An automatic synonym can map a brand to a competitor, or map an adult meaning onto a product for children. The review is the service. A blind import is how that mapping survives the move.

Catalogue gaps found in the sample may need rows added. That is a data task with an owner in merchandising, a source for the attributes, and a check that the new row is actually purchasable. It is not a licence to reload the whole catalogue over a weekend. Limit the load to the gaps you marked. Keep a way to hide a row if the attribute was wrong.

Redirects and bookmarks are easy to forget. A query that used to hit a landing page may now hit a product, or the reverse. List the landing pages you still need. Search should not quietly retire a page marketing still pays to promote. Put those URLs on the brief.

Synonyms that differ by locale stay in the locale. A migration that copies one dictionary into every language will be wrong in all but one of them. Price each locale you include. A cheap "all languages" line is the line that creates the incident.

#The hiring test

People search for the best cimulate Salesforce integration agency. Give the partner ten zero-result queries and a thin note on what you sell. Ask them to sort each query into gap, synonym, or absence. Ask who approves a synonym. Ask whether stock constrains the rank. Ask them to leave the storefront rebuild off the fee. Ask which commerce products are already on the contract. Ask them not to describe a search screen they have not opened in your org.

A pass sorts the ten, and the sorts are ones a merchandiser can argue with. They ask about locale. They ask about stock. They refuse a revenue figure. They keep B2C Commerce, if that is your storefront, on a different statement of work. They treat an honest empty result as a valid outcome.

A fail prices a replatform. It imports every old synonym unreviewed. It promises the empty queries will convert. It calls itself the best agency. The ten queries are enough to see the miss. You do not need a scoreboard.

Second test: a query for something you must not surface, because of age, region, or a brand rule. The partner should put it on a deny list, not "handle it in relevance." Relevance will eventually get it wrong. A deny list is a rule.

#The week you watch the export

Change one locale and one class of synonym. Read the next export for those queries. Look for a gap you filled that still returns nothing, which means the attribute is not the one search uses. Look for a synonym that maps too widely and drags the wrong product up. Fix that one before you add the next hundred words. A large dictionary loaded on Friday is how you spend the next week undoing it.

Tell the people who load products that search depends on the fields you named. If they stop filling an attribute, the dictionary cannot save you. That dependency belongs in the quote as a responsibility, not as a hope. The merchandiser hours continue. They are the cost of a dictionary that stays true.

The till is a third scope, even in the same retail week. PredictSpring and the receipt is the quote for stores, tenders, and inventory events at the counter. A missed search is not solved by a new register, and a receipt is not a synonym.

Send the ten queries. Ask for the three boxes in writing before a fee. A partner who can sort them, and who leaves the shop rebuild off the page, can price a locale. A partner who starts from a new site has answered a fear, not the export.

#The other searches

The project quote is on cimulate implementation cost. That page does not print a figure. Who bought the company is on what happened to cimulate. The connector is on integrating cimulate with Salesforce. Whether to move is on should I migrate from cimulate. The customer fee is on cimulate licence cost. Who to hire is on hiring for cimulate. This page does not rank an agency.