I kept principles intentionally tight. These acted as a decision filter when trade offs appeared:

• Clarity over density: Prefer structured experiences that reduce cognitive load, especially where data is complex or unfamiliar.

• Progressive disclosure: Put heavy detail (research depth, attribute level scoring) behind deliberate interaction, so the default experience stays approachable.

• Governance without friction: Make enterprise controls (admin editing, brand protection) explicit, but don’t punish non admin users who simply want to generate concepts.

• Trust through transparency: If we show a score, we also show the reasoning structure and the levers users can pull to improve outcomes.

Redesign the landing experience around momentum and re entry

Problem:
Users needed a clearer entry point and a faster way back into relevant exploration, without starting from a blank prompt every time.

Approach:

• Defined a landing experience with smarter prompts; dynamically generated from conversation history for returning users, and curated defaults for first-time users.

• Established the landing page as a stable home for Brand Configuration, so high-impact setup isn't buried mid-conversation.

Why it mattered:
It reduced blank page paralysis, supports returning users, and creates a predictable place for high impact configuration.

First time usage (left) vs returning usage (right)

Make Brand Configuration enterprise safe

Problem:
Brand inputs need careful management because they can be accidentally edited, set up inconsistently, or made visible to the wrong users.

Approach:

• Defined Brand Configuration as a first-class product capability; allowing users to add, edit, and manage brands, with logo support to influence concept imagery.

• Shaped a permissions model where clients can choose between open editing and admin-only editing, supporting governance for large organisations while preserving flexibility for smaller teams.

• Kept brand selection optional in the concepting workflow, so the tool still supports unbranded exploration.

Why it mattered:
This balances two competing enterprise needs; brand safety and creative iteration speed. Admin only editing reduces the chance of “brand drift” and accidental overwrites, while optional selection avoids blocking ideation when brand setup isn’t ready (a real concern raised in training feedback).

Various viewer vs admin user flows to consider

Brand configuration form for admins on first usage

Force filter confirmation up front to protect relevance and the conversation

Problem:
In a conversational UI, it’s easy for context to drift. If market/category/stage aren’t stable, outputs become inconsistent and users can’t tell what changed.

Approach:

• Introduced an explicit filter confirmation step at the start of each conversation, capturing Super-category, Category, Market, Innovation Stage (where applicable), and optional Brand selection.

• Defined single-select category and market as a product clarity choice, aligned with technical constraints.

• Introduced help content for innovation stages so users understand what they’re choosing.

• Added a toggle to show only Black Swan supported filters (external data source), preventing users from selecting contexts that won’t return the expected analysis.

• Made “start a new conversation to change direction” an explicit workflow rule, keeping each thread coherent.

Why it mattered:
We accept a small amount of upfront structure in exchange for much higher relevance, better explainability (“this score was for this context”), and fewer “why did the tool do that?” moments. It’s also a guardrail against conversational drift, which is a common failure mode in chat based products.

Replace the old quick prompt research flow with auto generated structured research

Problem:
Testing showed users were clicking through prompts without engaging in meaningful research, then generating concepts with shallow context.

Approach:

• Replaced the prior quick-prompt sequence with auto-generated structured research immediately after filters are confirmed.

• Defined research sections that are scannable and role appropriate.

• Ensured research adjusts to chosen brand and innovation stage, so concepts inherit the right strategic frame.

• Preserved flexibility: users can ask follow ups within research or move straight to “Generate Concepts”.

Why it mattered:
The workflow stops assuming users want to manually complete research steps and instead provides the right context by default, while still letting motivated users go deeper. It respects expert users’ time while improving concept quality for everyone.

The restructured workflow supports 3,800–6,500 concept searches per month, with ~81% week-one retention. This suggested users return once context is established, rather than abandoning after a shallow first visit.

Making scores understandable, actionable, and defensible

Problem:
Scores are only valuable if users can interpret them and act on them. We also saw internal concern about mismatches between tables, roll up scores, and explanatory text; a trust risk.

Approach:

• Defined an integrated scoring model within the concept view, with an overall rating built from three metrics; Demand Momentum, Distinctiveness, Brand Fit.

• Introduced a Key Data view to provide transparent, granular support for the score: short metric explanations plus attribute level scores, roles (hero/supporting), and importance weighting.

• Improved concept review for stakeholder sharing by enabling full-screen concept imagery.

Why it mattered:
This uses progressive disclosure to keep the primary concept experience fast and legible, while offering auditability for users who need to defend decisions. It also reduces the risk that scoring becomes a black box, especially important when Spark is positioned as evidence based and grounded in proprietary data.