Intent and customer context
Use the stated goal, current journey and behavioural context where permitted. A first-time visitor can still receive useful recommendations from explicit needs.
Recommendation Engine
Build Recommendations Around Intent, Context and Business Rules
Create recommendation logic that does more than “customers also bought”. Match products to the reason someone is buying, explain the fit and respect the rules of your business.
Built around your stack
Possible connections, scoped to your systems and permissions.
ShopifyCommerce platform
HubSpotCustomer context
BigQueryBusiness signals
PostHogJourney measurement
Start with the decision a recommendation must support.
A useful alternative, compatible accessory and upgrade serve different customer decisions. Define the recommendation placement and desired next action before choosing ranking techniques. The logic should distinguish a hard eligibility rule from a preference or commercial priority, so a high score cannot make an incompatible product acceptable.
See the workflow
See the logic behind a useful recommendation.
Select a step to see how the context, output and controls connect.
Context
Capture the reason for buying.
A running kit and a casual footwear purchase require different recommendations, even if they start from the same product.
Illustrative workflow
- Intent
- Prepare for a trail run
- Product context
- Running footwear
- Preference
- Carry water comfortably
Rank
Apply constraints before ranking.
Match accessories to the use case, check availability and respect merchant exclusions before preparing a bundle.
Illustrative workflow
- Eligibility
- Compatible · in stock
- Bundle candidate
- Water bottle and running pack
- Merchant rules
- Exclusions checked
Explain
Give the customer a reason to consider it.
Make the relationship between the need and each recommendation visible, with alternatives when a tradeoff matters.
Illustrative workflow
- Why this set
- Footwear · hydration · carrying
- Alternatives
- Compare size and storage
- Next step
- Review the suggested bundle
Recommendation inputs.
Catalogue and product relationships
Compare attributes, compatibility, alternatives, accessories and bundle relationships using reliable product information.
Inventory and merchant rules
Respect availability, exclusions and business priorities. Define freshness expectations and safe behaviour when a source cannot be checked.
Outputs that make the next step easier.
Ranked products and alternatives
Present a small relevant set with reasons to consider each option and clear tradeoffs.
Bundles and compatible accessories
Suggest combinations that satisfy documented compatibility rules and customer needs.
Upgrades with explanations
Explain the additional benefit and pricing context instead of assuming the most expensive option is the best fit.
Give merchants visible rules and overrides.
A merchant interface can expose exclusions, priority rules, preview scenarios and override history. Evaluate proposed rule changes before applying them broadly. If a product is unavailable or a compatibility relationship is uncertain, the system can remove it, flag it for review or explain the limitation. Business control belongs in the recommendation logic and the operator experience.
Measure whether recommendations help.
Define a baseline and measure recommendation click-through rate, product engagement, add-to-cart and conversion for the relevant journey. Average order value may be useful for a bundle or accessory experience. Assisted revenue should only be reported when attribution supports it. Review irrelevant matches, stock-related failures and abandonment alongside commercial metrics; no outcome increase is assumed.
From ranking assumptions to an evaluated MVP.
Lumethica can map product relationships and rules, define representative recommendation scenarios, prototype the presentation and build a focused integration. Rules or existing catalogue signals may be enough for a first version. More complex models should be introduced when the available data and measured results justify them.
Related
Explore related solutions.
FAQ
Common questions
Do we need historical purchase data?
Not necessarily. A first version can use stated intent, structured attributes and merchant rules. Behavioural or purchase signals can be evaluated when sufficient permitted data exists.
Can recommendations respect stock and compatibility?
Yes. Eligibility rules can filter products before ranking, with freshness checks and defined fallbacks when inventory or compatibility data is unavailable.
Can our team change recommendation rules?
A scoped merchant interface can support rule changes, previews and overrides with appropriate permissions and an audit trail.
How do we know the engine is working?
Agree on representative relevance scenarios and a commercial baseline, then evaluate customer actions and error cases in a pilot. Revenue attribution depends on the measurement setup.
Define the first useful release
Bring the workflow. We can help define and build the product.
Start with your catalogue, current process and the outcome you want to improve.