Merchandising Agent: what it does for a parts catalog

The Merchandising Agent in Salesforce B2B Commerce does back office work in conversation: it boosts or buries products in search, creates products and manages buyer accounts, and its merchandising skills summarise order value and conversion and draft promotions. For a manufacturer’s parts catalogue the useful part is search ranking: superseded parts down, current kits and wear parts up. Product master data stays in the ERP or PIM, and I would have every change the agent makes reviewed. Merchant Agent is listed in both editions; Salesforce dates the B2B Merchandising Agent to August 2026.

What is the Merchandising Agent?

Salesforce B2B Commerce, which Salesforce now also markets as “Agentforce Commerce for B2B”, has two kinds of agents. The Buyer Agent talks to dealers, which I cover in B2B Buyer Agent: ordering through WhatsApp and SMS. The Merchandising Agent works for the people who run the store.

Salesforce uses three names here, and they appear on different pages:

• Merchant Agent is the line item under Agentforce on the B2B pricing page1, in both the Growth and the Advanced edition.
• Merchandising Agent (B2B) is listed on the Commerce innovations page2 with “GA August ’26”. Salesforce calls it the back office agent that “comes to B2B” and says you can tell it to boost or bury products in search, spin up new products or manage buyer accounts, done in conversation instead of clicks.
• Agentforce Skills for Merchandising is the title of the Trailhead unit3 that explains KPI summaries and promotions.

What can it do for a parts catalogue?

A parts catalogue is a store where search decides almost everything. Dealers type a part number, an old number or a short description, and they expect the current part at the top. Three of the agent’s jobs fit this well.

Boost and bury. Boost and bury rules raise or lower products in search and category results. Salesforce’s July 2026 post on B2B Commerce4 names the examples: high margin, surplus or promoted products up, low stock or end of life items down. For spare parts this means superseded numbers down, the successor part and complete repair kits up. Today a merchandiser sets these rules by hand; with the agent, a request like “bury all filters marked end of life” becomes a rule.

New products. The agent can create products. That suits items that exist only in the store, such as a bundle of wear parts for a service interval. Parts that come from the ERP should keep arriving through the interface.

Buyer accounts. The agent can manage buyer accounts. For a manufacturer that means requests like “add the new branch of dealer X as a buyer account”. Which buyer group the account lands in decides what it sees and what it pays, so I would treat this skill with the most care. How buyer groups and entitlements are structured is the topic of the chapter Catalogue, search, entitlements.

Summarise KPIs, draft promotions

According to Trailhead, Agentforce Skills for Merchandising retrieve and summarise average order value (AOV) and conversion rate for a store and name derived insights such as the weakest category. Trailhead’s example is a B2B store. The skills also suggest promotions by matching predefined promotion templates in Salesforce Commerce and generate the text, terms and configuration.

Trailhead calls this “semiautonomous”: you review the promotion details and make changes before you send it out. For a manufacturer, promotions in a parts store are often small: a discount on a maintenance kit before the season, or free shipping above an order value. The agent saves the clicks. Whether a discount is allowed under dealer agreements and pricing rules stays a decision for sales.

Two KPIs matter more in parts stores than AOV: zero result searches and the share of order lines through the store. Search analytics through Data Cloud, which the July post lists, shows zero result searches. Whether the agent can summarise those as well is not documented on the pages I read.

Generative AI, Agentforce and Data 360 come first

Trailhead lists the prerequisites: Salesforce generative AI, Agentforce, bots and Data 360 activated for the store. Setup runs in the Commerce app with guided steps, and the topics and actions come preconfigured for Commerce. For licensing, Trailhead points to the Salesforce Help article on enabling Agentforce Skills for Merchandising.

The editions include Data Cloud credits: 250K in Growth and 500K in Advanced, according to the pricing page. What the agent consumes per request is not published. I would therefore measure consumption in a sandbox with real requests before go-live; how such an estimate works is shown in Flex Credits: estimating Agentforce cost per action.

The store’s limits also apply to the agent, and both rule limits are hard limits:

100

boost and bury rules per store

100

synonym rules per store

70

searchable and filterable attributes plus sort rules combined

An agent that writes a new rule for every request uses up the 100 quickly. Someone has to clean up rules regularly, whoever created them.

What should the agent not do?

Product master data belongs in the ERP or PIM. Material number, unit, sales status and successor part are maintained there, and the store receives a copy. If the agent changes a product in Salesforce, the next nightly load overwrites it, or worse, the store and the ERP diverge. I explain where which product data belongs in B2B Commerce product data: PIM, ERP or Salesforce?.

My rule for a parts store:

Task	Agent	Reason
Boost and bury in search	Yes, with review	Store-only setting, reversible
Promotion from a template	Yes, with release by sales	Price effect, needs approval
Product that exists only in the store	Yes, with review	No ERP master
Part from the ERP material master	No	ERP or PIM is master
New buyer account	Only with a named approver	Decides visibility and prices

Every agent change should be traceable: who asked, what changed, who released it. Which logs Salesforce offers for this is covered in Audit trail for AI agents in Salesforce: which logs?.

Worked example: 60,000 spare parts for 300 dealers

Assume a manufacturer of compressors runs a B2B store with 60,000 spare parts for 300 dealers. Every quarter, engineering supersedes a few hundred part numbers. Today one merchandiser maintains search rules by hand and is two months behind.

With the agent, the merchandiser asks after each ERP update: “bury every part whose sales status is superseded and boost its successor”. The agent proposes the rules, the merchandiser checks a sample and releases them. Once a month the merchandiser asks for the zero result searches of the last month and passes missing old numbers to the PIM team as synonyms or attributes. The search side of this is covered in B2B Commerce search: finding parts by part number.

Clarify the licence and included skills

1. Clarify the licence. Ask which skills the included Merchant Agent covers and how usage is counted.
2. Fix the data first. A clean sales status and successor part in the ERP are what make good boost and bury requests possible.
3. Start with search ranking. It is reversible, stays within the store and shows results in zero result searches quickly.
4. Set review rules. Promotions and buyer accounts only with a named approver.
5. Watch the limits. 100 boost and bury rules per store, with a monthly cleanup.

The chapter Agents, migration, rollout of the B2B Commerce masterclass places the Merchandising Agent next to the Buyer Agent and the rollout. For the question of who owns which data between ERP, PIM and Salesforce, see Salesforce architecture.

Sources4

1. 1B2B pricing page (opens in a new tab)salesforce.com
2. 2Commerce innovations page (opens in a new tab)salesforce.com
3. 3Trailhead unit (opens in a new tab)trailhead.salesforce.com
4. 4July 2026 post on B2B Commerce (opens in a new tab)salesforce.com