5 AI POCs – 1 Coherent
Workflow
Project Details
Five engineering teams each built a separate AI proof of concept for HubSpot’s Future of Marketing initiative. Each showed a promising capability in isolation—but together they created a fragmented, inconsistent experience no marketer could realistically use. Yvonne led the design effort to connect these disconnected experiments into one coherent, end-to-end AI workflow for Marketing Studio.
The challenge was systems-level: map five independent AI explorations to a single user journey, align interaction patterns across teams, and produce a workflow credible enough for leadership to evaluate, fund, and plan around for 2026.
Role & Scope
Yvonne owned the design vision across all five POCs, serving as the single UX lead responsible for turning independent team outputs into a unified product experience.
- Audited each POC to identify UX fragmentation: inconsistent flows, duplicated patterns, conflicting mental models, and gaps in the end-to-end journey.
- Defined the information architecture and workflow model that connected all five POCs into one continuous marketer journey—from planning through execution to optimisation.
- Established shared interaction patterns using Trellis and HubSpot’s AI Design System, replacing one-off solutions with reusable, system-aligned components.
- Made scope decisions about what to keep, consolidate, or remove so the combined workflow was coherent rather than merely comprehensive.
- Partnered with PMs, engineering leads, and leadership to align the design direction with HubSpot’s 2026 product strategy.
This was not a visual polish exercise. The work required deciding how five separate AI capabilities should sequence, connect, and hand off to each other—creating a shared structure that let multiple teams build toward the same product direction.
Strategy & Discovery
Each POC demonstrated a compelling AI capability on its own. Viewed together, they created friction: different navigation models, inconsistent AI behaviours, overlapping functionality, and no shared sense of where the user was in the overall journey. The core design challenge was reframing five standalone demos as one credible day-in-the-life workflow for a marketer using AI-powered Marketing Studio.
- Mapped each POC to stages of the marketer lifecycle (plan → create → launch → analyse → optimise) and identified where capabilities overlapped, conflicted, or left gaps.
- Designed a shared navigation and entry-point model so each team understood exactly where their AI capability fit within the overall workflow.
- Defined the states, transitions, and decision points across the journey—when AI suggests, when the marketer chooses, what happens next—so behaviour remained consistent and predictable across all five touchpoints.
- Sequenced the POCs into a continuous flow that demonstrated how a marketer would move through the full workflow, surfacing the connections and handoffs that made the experience feel unified rather than stitched together.
This reframing gave every team a common reference point: where their work started, how it handed off to the next step, and what “done” meant in the context of the larger product story.
Workflow Design & Execution
Designing a coherent AI workflow across five teams required more than consistent styling—it required shared decisions about how users move through AI-assisted steps, what they control, and how the system communicates its actions.
- Designed for flow continuity across multiple AI touchpoints, ensuring each transition felt intentional and each step built on the last rather than resetting the user’s context.
- Applied Trellis layouts and components as structural constraints—same shells, same side panels, same action areas—so every POC used a consistent spatial language the user could learn once.
- Aligned AI interaction patterns (conversational panels, insight cards, suggested actions) with HubSpot’s AI Design System, replacing bespoke controls with reusable patterns that reduced cognitive load across the journey.
- Defined clear boundaries between AI-driven and user-controlled actions, making it explicit where the system was suggesting, generating, or waiting for human input.
- Established review states and refinement loops so marketers could evaluate, adjust, and approve AI outputs before moving to the next step in the workflow.
- Evaluated every proposed UI pattern against Trellis and AI guidelines—if a pattern could not be justified within those systems, it was redesigned or cut.
By treating design-system and AI interaction patterns as non-negotiable constraints, teams could move quickly without diverging. The result was a workflow that felt like one product, not five prototypes side by side.
Validation & Alignment
The sprint moved quickly, with shifting requirements and strong opinions from multiple teams. Validating workflow coherence required continuous, structured feedback loops—not a single research phase.
- Ran short, focused alignment sessions where engineers walked through their latest POC builds and mapped them directly back to the shared workflow, surfacing breaks in continuity or consistency.
- Used quick sketches and updated Trellis-aligned frames to present alternative flow connections, giving teams concrete options to evaluate rather than abstract feedback.
- Worked with PMs and leadership to define evaluation criteria—feasibility, user value, and system fit—so decisions about which POCs advanced and which stayed exploratory were grounded and transparent.
- Iterated on interaction patterns based on team feedback, standardising patterns that worked across POCs and refining those that created confusion or inconsistency.
This approach reduced ambiguity incrementally. Each session tightened alignment, and by the end of the sprint, teams shared confidence in a workflow direction rather than holding individual conviction about separate POCs.
Scaling & Reusable Framework
The value of this work extended beyond the five POCs. The workflow model, interaction patterns, and design decisions created a reusable foundation for future AI experiences in Marketing Studio.
- The unified workflow model established a repeatable structure—plan, create, launch, analyse, optimise—that new AI capabilities can plug into without requiring a from-scratch design effort.
- Trellis-aligned patterns and AI interaction components validated during the sprint became reference implementations for how AI should behave across Marketing Studio.
- The evaluation criteria and alignment process used during the sprint provided a template for how HubSpot can assess and integrate future AI explorations into the product.
This shifted the conversation from “which POC should we ship?” to “how do we build on this workflow model as new AI capabilities mature?”—a more scalable question for 2026 planning.
Impact
The unified workflow became the primary artifact used in Future of Marketing research and planning, shaping which AI investments moved forward and how AI work is framed at the portfolio level.
- Delivered a single, coherent AI workflow narrative that leadership used to evaluate feasibility, prioritise bets, and inform the 2026 roadmap and INBOUND strategy discussions.
- Aligned five engineering teams around a shared product direction, reducing duplicated effort and conflicting assumptions.
- Established a reusable workflow model and interaction pattern library that scales beyond the initial sprint to future AI capabilities in Marketing Studio.
- Shifted the organisational conversation from evaluating isolated AI demos to planning around a connected, system-level product vision.