AI with strict / balanced / flexible interpretation modes

The user picks how literal the AI should be when mapping the request to existing supported filters.

Why useful

Adds controllability without changing the core system; bridge to future AI maturity.

Key risk

Adds cognitive load — users may not understand the mode tradeoffs.

Example query

"Affordable business hotel with breakfast"

Section — Interactive demo · Click ✨ AI Filter to try

Interactive demo · Click ✨ AI Filter to try

All filtersHotel ratingHotel nameLoyalty & brandsIncludes breakfastPrice

Section A — End-to-end interaction flow

1 · Default

All filters✨ AI FilterHotel ratingHotel nameLoyalty & brandsIncludes breakfastPrice

›

2 · Click AI Filter

All filters✨ AI FilterHotel ratingHotel nameLoyalty & brandsIncludes breakfastPrice

›

3 · Input + mode toggle

All filters✨ AI FilterHotel ratingHotel nameLoyalty & brandsIncludes breakfastPrice

StrictBalancedFlexible

Affordable business hotel with breakfast

›

4 · Applied (Balanced)

All filtersPrice ≤ $200×Business services×Breakfast×

›

5 · Switch mode → chips update

StrictBalancedFlexible

All filtersPrice ≤ $200×Business services×Meeting facilities×Breakfast×High-speed Internet×

Section B — Static state breakouts

Strict — only exact matches

Includes breakfast

"Affordable" and "business" not mapped.

Balanced — likely matches

Price ≤ $200Business servicesBreakfast

Flexible — broader inferences

Price ≤ $200Business servicesMeeting facilitiesBreakfastHigh-speed Internet

Mode toggle inline

StrictBalancedFlexible

Mode tooltip

Balanced — maps the most likely supported filters from your request.

Default mode badge on applied chips

Breakfast×Mode: Balanced

Switch mode after apply → recompute

Mode changed → mapping recomputed automatically.

Default selection persists per user

User's last-used mode is remembered.

No supported match

✨ Strict

No matches in Strict; try Balanced.

Section C — Notes & tradeoffs

What changes in the bar: Same applied state as other concepts; mode is a control inside the AI input.
Existing pills: Untouched.
AI representation: Standard AI chips; mode is shown as metadata.
Edit / remove: Per-chip remove; switching mode re-computes the chip set.
Cross-LOB scaling: Mode concept reusable; each LOB tunes what 'Flexible' means in its domain.
Unsupported terms: Strict mode skips more aggressively; Flexible mode skips less.

← Concept 08: Suggested prompts

All concepts

Concept 10: AI session banner →