Building Human-Centric AI Applications: A Step-by-Step Guide
Editor’s note: Afrozy Ara is a speaker for ODSC West this October 29th-31st. Be sure to check out her talk, “Designing Human-Centric AI Interfaces,” there!
As AI adoption accelerates, teams face pressure to create AI that not only boosts efficiency but is also intuitive and trustworthy. While data quality and model performance are crucial, building human-centric AI requires collaboration across data, business, design, and legal teams to ensure AI tools genuinely improve productivity.
In this guide, we’ll walk through steps to build AI applications that empower users and drive better decision-making. To make it more relatable, let’s use a real-world example of an AI tool designed to help an Accounts Payable (AP) team process invoices.
Step 1: Understand the User’s Workflow and Define KPIs
The first step is understanding the user’s role and workflow. For an AP team, this means processing invoices, matching them to purchase orders (POs), and ensuring timely payments. Understanding the user’s day-to-day tasks and challenges helps in designing AI that genuinely fits their needs.
Key objectives for the AP team include reducing errors, speeding up payments, and improving vendor satisfaction. For AI to be impactful, it should align with these objectives. Success can be measured by KPIs like processing time, error rates, and payment cycle times. For example, aim to reduce manual processing time by 30% and improve accuracy by minimizing mismatches between invoices and POs.
Step 2: Identify Work States and Ensure Seamless Movement
Users often shift between high-level overviews and detailed tasks. The user experience should support these shifts seamlessly. For instance, AP managers may start with an overview of all invoices, then zoom into specific discrepancies for resolution. The AI tool should enable quick toggling between summary and detailed views without causing frustration or confusion.
For example, if an AI system flags an invoice that doesn’t match a PO, the user should easily be able to zoom in, review the discrepancy, and then return to the overview. This seamless navigation enhances productivity and reduces the cognitive load on users, making the AI a tool that fits naturally into their existing workflow.
Step 3: Ensure Transparency and Trust
Trust is crucial in AI, especially in finance, where compliance and accuracy are essential. Users need to understand how and why AI decisions are made to trust the system and rely on it in their workflow.
The AI should explain its reasoning clearly so users can validate decisions. For instance, if it flags a discrepancy, it should specify why (e.g., “Invoice amount does not match PO by 5%”). Providing clear explanations helps users understand and accept AI recommendations, rather than questioning or ignoring them.
Moreover, empower users to adjust thresholds and parameters to maintain control. For example, let AP managers tweak the acceptable variance between an invoice and a PO. This flexibility not only keeps the AI compliant with company policies but also builds trust by giving users some control over AI behavior.
Step 4: Empower Users Without Overwhelming Them
Successful AI provides actionable insights without overwhelming users. Use progressive disclosure — reveal more information only when needed. Keep the interface clean and intuitive, alerting users only when issues arise.
The goal is to make AI feel like a helpful assistant, not a complex black box. For example, an AP manager might only need to see the number of successfully processed invoices and receive alerts for discrepancies. By focusing on what’s most relevant at any given time, AI can help users stay focused and efficient.
Step 5: Keep Humans at the Center of Decision-Making
AI should support, not replace, human judgment. While AI can automate tasks like invoice matching, larger discrepancies should still be reviewed by a human. This human-in-the-loop approach ensures that users remain in control of important decisions, particularly those involving high-risk or complex issues.
AI should present its findings clearly and allow users to make adjustments, supporting rather than overriding their expertise. For example, if there is a significant mismatch, the AI might recommend a course of action, but the final call should be left to the AP manager. This approach helps users feel empowered rather than displaced by the AI.
Conclusion
Building human-centric AI is about enhancing users’ ability to work efficiently and confidently. By defining clear KPIs, understanding workflows, ensuring transparency, and empowering users, AI can become an indispensable partner. AI should be like a good sidekick — helpful, smart, and always ready to assist, while knowing when to let the hero (that’s you) take the lead.
To learn more about building AI that empowers and supports users, join Afrozy Ara’s talk at the ODSC conference, where these principles come to life with real-world examples. Don’t miss it!
About the Author/ODSC West 2024 Speaker:
Afrozy Ara is the Co-founder & CEO of LuminaData (https://www.luminadata.ai/), a company building AI coworkers to streamline and automate finance roles like Accounting and FP&A. Prior to this, she was VP, Head of Diagnostics at Incedo Inc., specializing in benchmarking finance operations and implementing AI-driven solutions. With experience advising Fortune 500 companies, Afrozy is passionate about designing human-centric AI that reimagines work in the age of AI.
