Cross-Sell and Upsell Template: Recommend Only When It Fits

A recommendation is not useful merely because it contains the customer's name. It is useful when the team can explain how an option fits a need the customer has stated—and when the team is willing to recommend nothing.

Use this record to start from the customer's current need, remove unsuitable options, and explain every option that remains. It can be completed in a document or spreadsheet before any automated recommendation is configured.

Copy the recommendation record

Customer's stated need and current context:
Information allowed and date checked:
Reason to consider a recommendation now:
Reasons not to recommend:
Options considered:
Compatibility and policy checks:
Current price, stock, and source:
Why each remaining option fits:
Uncertainty and help owner:
When to stop:
Proof of completion and review owner:

Estimated completion time: 20 minutes once the team knows which customer and product information it may use. Copy and complete every line, save the approved record with its test cases, and do not paste real customer data into this article.

Field guidance

  • Use a current purchase, stated goal or approved service interaction; record freshness and never infer sensitive traits.
  • Name the exact eligibility event and evaluate exclusions/suppression before creating candidates.
  • Start with an approved candidate set, then remove incompatible, unavailable, redundant or poorly timed options.
  • Check compatibility, policy, price, stock and source independently; stale evidence produces no recommendation.
  • Explain the factual relationship between each surviving option and the customer's job.
  • State uncertainty and route consequential compatibility questions to an accepted human owner.
  • Define exit events, completion writes and QA criteria for both recommendations and correct no-recommendation outputs.

Microsoft publishes a vendor scenario for context-based cross-sell/upsell support. It informs the workflow shape but does not prove recommendation quality, outcome, or an Easy AI feature. Review the scenario.

Fictional filled example: compatible replacement filter

Customer's stated need and current context: Customer with fictional purifier P-20 asks which replacement filter fits
Information allowed and date checked: Purchase SKU, customer's question, and current compatibility table checked 15 August 2026
Reason to consider a recommendation now: Customer asked a compatibility question; elapsed time alone is not enough
Reasons not to recommend: Open return or complaint, opt-out, unknown model, outdated compatibility source, or unavailable item
Options considered: Fictional filters F-20 and F-30
Compatibility and policy checks: F-20 matches P-20; F-30 does not and is removed
Current price, stock, and source: Link to the current product page; do not repeat a saved price or promise stock
Why each remaining option fits: “F-20 is listed as compatible with P-20. Confirm the model label before ordering.”
Uncertainty and help owner: If the label differs or the source is unavailable, send the model details to the product specialist; request a photo only when permitted
When to stop: Purchase, reply, service issue, opt-out, product change, or missing reliable information
Proof of completion and review owner: Removed options, final explanation, source, and outcome saved; commerce operations checks wrong, irrelevant, and correctly withheld recommendations weekly

The exception is F-30: it is removed after the compatibility check. If the model label cannot be confirmed, the correct result is no recommendation and a clear help route.

Quality checklist

  • Context, purpose and evidence freshness are approved.
  • Every candidate passes eligibility, exclusion and compatibility checks.
  • Price, stock and policy come from current sources.
  • Each option has a factual explanation and uncertainty route.
  • Service issues and suppression can produce no recommendation.
  • QA includes incorrect, irrelevant and correctly suppressed outputs.

Common mistakes and controls

Mistake Control
Using correlation as a need Require a stated job or approved eligibility event
Recommending during a complaint Service state suppresses commercial messaging
Inventing compatibility Check the current official catalog/table
Optimizing only revenue Measure relevance, corrections, complaints, returns and suppression failures

What to do next

Build that test set, and suppress this workflow whenever the post-purchase state record shows an unresolved issue.

Evidence and limitations

This template does not claim personalization performance, product compatibility beyond the fictional example, revenue uplift, or Easy AI capability.

FAQ

Recommended for you