Reimagining NYX.today
Transforming a fragmented adtech SaaS into an integrated conversational AI ecosystem.
I was the only designer at Nyxify, redesigning a live AI marketing platform for Indian agencies whose demos kept stalling on a fragmented interface. I replaced six tool silos with one conversational workspace, inside frozen engineering architecture, lifting task efficiency 24% and cutting support tickets 40%.
View Site
Goal
Transform a fragmented, click-heavy SaaS layout into a centralized, conversational AI workspace that lifts the successful campaign generation rate, while working strictly within existing engineering and architecture constraints.
Challenge
The original modular homepage layout hid the true interconnected power of NYX.today, forcing users into tool-specific silos that increased cognitive load and delayed campaign execution. We needed to transition the platform into a centralized, conversational workspace without altering the underlying application architecture or breaking developer constraints.
Outcome
A complete product overhaul that unified creative generation and campaign deployment into one conversational workspace, raising task efficiency by 24%, cutting support tickets by 40%, accelerating activation by 30%, and reaching 100% WCAG AA compliance.
I delivered a complete product overhaul that transformed a fragmented, click-heavy SaaS layout into an integrated conversational AI ecosystem, helping agencies move from creative generation to campaign launch without switching tools.
Project overview
Nyxify Technologies operates NYX.today, an AI-driven marketing automation platform built for high-growth brands and mid-to-large digital agencies across India. The company operates on a B2B SaaS subscription model, scaling revenue based on active workspace features and processed advertising volume.
- Neo: the primary conversational orchestration hub
- Pixieo: a unified image and video generation core
- Xeno: a single-prompt, fully autonomous campaign engine
- North Star Metric: Successful Campaign Generation Rate
- Demos were stalling on the interface, not the AI
- Agencies described the same break in the same place
- Three audits of the live product, one repeated failure
- Problem statement
- The gap was between the tools, not inside them
- Three directions considered, one that engineering could actually ship
- Wireframing the split-screen before committing a pixel
- Neo routes, Pixieo makes, Xeno ships
- Four problems that only exist in AI interfaces
- 43 contrast failures, found and fixed
- The final product
- What it moved
- What I got wrong
Demos were stalling on the interface, not the AI
NYX.today could already generate creative and deploy campaigns. What it could not do was look like one product while doing it. Demonstrations with marquee Indian agencies kept stalling because the interface failed to show the automation actually working, putting enterprise pipelines worth lakhs at risk over perceived instability.
The trigger, and the market gap it sat in
During the pilot phase the product team saw that users increasingly expected generative AI to behave conversationally rather than through rigid navigation. Configuring an ad campaign through endless form fields felt outdated next to modern conversational tools, which created pressure to abandon the modular homepage entirely.
Mapping the Indian AdTech landscape showed why the gap mattered. Most creative marketing software stops at static visual or content generation with no downstream implementation. Most performance campaign utilities start at budget deployment with no built-in creative generation. The opening was an ecosystem that bridged asset creation and live media execution inside one continuous context.
Agencies described the same break in the same place
I ran testing loops with twelve participants across the Indian agency and brand landscape. Every group described losing their rhythm at the same seam: the handoff between making an asset and spending money on it.
- “The tools work fine individually, but constantly moving between them to download and upload files completely slows down my creative rhythm during busy client seasons.”
- “Creating an ad asset in one menu and then setting up the budget inside another feels like restarting my entire workflow from scratch every time.”
Who was in the room
As the solitary UI/UX designer I owned the translation of this pivot into a usable interface. Daily alignment ran with the Product Management team on milestones and scope trade-offs, with periodic syncs to the founder on strategic shifts and fundraising direction. Because the development team sat in a separate department, I built the design-to-development pipeline from scratch: asset handoff, functional interaction states and systematic design QA inside their engineering sprints.
- Digital Agency Account Directors (4): recruited via the partner network in Mumbai and Bengaluru, managing ad budgets exceeding ₹5 lakhs monthly, evaluating software on speed and multi-client transparency.
- Performance Marketers (5): sourced from tech co-working hubs in Delhi NCR, highly technical, requiring immediate access to campaign ROI metrics, budget burn rates and asset generation history.
- D2C Brand Founders (3): reached through startup ecosystem referrals, with limited formal design or ad-buying expertise, relying on the platform to generate instant, trustworthy creative assets.
Three audits of the live product, one repeated failure
Reviewing what was running when I joined turned up three problems that all traced back to the same cause: tools built as islands. Each one broke user trust in a different way, and none of them could be fixed by making an island prettier.
What each audit found
- Asymmetrical homepage card silos: five primary launch cards in an inconsistent arrangement with a lone sixth card (Generate Video Ads) dropped onto its own row, creating an awkward anchor, dead space, and project history buried under text-heavy blocks.
- Chaotic analytics visualisation: a dual-axis area chart plotting a yellow spend line and a magenta CTR trendline over an identical coordinate grid without distinct background separation, producing an unreadable intersection layout.
- Disconnected creative editing: a detailed photo manipulation space with right-hand property sliders (blur, transparency, brightness, contrast, saturation, exposure, warmth, shadows, highlights) that functioned in complete isolation from campaign deployment.
Problem statement
The original modular homepage layout hid the true interconnected power of NYX.today, forcing users into tool-specific silos that increased cognitive load and delayed campaign execution. We needed to transition the platform into a centralized, conversational workspace without altering the underlying application architecture or breaking developer constraints.
The gap was between the tools, not inside them
The individual engines worked. The workflow between them did not. Five steps with two file transfers became four steps with none, which is the entire redesign stated in one line.
- Old, fragmented: generate asset → download file → open campaign tool → upload again → launch ad.
- New, unified: prompt input → concurrent creative generation → instant review → live launch.
Three directions considered, one that engineering could actually ship
The platform was live with active enterprise configurations, so a database or workflow restructure was off the table. Every deep UX problem had to be solved through interface logic alone. That constraint eliminated two of the three options before aesthetics entered the conversation.
- A. Improve the existing modular dashboard: refines component metrics but preserves tool context switching and never bridges the creation-to-deployment gap. Rejected.
- B. Conversational overlay panels: a floating assistant window that blocks the asset previews and analytics tables underneath it. Rejected.
- C. A unified split-screen Neo Hub: removes fragmentation while keeping output assets visible alongside the live chat context. Selected.
The trade-off, and the stakeholder tension behind it
Rather than build alternative multi-step wizards, I recommended replacing the static homepage entirely: merging text-to-image and text-to-video into one multimodal application called Pixieo, integrating the single-prompt campaign infrastructure called Xeno, and tying both together under a central chat orchestration layer called Neo.
To keep this inside a separate team's bandwidth I postponed complex file management and asset sorting, and prioritised a clean split-screen workspace that shows previews alongside live conversation. Straightforward to implement, and the part users actually felt.
Alignment syncs surfaced a three-way tension. The founder wanted maximum visibility of every sub-agent calculation, to prove the AI's power to investors. Engineering wanted the simplest possible data grid to stay inside rapid backend sprints. Users wanted fast workflows stripped of technical noise, to hit agency client timelines.
I resolved it with an expandable persistent sidebar paired with responsive action chips: implementation stayed simple, agent sub-processes hid behind contextual confirmation states, and technical validation remained available on demand.
Wireframing the split-screen before committing a pixel
I mapped the conversational workspace and every supporting surface at low fidelity first, validating hierarchy against the engineering architecture before any visual design existed.
Neo routes, Pixieo makes, Xeno ships
One prompt enters the Neo Hub, which detects intent and orchestrates the rest end to end. Neo routes to Pixieo for creative generation and to Xeno for deployment, and both converge into a single live campaign stream.
Navigation architecture
To give a customisable layout without database-level changes, I built a global navigation structure with a dedicated user pinning zone.
- Core orchestration anchor: the Neo chat core, the central conversational homepage.
- System toolsets index: the Pixieo workspace and the Xeno automation engine.
- User-customised pinning zone: a drag-and-drop area for pinned feature slots, such as the Pixieo generation feed or a Xeno real-time analytics card.
- System configurations: profile, billing and API connections.
Four problems that only exist in AI interfaces
A conversation-driven interface introduces behaviours traditional SaaS never has to answer for. Each one got a specific layout response rather than a general one.
- The blank slate: users froze at an empty chat container. Quick-start prompt chips around the input give visual entry points instead of a cursor.
- Agent confusion: users did not know which engine to use. Neo parses intent and activates Pixieo or Xeno in the background, so nobody has to choose.
- Automation mistrust: media buyers feared autonomous spend. Xeno cannot deploy without rendering a confirmation card stating audience, budget caps and channel breakdown.
- Losing your place: deep conversational flows filled the screen. A strict immutable global sidebar keeps core workspaces anchored at all times.
The four layout decisions, with the reasoning
Usability testing drove the last two. I made the sidebar persistent, added agent state indicators that flash which engine is running, and shipped the drag-and-pin mechanism so any deep tool module could be locked to the navigation sideboard.
- Persistent global sidebar. Problem: deep conversational flows filled the screen and users lost their position. Insight: complex multi-agent workspaces need a fixed visual baseline. Decision: an immutable global sidebar anchoring core platform spaces.
- Split-screen content architecture. Problem: scrolling or window-swapping to check generated assets broke focus. Insight: prompts and their outputs must share one visual frame. Decision: a twin-panel model, conversation left, live previews and data grids on the wider right canvas.
- Custom navigation pinning. Problem: performance marketers used optimisation reports far more than creative feeds. Insight: efficiency scales when operators personalise their own navigation. Decision: a drag-and-pin area inside the global sidebar.
- Signal-driven recommendation cards. Problem: automated suggestions were ignored because uniform grids hid financial risk. Insight: automation insights need distinct grouping and explicit impact labelling. Decision: a modular feed isolating category, data signals and risk, each with a prominent command button.
43 contrast failures, found and fixed
Agency monitoring sessions run long, under variable lighting. I owned the accessibility overhaul personally and took the platform to full WCAG AA compliance.
- Contrast repair: 43 individual failures fixed across high-density analytics components and input fields.
- Token architecture: a functional colour token system anchored on a high-contrast primary purple (#7648EF) with an accessible secondary text variant (#B39DFF), exceeding 4.5:1 against dark canvas backdrops.
- Screen-reader infrastructure: descriptive aria labels and precise focus states across all charting elements and operational toggles.
The final product
One workspace, four surfaces, no file transfers between them.
- Neo: the cluttered legacy dashboard replaced by a focused conversational panel with context-aware action chips.
- Pixieo: independent creative tools merged into one space, text-to-image and video side by side.
- Xeno: multi-step wizard fields replaced by single-prompt setup with high-visibility validation cards.
- CamPulse and assets: scannable cards and readable analytics, with integrations in both themes.
What it moved
Measured against internal databases, marketing logs and partner tracking systems rather than estimated.
- Task completion efficiency +24%, from database transaction timestamps tracking prompt input through to live budget deployment.
- Support tickets -40%, from customer success logs over a 60-day window after rollout.
- Workspace activation +30%, from GA4 events tracking multi-agent prompt configuration.
- Demo-to-trial conversion +35%, from enterprise CRM deal-stage movement during live demonstrations with Indian agency prospects.
- Enterprise evaluation cycle -28%, measured as average days in active trial before technical sign-off.
- Deal qualification rate +20%, from sales feedback showing less client hesitation about reliability during pitch cycles.
Why the pipeline metrics, and not revenue
Framing impact through conversion velocity rather than speculative financial projection let the design prove itself as an enterprise sales acceleration engine without inventing numbers. The premium feel of the interface let the product team advance high-value agency accounts through evaluation far more predictably than the legacy layout allowed.
What I got wrong
I assumed performance marketers wanted dense, compact data to maximise screen real estate, and designed an ultra-compact table packing every asset output and campaign metric into one view. Testing with PMs and trial users disproved it immediately: people found it overwhelming and could not find the actions. I introduced structured tab groups and clearer spacing, which fixed readability without touching the backend.
What the constraint taught me, and what comes next
Designing inside strict engineering boundaries was the real lesson. Being the only designer meant validating user needs against business reality with product managers while managing handoff to a separate development department. I discarded several complex interaction concepts that exceeded implementation capacity, and learned that a perfectly executed standard component beats an ambitious custom interaction that fails in development.
The design system this established sets up the next cycles: natural language campaign editing inside the optimisation view; multi-agent collaboration flows where Pixieo and Xeno swap underperforming creatives automatically on cost metrics; AI-generated monthly retrospectives benchmarked against regional e-commerce patterns; and cross-workspace memory so Neo retains historical prompt settings across accounts.
“The problem wasn't that the individual tools were weak. The problem was the massive operational gap between them. Closing that gap, without touching the underlying architecture, is what turned a fragmented product into a conversational AI ecosystem agencies could trust.”