Same question, different brand, different answer: the agent rule matrix

Three brands of a fashion group sell on the same marketplace. Customers ask all three the same question: will this fabric shrink. The answer cannot be the same. One brand speaks formally, another casually; one has a fourteen-day return window, another thirty; one covers shipping above a certain amount, another does not. Three brands of the same group are three different contracts.

The problem is not the rule but where it lives

Most chatbots carry their rules inside themselves. One bot per brand, one rule set per bot, three separate edits for every change. Over time the rules drift apart: one brand's return window is updated, another's is forgotten. Nobody knows which bot speaks with which rule; the customer knows, because they get the wrong answer.

The same problem exists inside the company. The assistant the sales associate uses in store, the agent the call centre uses and the bot on the marketplace can give three different answers about the same product. The problem is not the quality of the AI; it is the rule living in three places.

What the matrix does

SNR Core keeps the rule in the core: brand- and unit-level instructions in one matrix. Rows are brands and units; columns are behaviour areas: tone, return window, shipping policy, forbidden phrases, routing rules, which questions hand over to a person. Whichever brand the helpdesk agent speaks for, it reads that row.

Written once; every agent uses the same row. The marketplace agent in S0 Ecom, the "where is my parcel" agent in S0 Logistics, the sales associate assistant in S0 Store, the supplier correspondence agent in S0 Supply. Four agents, one source of rules.

What happens when a rule changes

If one brand's return window changes, one cell changes in the matrix. Every agent updates at once; no bot is retrained, no prompt is rewritten. A rule change is a management task, not a software task; that is how it should be. The brand manager owns their row; group communications owns the forbidden-phrases column.

Changes are logged. Which rule changed when and by whom; which way the customer satisfaction signal moved afterwards. S0 Customer reads that signal and feeds it back to the matrix. The rule learns too.

The boundary: the matrix does not know the product

The matrix knows behaviour; it does not know the product. The PIM and S0 Ecom's Product Enrichment module know the product: fabric composition, care instructions, size chart, stock status. The agent combines the two. The answer to "will this fabric shrink" comes from product knowledge: one hundred percent cotton, some shrinkage on first wash. The tone, length and the return information added come from the matrix.

This separation matters because product knowledge changes often and behaviour rules rarely. Keeping both in one place pollutes both.

Handover to a person

One column of the matrix is the handover rule: on which question, which word, which customer segment does the conversation pass to a person. If the complaint tone rises, if a legal phrase appears, if the customer is high-value. The handover rule is brand-level too; one brand hands over earlier, another later. The handed-over conversation lands in S0 Customer's Complaint Resolution module and is tracked to closure.

In the field

This matrix runs the marketplace agent across several brands in live retail operations; each brand speaks in its own tone. The field post lists what is live; the SNR Core post places the matrix inside the core. To see the matrix for your own brands: book a demo.

Short answers

Summary

• The problem is not the quality of the AI but the rule living in three separate places.
• The matrix keeps the rule in the core; four different agents read the same row.
• A rule change is a management task; one cell, every agent.
• The matrix knows behaviour, not the product; the agent combines the two.