How does AI handle a multi-product company's support?

When a company sells more than one product, customer service has to handle questions that may span products, with policies that differ by product, and customers who do not always know which product they need help on. AI's job is to route to the right product team's knowledge and policies, share customer context across product boundaries, and shield the customer from the internal structure.

A customer of a multi-product SaaS company messages support: "I'm having trouble with the dashboard." The company has three products; two of them have dashboards. Before AI, the agent had to ask which product. With AI, the customer's account state shows which products they use; the AI infers the likely one; the customer is asked only if the inference is wrong.

What people in the field are saying

Pylon and DevRev are explicit about the multi-product B2B support problem: customers do not segment their experience by your product map. The AI either bridges the gap or the customer has to.

What does AI do well here?

Infers the product from context (the customer's usage, the page they came from, the tone of the question). Routes to the right product's knowledge and policy. Shares context across products: the customer's history with product A is visible when they ask about product B. Avoids forcing the customer through "which product are you contacting us about?" when the answer is knowable.

Where do the policy differences land?

Products often have different refund windows, different SLAs, different feature sets. The AI has to apply the right policy to the right contact. A customer with both product A (annual contract) and product B (monthly) gets different answers for the same kind of question depending on which product.

Where does this go wrong?

The AI mixes products: applies product A's policy to a product B contact. The product teams disagree on policy and the customer sees the inconsistency. The customer has both products and asks a question that spans them; the AI handles half and drops the rest. Each of these is a multi-product-specific failure mode.

What about acquired products?

The hardest case. The acquired product has its own knowledge base, its own policies, possibly its own back-end. Integration takes years. In the meantime, the AI has to know which product's question this is and apply the right rules. This often looks like two AIs sharing one front door.

What is the practical first step?

Make the product field explicit in the AI's context. Tag every conversation with the inferred product, with confidence. Audit a sample of low-confidence cases monthly to verify the inference logic is improving. The cleanest test of multi-product AI is whether the customer ever had to be asked which product they meant.