Filters do not answer the question
Color is easy to choose; fit with the space and the intended use is not.
WHAT WE WOULD TESTExplaining relevant differences and acknowledging missing information could help shoppers make a more suitable choice.
All examples
A clearer choice, from question to order.
A shopping assistant could turn product questions into a useful comparison using the store’s current catalog and policies.
“Will it fit a 115 cm space and hold two monitors?”
A visitor needs a desk for a small room. Filters help with color, but the question is more specific. They need dimensions, usable surface information and a comparison they can inspect. We would design the experience around a suitable choice, with a natural path to the product, checkout or the support team.
- The scenario
- A conceptual scenario for a home-office furniture store. Products and demonstration conversations are fictional. They are not commercial results or recommendations for real products.
- What we assume
- Product specifications, variants, stock data and delivery policies. We would check completeness and consistent identifiers across the catalog, cart and order system.
- The boundary
- Payments, order changes, refunds and commercial exceptions remain within approved processes.
Start with the questions that matter.
Working hypotheses
A plausible answer replaces verification
A 120 cm desk can be wrongly presented as suitable for 115 cm.
WHAT WE WOULD TESTApplication rules check firm constraints, such as maximum width or a declared price limit. What cannot be verified stays unconfirmed.
The problem is not the choice
An assistant alone would not fix misleading photographs or incorrect dimensions.
WHAT WE WOULD TESTFirst establish whether the difficulty is selection, missing information or commercial terms.
The evidence we would need
Questions received
Support questions, searches without results, exits from product pages and documented reasons for returns.
Simpler alternatives
Situations where a clearer page, filter or comparison table answers the question more directly. Conversation would be available where additional context is useful, alongside ordinary navigation.
Catalog data
Complete specifications, variants, stock and consistent identifiers across the catalog, cart and order system.
WORKING IDEA / A RECOMMENDATION THAT CAN BE EXPLAINED
Not the most. The one that fits.
CREATIVE PROPOSAL / ILLUSTRATIVE EXAMPLEHow the idea could take shape.
AI would interpret the request and phrase the comparison. Application rules would check firm constraints, and the system would retain the variant identifier and source for each relevant attribute. We would define how options are ranked: fit with the shopper’s requirements is assessed explicitly, and commercial rules must be consistent with how the recommendation is presented.
“I have a 115 cm space and need a desk for two monitors.”
The assistant compares desks narrower than 115 cm in the demonstration catalog and asks how large the monitor stands are.
“Space for both stands still needs checking.”
The 100 × 60 cm model fits the width given. The 120 cm model exceeds the limit and is left out of the comparison.
The product page, without contact details.
The visitor can continue to the product page without supplying contact details. If they ask for support, the handover preserves their measurements and the open question.
From an idea to an experience that works.
What the project would deliver
Information and comparison
The information structure, comparison components, conversation and catalog connection.
Separate access after an order
Public product information and personal order data require different access levels. Order status would use a verified route, and return explanations would use the approved policy.
Tests for real situations
Variant changes, unavailable stock, price updates, unanswered questions and requests to stop communication.
The proposed journey.
From the first question to the relationship after an order.
| Stage | What the shopper can do | What we would build |
|---|---|---|
| Discovery | Arrives through search, a campaign or content | Pages with consistent promises and information |
| Clarification | Describes the space, use and preferences | Short questions, with nonessential details optional |
| Comparison | Sees options and reasons for their fit | Cards with verifiable attributes, tradeoffs and unknowns |
| Decision | Opens the product page and cart | Links to the exact variant and checks at purchase |
| After purchase | Checks a detail or asks for help | A separate, verified route to order information |
| Ongoing relationship | Receives communication suited to the situation | Rules informed by delivery, support status and communication preferences |
Conceptual scenario. Fictional products and conversations.
What happens when…
A choice supported by relevant criteria and accurate information.
Demonstration. Fictional data. Simulated actions. Scripted responses, with no connected AI, personal data, or real operations.
Product catalog
Available width: 115 cm. Two monitors, with stand dimensions not provided.
DESK-100100 × 60 cmDimensions from the fictional record
DESK-120120 × 60 cmDimensions from the fictional record
Available in the fictional catalog only. Two-monitor compatibility: unknown.
Browse the demonstration records directly
DESK-100: 100 × 60 cm. DESK-120: 120 × 60 cm. Finish: oak, fictional variant. Monitor stand measurements are unknown. These are not products for sale on this website.
Fictional data from October 8, 2026. This is not current inventory or appointment availability.
From information to action
Sources and question
Choose a situation and follow what the system can verify, what stays unknown, and when a person needs to step in.
Choosing another situation restarts the demonstration.
Measure the objective, one step at a time.
- A suitable choice
- Selection of a suitable variant and movement to product and cart.
- Orders and returns
- Orders and return reasons, especially those connected to product mismatch.
- Wrong recommendations
- Cases where the assistant recommended something incompatible or concealed missing information.
- Support effort
- Support handling time, repeated questions and corrections.
People choosing to use an assistant may already differ in buying intent from other visitors. To assess its effect, we would select a method suited to traffic volume and document the limits. More messages do not establish a better experience on their own.
Relevant sources: Google Cloud / Grounding overview
What we would hold back, and what would change the plan.
What we would not do yet
- Payments, order changes or refunds handled by the assistant.
- Access to order data without identity verification.
- Commercial messages that ignore a delivery delay or an unresolved issue.
What would change our mind?
If missing dimensions are the main obstacle, completing the catalog may come first. If question volume is low, a static comparison and existing support could be sufficient. Those alternatives would remain part of the investment decision.
First understand. Then test and adapt.
Questions and data
Review questions, searches without results, returns and the catalog.
A comparison shoppers can check
The assistant explains choices and directs people to the existing purchase flow.
Usefulness, not volume
Measure suitable choices, returns and support effort, with documented limits.
An illustrative sequence; timing and investment follow the real context.