Designing the trusted assistant — agentic design principles, a decade early
How an AI-era design philosophy, applied to enterprise payroll in 2015, lays the foundation for how I build AI products today
Innovation lab · FinTech · B2B · B2C · agentic design · explainability · data visualization · multi-user platform · regulated FinTech · AIThe challenge
For the previous 40 years, ADP processed payroll in batches on a mainframe computer. If you had to re-calculate something, it took a lot of time and sometimes required ADP to help or write custom code to achieve desired outcomes.
Customers wanted a modern, real-time, easily-configurable, cloud-based technology. Between 2010 and 2015, ADP’s Payroll Innovation Lab (aka “Pi”) built that real-time, web-based engine.
My contribution
I was hired as Director of UX, Payroll Innovation in 2015 to:
build a multi-disciplinary team (UX, Research, prototype development, and Content Strategy)
grow a culture of innovation
collaborate with Dev to execute meaningful experiences that leverage the capabilities of the new cloud-based engine
collaborate with Product and Dev to design for a multi-product landscape (payroll, time, tax, Implementation, IT, and adjacent products such as HR)
socialize the experience vision with multiple ADP product teams, C-suite, and industry analysts
Why this project still matters
The principles we defined for Pi in 2015 — automate the mundane, show your work, guide users through complexity, speak their language — are the same principles driving well-designed AI products in 2026. We called the system a "trusted assistant" before that phrase meant anything in an AI context. We designed explainability into payroll calculations before explainability was an AI design discipline. We built wizard-based wedge experiences to reduce the cognitive load of complex decisions before anyone called them wedge experiences.
I didn't know it at the time, but this project was where I developed the design instincts I now apply to every AI product I build.
The strategy
Hire the right people
Recruit the necessary UX team disciplines:
6 designers, 5 researchers, 2 content strategists, and 1 Ruby developer (to build prototypes)
Create a culture of innovation
Implement Design Thinking methods and processes
Focus on problem framing to ensure we solve the right problems
Standardize and democratize research techniques to limit bias, increase capacity, and increase trust
Encourage controlled risk-taking, based on insights, data, and hypotheses
Encourage early sharing of concepts and ideas, to eliminate waste and shorten timelines
Encourage trust and collaboration by setting up:
Design pairs (who provide constant feedback and support)
“Secret powers” to help team members know who has skills they can leverage (e.g. visual design, video production, etc.)
Be a coach and mentor
Hire team members who are self-motivated professionals
Be a facilitator who empowers designers and eliminates roadblocks
Understand each team member’s career goals
Be a trusted advisor (show I’m sincerely interested in their growth)
Use case studies to track, evaluate, and drive desired growth
Be open to feedback and self-improvement
Formally articulate my commitment to the team
Video: An excerpt from a presentation about the UX process (3 min)
I built a research program that changed how Product made decisions — establishing structured research requirements that ensured every feature recommendation was evidence-backed, not opinion-driven. Product began inviting us into long-term roadmap planning as a result.
What we learned
Payroll is labor intensive
Payroll practitioners want insights that the organization can use to make decisions, not just data (“Make payroll strategic!”)
They’re open to payroll that runs itself if the system can be trusted to let them review and resolve issues before payroll is complete
In many payroll departments, there’s one person who’s used ADP for several years (the guru) and feels they have to explain it to the others
If the guru can’t help, customers rely on ADP support (who provide excellent service!)
It takes too long to set up and change ADP payroll configurations
Payroll is stressful!
There are just a couple of days after employees report their time to complete payroll
If payroll is wrong, it’s likely that an employee will quit
Practitioners are personally and criminally liable for mistakes, so they double-check everything
They’re uncomfortable searching for patterns of fraud; they want tips about what to look for
They get user-friendly experiences at home, so why not at work?
Video: Overview of research outcomes (2 min)
Pay is confusing for employees
Employees want:
More and better information
To know how pay is calculated
To know that withholdings are correct
And to get paid immediately
“Why should the company keep my money for 2 weeks? I have bills to pay today!”
The vision
We created a vision where we delight users
First, we looked to ADP’s corporate pillars to inform the important goals we need to achieve.
Then, we explored attributes of delightful, useful payroll experiences that align with the brand and what we learned during research.
As we evolved designs, we evaluated them based on how well they aligned with the vision and brand.
Guiding principles
Provide explainability (show the calculations. Users need to see and verify what the system decided before they commit to it. In a regulated domain like payroll, this isn't a nice-to-have; it's the difference between a system users trust and one they work around. This is the same principle now called explainability in AI product design — and it's why I build it into every AI experience from the start.)
Provide insights, not reports (but make the raw data readily available)
Automate the mundane (like when designing for agentic, we let the system do the heavy lifting. The best AI products don't just answer questions; they handle the repetitive, error-prone work so users can focus on decisions that require human judgment. We designed Pi to process, calculate, and surface what mattered — so practitioners could stop managing payroll and start managing exceptions. That's what agentic design means, whether the intelligence underneath is a rules engine or a large language model.
Speak the user’s language (avoid “ADP-isms” that only insiders know)
The system is a trusted assistant that keeps things running smoothly
Clarity through visualizations — Designing for financial data at scale meant making complex, high-stakes information scannable without losing the detail practitioners needed. Every dashboard and calculation view is a data visualization problem first, a layout problem second.
Video: Intro to the Pi experience vision (2:37 min)
We crafted narratives
For each major body of work, we conducted ideation sessions with subject matter experts (SMEs) and customers to explore stories of delightful experiences that solve real user problems.
Stories are easy to remember and help us focus on the user’s work and goals
We told stories such as short narratives, storyboards, mock press releases and datasheets, and videos. Stories take as little as 20 minutes to create and are easily tested in moderated or unmoderated tests.
“Please explain my pay” video (1:17 min)
We defined goals and scope for each core capability
For each idea that we decided to pursue, we wrote a 1-page brief that helps the team align and keep focused on:
the business problems to solve
user problems to address
stakeholders to keep informed, collaborate with, or who will approve the effort
insights to leverage in the solution
the MVP and the metrics we’ll track to validate the efforts as we iterate toward an ultimate (“concept car”) design
Document jobs-to-be-done (JTBD), prototype, and test
The researchers, content strategists, and UX designers collaborated to design UIs and content that support one or more stories. We:
journey-mapped the way users worked to document the JTBD
synthesized the journey and insights to help Product and Dev determine the MVP, capabilities, constraints, and subsequent enhancements
user-tested long-term concepts using the lowest fidelity artifacts (narratives, sketches, wireframes, mockups, and prototypes) to inform the direction
evolved the UI microcopy and other content as the design evolved
built and managed a design system that facilitated development and standards, and supported Legal reviews
analyzed messaging throughout the experience to strategically drive the funnel from first awareness through to full engagement
conducted peer reviews before testing
built prototypes using Axure or live Ruby code (because of the data required to test some scenarios)
validated outcomes with Google Analytics, Full Story, and other tools
Some challenges to overcome
To ensure we provided solutions that the company could sell and support, we regularly shared our vision and concepts with the company’s Brand, Strategy, Marketing, and Support teams, We shared the stories of our innovations as they evolved to ensure solutions aligned with the current and future company brand and product portfolio.
Other Product teams needed to see the vision
In payroll there are four distinct user types — practitioners, employees, implementers, and IT — each requires a different level of data visibility, workflow complexity, and technical vocabulary. Designing coherently across all four was the central design architecture challenge.
To help our Product partners have confidence in our qualitative techniques and hypothesis, we exposed them and our Dev partners to various research sessions.
We established a program to ensure that all Product partners attend at least one of the following types of sessions:
Site visit (contextual inquiry) where we see real users performing the types of tasks that we need to automate or improve
Participatory design with users who would design something so we can learn how they think about their solutions and workarounds
User tests, where we test a concept (story or low-fidelity prototype) to see and how users respond to a design
As a result, Product:
trusted that our feature recommendations were not opinions but evidence-backed
collaborated with us early during long-term planning, because they wanted to include our insights in their roadmap planning
After we validated concepts in the innovation lab, we needed to work with other teams to integrate the solutions into existing product lines. Those teams would also continue to iterate and evolve the work. We needed them to believe in the vision and carry the work forward.
So we regularly shared what we learned during research and invited those teams to participate in frequent design reviews so they could understand our design choices and provide additional insights.
We needed Product to trust us
Leadership needed to support the vision
We told the story of our innovations using mock datasheets, press releases, and videos to ADP’s C-suite, industry analysts, and other stakeholders. It was important to help them see how our solutions supported the company vision and potentially impacted company value.
This also gave us the opportunity to receive their feedback and insights.
A good feature that can’t be sold, may not have value
We need to bridge and acknowledge the entire user ecosystem
Then we continued to share the vision
We shared insights with internal product teams who would assume responsibility for the innovations and:
held regular sessions where we demonstrated innovations
participated in a company re-branding exercise
developed datasheets to internally share plans and outcomes
The solutions
In this section see the many applications we created during this 2-1/2 year project.
The colors vary because some were added to existing ADP applications and needed to adhere to the colors and design patterns used in those products.
We defined a new visual design that uses a lot of white space and avoids non-functional visual artifacts. Many users work all day in the application; we wanted to make it easy to focus on key messages and data. We also wanted the application to be a calming experience; in research, users told us the new visuals were more modern.
We published a style guide with recommendations for:
tone and voice
techniques for displaying and emphasizing information
information hierarchy and navigation
Tip: click screenshots to display a larger version
Ava, a personal assistant explains pay and saves time
Employees can see how pay is calculated on their MyADP mobile app
Ava, the chatbot, answers common payroll questions
When an employee’s payroll changes, Ava notifies them (many employees only look at the bottom line and don’t notice or understand changes)
A UX designer designed the framework for the chat experience — the content strategist worked with a machine language expert to train the bot to answer questions and provide explanations.
Usage of the MyADP mobile app increased by 6%.
My ADP chatbot: Employees can readily see how they were paid and ask questions.
Video: A demo of the early chatbot concept (2:17 min)
Anyone can do payroll
Employees can see how much they might be paid before payday, based on hours already worked and mandatory deductions. With this capability, employees can choose to be paid before payday for wages already earned.
At the end of the pay period, the practitioner simply reconciles the taxes and makes any necessary adjustments
Factoids:
Companies who offer same day pay attract more talent
Pay Me Now mobile app: Employees can see how much they’ve earned and mandatory deductions to get paid before payday.
Paycheck Calculator app: Small employers who need to produce 1 or 2 paychecks can see easily process payroll. This was available on ADP’s website for free.
Video: A demo of an early version of the paycheck calculator (1:32 min)
Payroll practitioners get a dashboard that delivers insights
Focus on what’s important
See outcomes based on preset thresholds
Run payroll automatically if nothing unusual is found
See details when desired
See potential fraud and the severity
Share insights and data with others in the organization
Eliminate menus and allow users to search for features (no more “3 levels deep menus”)
Remember scheduled tasks and automatically display required information (e.g. tax payments)
Payroll dashboard: Payroll practitioners can readily see the most important information and still access the underlying details.
Calculations are explained using natural language
Display calculations in natural language to take the complexity out of interactions and increase practitioner confidence
Hide the details to eliminate clutter — allow users to expand a section to see more
Calculation logic: Payroll practitioners, employees, and implementers can readily see the inputs and formulas to understand how an item is calculated.
An interactive view of payroll events explains changes
Practitioners see a visual payroll history
Items that changed significantly or exceed thresholds are highlighted
Selecting a node on the timeline displays inputs and events that contributed to the outcome
When the practitioner wants to test payroll changes, a timeline displays the impact of changes on past and future events
Interactive calculations: Payroll practitioners can see changes in an individual’s pay or over time and expose the details behind changes. This helps them troubleshoot or see the input of changes before committing to them.
Payroll is one of the most regulated domains in enterprise software. Designing for auditability — making it possible for a practitioner to trace every output back to its inputs — is the same design requirement I now apply to AI systems in regulated industries like healthcare and financial services. The principle is identical: humans need to be able to verify what the system did, and why, before they act on it.
Video: A demo of dashboard solutions (1:02 min)
Employees can understand and get tax withholdings right
Users can choose to complete forms or use a wizard
A wizard guides users to answer questions — the app uses those answers to determine the proper exemptions
To avoid clutter, we display verbose descriptions in a panel the user can choose to display
Unfamiliar tax terms are replaced with legally compliant everyday language
This structured, question-by-question approach — rather than asking users to figure it out themselves — is the pattern I now call a wedge experience: a designed layer that reduces friction, collects exactly the input the system needs, and produces a reliable outcome without requiring users to understand what's happening underneath.
Research insights behind the solution
Employees usually change withholdings once every few years, so they’re unfamiliar with the terms and concepts and need to research descriptions on federal and state sites
Some employees know the appropriate number of exemptions and just want to complete the form (so we don’t force them to use the wizard)
Most employees over-withhold, so they get a tax refund (not realizing that the IRS is holding their money interest-free throughout the year — meanwhile employees pay interest on credit card debt)
2018 tax withholding forms: Employees can figure out how many exemptions they should claim.
Configuration is self-service
Payroll calculations are defined in pay policies (descriptions written in natural language)
To configure how pay is calculated, practitioners can see and change the descriptions
A workflow helps practitioners manage changes and get pay policies reviewed, approved, and published
Practitioners can export pay descriptions and add them to the company’s HR handbook
Because ADP is the payroll expert, we display tips in a panel on the right
Self-service configuration: Practitioners can readily see inputs and formulas and change them. They also see tips from ADP that help them make decisions.
Video: A demo of the new configuration process (2:16 min)
Global deployment is easy
A global payroll configuration tool helps ADP employees and consultants configure payroll solutions for any country (ADP provides payroll products in over 100 countries)
Because country-specific pay policies are based on globally-shared policies, users can easily see which global policies are available and available for use in their country
Users can search for policies instead of scanning lists and hierarchies to find them
Natural language policy descriptions allow users to understand and change how pay is calculated
An Operations dashboard helps Pi IT monitor system performance and quickly respond to issues.
Easy global configuration: ADP employees who set up and manage ADP’s payroll configuration for a country, can readily see available global payroll policies and assemble them into a policy catalog for a country.
Payroll engine monitor: Pi IT can readily see how the payroll engine is performing to quickly resolve issues.
Pay statements are more informative
We replaced acronyms and pay codes with regionally appropriate terms and phrases
We show the history of vacation hours so employees can determine if the totals are correct (research shows they previously had to go to the employee portal and calculate it)
A visual summary highlights income and deductions
Employer contributions are separated from other deductions, so organizations can better show employees the benefits they provide to them
Employee pay statements: Employees can readily see how they were paid and how vacation time was calculated.
What Pi taught me about AI design
Every principle that made Pi successful maps directly to what makes AI products work today.
Show your work
Users won't trust a system they can't inspect. Whether it's a payroll calculation or an AI-generated recommendation, the design job is the same: make the reasoning visible enough that users can verify it and correct it when it's wrong.
Automate the routine, elevate the judgment
The goal was never to replace payroll practitioners — it was to handle everything that didn't require them, so their expertise could go where it actually mattered. That's still the goal with AI: not automation for its own sake, but automation that amplifies human capability.
Guide users through complexity
Open-ended interfaces fail when the decision space is large and the stakes are high. Wizards, structured inputs, and guided flows — what I now call wedge experiences — produce better outcomes because they meet users where they are, not where we wish they were.
In regulated domains, governance is design
Payroll compliance shaped every transparency decision we made. Healthcare, financial services, and other high-stakes AI domains require the same thinking: human oversight and auditability aren't features you add later. They're the foundation.
The results
The Pi platform launched to internal adoption across ADP's client base and contributed directly to reducing reliance on ADP support for payroll troubleshooting — one of the original drivers for the lab's investment.
“Automatic Data Processing's (ADP) next-gen payroll platform blew expectations out of the water...
Shares of ADP stock are up by 3% to $145.18 per share”
Business Insider article
