Designing Human-AI Workflows
This case study showcases how I worked with stakeholders and end-users to lay out the roadmap for integrating AI into existing workflows using a problem-first approach.
The Goal
The rapid advancements in Generative AI technology sparked an internal race to integrate AI into existing workflows. A product team that is building a collaborative financial service wanted to identify opportunities for using AI to increase Advisor efficiency and scale Client capacity.
The Problem
The tech team was eager to deploy several early-stage tech demos to production. However, I recognised the risk of "feature fatigue" and advocated for a problem-first approach. My goal was to move beyond the "race" to implement AI and ensure we don’t overwhelm Advisors with features that didn't address their daily operational hurdles.
Strategic Scoping & Research
To ensure the project was both impactful and compliant with firm-wide standards, I began by auditing the end-to-end Advisor workflow. Utilising existing research, I mapped every task performed by an Advisor during a typical day.
Then, I narrowed the scope to tasks that did not involve direct client communication. This was a strategic decision based on the firm's policy of not using AI for client-facing tasks.
The two tasks selected for further study were:
1. Monitoring client portfolios for performance and opportunities.
2. Creating investment proposals for clients.
Co-Creation With Advisors
I conducted semi-structured interviews and co-creation sessions with highly active Advisors (those managing 20+ ENGAGE clients). We mapped their current challenges and collaborated on "future-state" workflows where AI could take over manual labour while keeping the Advisor in control.
The findings
The research surfaced 10 distinct AI capabilities desired by Advisors. To provide a clear path forward for the product team, I structured these findings in two ways:
-
Evidence-Based Prioritisation: I ranked these 10 capabilities based on frequency of mention, ensuring the roadmap was driven by the most common and pressing Advisor needs.
-
Workflow Integration: Beyond a static list, I visualised exactly how these AI capabilities would integrate into existing Advisor journeys. This mapping clarified which stages should be fully automated and which required the "human-in-the-loop”.
A framework for Advisor-AI Workflows
Beyond specific capabilities, I also synthesised the findings into a 3-part Strategic Framework for how Advisors want to interact with AI:
-
Proactive Intelligence: "Don't make me look for information; bring it to me." AI should crawl data and surface only what requires immediate attention.
-
Guided Decisioning: "Give me options and let me decide." AI should provide actionable paths with clear rationales rather than a single "black box" answer.
-
Minimal Interactions: "Help me be done in a few clicks." The goal is to reduce complex administrative tasks to a few simple interactions.
Prioritisation & Roadmap Development
After identifying the core user needs, I led a two-phased prioritisation process to transform our research into an actionable product roadmap.
-
Assessing Technical Feasibility: "Don't make me look for information; bring it to me." AI should crawl data and surface only what requires immediate attention.
-
Defining Business Impact: Then, I presented the technically viable shortlist to the product and business leadership. Together, we analysed each workflow to determine which would deliver the highest Return on Investment (ROI) and provide the most significant value to the firm's growth strategy.
This dual-lens approach allowed us to identify the top three workflows for immediate development, ensuring that our first step into AI was both technically sound and commercially impactful.
Next Steps
With the top three workflows prioritised, the project has transitioned into a high-fidelity research phase to ensure successful execution. I am currently leading deep-dive discovery sessions to define the granular requirements for each workflow.
My focus remains on making sure the final implementation satisfies user needs as well as supports the firm’s long-term goals.
