A cross-sell workflow can be useful when it connects a genuine customer need with a relevant additional service. It becomes unhelpful when the CRM simply identifies products the customer has not yet bought and creates a campaign. Absence of a purchase is not proof of demand. For a small UK team, the sensible starting point is to identify a clear complementary use case, verify the records that support it and let the account owner decide whether the timing is right. The comparison below considers human-led CRM recommendations alongside automated audience selection without inventing conversion percentages.
Consider an existing customer buying a product that has a documented maintenance requirement. A relevant service discussion might help, but only if the product, installation date, coverage and customer preferences are accurate. A company buying a similar-looking but different product might not be eligible. A human colleague should be able to understand why the CRM suggested the additional service instead of trusting an opaque score.
1. Choose a customer problem before choosing the offer
Define what the extra product or service would enable. The offer may address a genuine compatibility requirement, reduce a known operational risk or support a future change the customer has explicitly discussed. Write down the customer circumstance that makes the offer appropriate. 'Has not purchased service B' is not enough; that describes a database gap, not a reason to contact someone.
Ask account managers to identify three realistic situations where an additional offering would help and three where it would not. Test the proposed rules against both sets before automating anything. A rule that matches many unsuitable accounts is likely to create additional manual filtering and may damage confidence in the CRM.
Distinguish cross-sell from upsell and renewal. A complementary service is different from a more expensive version of an existing subscription or an approaching contract extension. These activities may need separate ownership, approval and communication rules even when they use the same customer history.
2. Establish reliable evidence for the trigger
Identify which application owns product purchases, active services, delivery status and customer identity. A CRM may hold relationships and notes, whilst an order system records what was actually sold. Do not assume an old proposal became a purchase or that a returned item is still being used. Verify source dates and the relationship between the customer record and relevant transaction.
Build a concise evidence packet with product or service reference, observed event, proposed complementary offering, relevant conditions and an account owner. Do not fill missing compatibility information with an AI-generated guess. Where multiple branches or subsidiaries share a name, check that the offer relates to the correct organisation and location.
Sample the output of every rule after changes to the catalogue, pricing or CRM data model. A recommendation that was valid under last year's product range can become misleading when dependencies or availability change. Give responsibility for maintaining those mappings to someone who understands the offering.
3. Exclude customers for whom contact would be inappropriate
Before creating an outbound task, check for unresolved service complaints, sensitive account situations, known procurement restrictions, recently declined offers and communication preferences. Someone dealing with a major delivery failure is unlikely to welcome a generic promotion, even if the additional service would otherwise be useful. Separate potential product relevance from the appropriateness of contacting the customer now.
For electronic marketing, UK rules concerning the subscriber, consent, applicable soft opt-in and objections must be assessed. The Information Commissioner's Office explains that a person can object to use of personal data for direct marketing and such processing must stop. Keep a suppression mechanism that remains effective when records are imported or connected systems resynchronise. An existing customer relationship does not grant permission for any promotional email whatsoever. ICO guidance on respecting marketing preferences is relevant to this design.
An internal opportunity suggestion need not automatically become a campaign. For high-context accounts, a personal review task may be the more responsible first step, allowing the owner to check timing and legitimate communication routes.
4. Explain why an opportunity was generated
A task saying 'sell additional product' is barely more useful than a blank call list. Show the event that triggered the suggestion, the complementary benefit, any uncertainties and the source record. The account owner should be able to approve, defer or dismiss the recommendation with a short reason. Keep the explanation concise enough that staff actually use it.
When two offers could both be relevant, apply clear priority and frequency limits rather than sending multiple messages. Product availability, installation compatibility and financial terms may require checking by the relevant team. A model may help summarise records for review but must not present predicted intentions as statements the customer made.
Avoid using potentially sensitive personal characteristics to infer buying interest. Where the CRM holds identifiable individuals' data, apply appropriate lawful processing, access and data-minimisation rules. Do not train account owners to treat a marketing score as a verified fact about a person.
5. Choose the next action according to relationship context
For some accounts, the relevant discussion fits a scheduled annual review. For others, a colleague may ask whether the customer is experiencing a problem that the additional service solves. A simple informative message may be suitable where marketing rules and preferences allow it. There is no reason every matching record must create an immediate outbound call.
Make the account owner responsible for the actual decision, especially where pricing, compatibility, promises or existing contractual commitments are involved. A proposal can be drafted using approved product information, but a confirmed price or delivery date must come from the relevant authoritative system. Stop generic follow-up once the customer replies with a different need or declines further contact.
Where an opportunity becomes a formal quote, link the discussion to the established sales process instead of maintaining a separate cross-sell-only shadow pipeline. Keep the originating recommendation for analysis without allowing it to dictate commercial outcome.

6. Record dismissals and corrections as useful evidence
A good workflow gives staff a quick way to label a recommendation irrelevant, mistimed, based on wrong data or already discussed. Those are different problems and need different repairs. An irrelevant product match suggests the rule requires improvement; a mistimed recommendation may be perfectly valid once an unresolved customer issue is addressed.
Preserve necessary provenance when product records are corrected. If the CRM believed a customer owned an item that had been returned, update the source rather than merely suppress one marketing task. Where automatic processing retries after a connection error, use stable identifiers to avoid duplicate recommendations or repeated customer messages.
Review a small sample of both accepted and dismissed opportunities at regular intervals. The feedback should improve customer-data quality, catalogue mapping and team judgement, not simply increase the number of recommendations marked positive.
7. Test the same accounts under manual and automatic selection
Build a representative test set using appropriately protected records. Include a valid complementary product, an incompatible item, an unresolved complaint, a contact who opted out, a recently returned purchase and an account already discussing the proposed service. Agree the expected handling before running the rules. For several cases, the correct result is no outbound communication.
Compare two approaches: a salesperson reviews a targeted CRM list, or configured rules prepare an eligible suggestion for their consideration. Measure correct recommendations, staff corrections, review effort and actual relevant conversations. Do not assume every accepted suggestion became a sale or claim that the workflow caused an increase without an appropriate baseline.
Simulate a failed CRM sync or missing purchase record. The system should mark the opportunity uncertain and queue it for review rather than inventing a purchase history. Check that communication suppression remains effective after retries and imports.
8. Calculate the commercial value and operational cost together
Consider licences, integration upkeep, data cleansing, account-manager review, product mapping and customer-response handling. A low-cost manual review may outperform an elaborate automation where opportunities are infrequent and customer relationships are complex. At larger volumes, automated detection can reduce missed relevant cases, but only when records and permissions remain dependable.
Judge outcomes by useful conversations, appropriately qualified opportunities, customer feedback and the number of irrelevant contacts prevented. A large list of 'cross-sell opportunities' is not a benefit if staff must discard most of it. Avoid made-up savings or conversion rates. Capture enough information to learn which offerings genuinely complement each other and where the customer had an unmet need.
The lasting objective is a CRM that assists human judgement instead of replacing it. Pick a narrow complementary offer, document its evidence, respect contact preferences, pilot both the ordinary and difficult cases and expand only when customers and account managers see practical value.
Sources: ICO — Accuracy principle and ICO — Respect people's preferences. These support data integrity and contactability controls, not invented sales performance statistics.
