Robert Half Direct

Humanizing AI & Machine Learning to create the job platform of the future.

Project Details

  • Product Design

    Primary Practice Area
  • 2019

    Year
  • VP Practice Director, Experience Design Consulting

    My Role

Organization Details

  • Staffing & Consulting

    Industry
  • $7B

    Yearly Revenue
  • 13,000+

    Employees

Background

Robert Half prides itself on a class-leading physical job search experience. Grounded by the principles of Match, Speed-To-Market, & Personal Touch, Robert Half's Recruiting Team are world-class talent finders. In the early-to-mid 2010's, the organization underwent a large-scale digital transformation, unifying the internal digital ecosystem, and setting the stage for scalable productivity. Once scalable productivity was achieved internally, the target was then set on achieving scalable productivity in its service offerings for clients & candidates.

Project Overview

The Problem Space

The digital transformation successfully created scalable productivity for its internal employee. However, the digital customer experience was almost completely unchanged by the transformation.

With an almost non-existent digital experience for client users, & a very limited digital experience for candidate users, Robert Half wanted to bring the digital experience fully in-line with their best-in-class physical experience, unifying the physical and digital worlds, to launch their vision of the future of job platforms: an AI-Driven self-service product.

Challenge

My team was engaged to design Robert Half Direct, a self-service job search platform, where scalable productivity is achieved through AI & Machine Learning.

The Design Process

For Whom We Designed

Client Users

Client users are HR Professionals or Hiring Managers whom engage Robert Half to hire new talent.

Candidate Users

Candidate users are job seekers whom engage Robert Half to connect them with clients looking to hire new talent.

Setting The Stage: Mapping The Physical Experience

To set the stage for research, I met with stakeholders & field employees to generate a deep understanding of both client & candidate experiences.

I mapped the entirety of the existing (ideal) physical experience, between both user groups, which is facilitated by & through Robert Half Recruiters. This comprehensive experience map ensured my team & I had a solid foundational understanding of the physical experience to use a point of reference, to which we could continuously refer, as we worked through the design process.

Understanding What Users Actually Want & Need

Designing a research strategy

At the outset of the project, I had a blank slate, with full autonomy to explore, create, & design.

This fostered a mindset for innovation, but without pre-existing insights, I needed to develop a comprehensive research strategy capable of producing high-quality, actionable insights that I could utilize to define the mission, goals, design strategy, & initial ideation points.

Synthesizing Research Data

Affinitizing the data

To effectively synthesize the research, I affinitized the research data, and created affinity diagrams to organize the research into actionable insights, on a categorical basis.

Client Users - Top Insights

  • Prefer a desktop experience, over mobile
  • Want to be in control of the hiring process
  • Want the flexibility to use their own  process
  • Want to see only the most relevant 5-20 matches
  • Want to communicate directly with candidates

Candidate Users - Top Insights

  • Want both mobile & desktop experience
  • Want to consistently know where they are in the hiring/interview process
  • Want to know the hiring process up-front
  • Want to know what jobs best fit their needs
  • Want to communicate directly with clients

Key user insights

The insights from initial user research were high-value; the data helped generate a firm understanding of the users, their wants & needs, and the friction they commonly encounter. From a generative standpoint, the insights served as the foundation for our concept discussions surrounding the design of Robert Half Direct.

Making Research Actionable

My focus was on creating high-value research assets that help foster my team's human-centric mindset, serve as signposts for user empathy, & highlight/emphasize the core user problems we hope to solve through a new product driven by AI & Machine Learning.

Personas based on real users

Robert Half is made up of 6 lines of business, comprised of users who have different & unique needs & expectations. I created personas on a per-line of business basis, in order to highlight the unique needs of each user sub-group. As the Experience Design Director, this allowed me to focus on setting a minimum bar of quality to impact all users, while enabling my design team to focus on the unique needs of each user, on a per-line of business basis.

Empathy maps

First, I created empathy maps on a per user basis, then bubbled-up the most critical insights into summary empathy maps for each primary user group.

Client user journeys

I mapped several different client user journeys to understand the range of outcomes, positive, negative, and neutral.

Candidate user journeys

For candidate users, I also mapped a full range of journeys; positive, negative, and neutral.

Combined user journeys

Since the new experience, we are thinking about, connects two user groups through a single app, it was important to map both journeys on top of each other. This helped my team visualize the friction points of each user group, as they occur contemporaneously.

The MVP: Robert Half Direct

Without restrictions, we ideated solutions, and arrived at a concept model which we called Robert Half Direct, an AI-driven, self-service job platform.

The concept for the MVP was driven by incorporating the best of what users like in the physical experience & designing solutions that solved for the friction points we uncovered in the physical experience; then combining the two, to design a new fully digital experience for Robert Half Direct.

The candidate flow

Based on our initial research, we found Candidate Users wanted to primarily engage with Robert Half through a mobile experience.

The client flow

Again, based on our initial research, we found Client Users overwhelmingly preferred to engage with Robert Half through the desktop experience. There was little-to-no interest in conducting their job search through a mobile experience.

Mapping The Ideal Journey

With the concept model for Robert Half Direct established, to begin the prototyping process, I started by mapping the ideal journey, for both user groups. Again, I built the journey maps so both user's journey's could be viewed contemporaneously on the same map.

Mapping The Ideal Experience

The ideal user flow

Based on the ideal user journey, I designed user flows to start conceptualizing the digital experience for both user groups. I started to think about how the experience should derive parity between the physical & digital, & how we would establish parity between the mobile & desktop experiences for candidate users.

Key Problems To Solve

How might we establish a self-service workflow that enables both user groups to progress through the experience at their desired pace?

The key research insights from both user groups let us know they desire a self-service, self-paced job search experience. The RH Direct experience needs to establish a powerful way to self-manage the experience.

How might we collect the high-level data that the Data Science team needs to build a high-precision Machine Learning model, capable of high-precision AI matching?

For the Robert Half Direct experience to be both viable, and highly scalable, we need to feed the Machine Learning model with massive amounts of data from both user groups without weighing down the flow of the experience.

How might we ensure the product gracefully degrades in edge cases where there aren't high precision matches available?

AI isn't perfect; there will always be scenarios where AI doesn't produce the desired result, or is incapable of producing the desired outcome. In those circumstances, it's important the user still has the ability to engage in a powerful job search experience.

How might we ensure parity between the mobile and desktop experience, for candidate users?

Research uncovered candidate users want to utilize both a desktop and mobile experience. However, client users overwhelmingly want a desktop experience, and don't desire a mobile experience. I didn't want to waste valuable time and resources designing a mobile client experience that would have little-to-no utilization. But, I needed to make sure there was strong parity between both experiences for the candidate users.

Prototyping The Self-Service Workflow

Robert Half Direct is all about the self-service experience. Research gave me high-level insights on what users wanted out of the experience. I led my team through several iterations of what the self-service experience might look like, conducting moderated usability tests after each iteration. Below, I walk-through the final validated design.

The client user tracking dashboard

The key to achieving a quick, self-paced, self-service workflow is a comprehensive tracking dashboard. Through both research & usability testing, we found users wanted a single pane to manage the flow of the hiring process.

Drag & drop

I designed and tested several different methods to enable client users to linearly manage the hiring process. Laying-out the process vertically, required too much scrolling up and down for the user; so much so, that the scrolling made it hard for the users to maintain their cognitive footing. The solution was to lay-out the process horizontally so that users could see an overview of the end-to-end process in one view, without needing to scroll. I designed a drag-and-drop system, using large drop-zones, to streamline how Client Users move a candidate through their own process. Utilizing the drag-and-drop system eliminates the need for both scrolling, & clicking through auxiliary screens, & makes the single screen flow come to life.

One-step to make an offer

I tested both single and multi-screen process flows that enable the client user to extend an offer. Both research and testing showed me the client user valued a quick & logical method to extend the offer. Through the final usability testing, I found the shorter the offer flow was, the more likely the client user was to immediately post another position. User satisfaction at the end of the offer flow had a direct correlation to increased repeat business, so I knew I needed to make the offer process as quick as possible.

To achieve the quick flow, the offer extension process needed to be a direct extension of the drag & drop system. Through moderated usability testing, I validated the single screen, one-step, drag & drop offer. To make an offer, all the client needs to do is make the final drop into the offer drop zone. Extending an offer is contained in a single popover window. Once the client makes the decision to extend an offer, it takes less than 30 seconds to make, confirm, and send the offer to the candidate.

The candidate dashboard

Candidate Users almost completely equally used both the desktop and mobile experience, so It was important to establish strong parity between the two. In both experiences, the candidates are notified in the main dashboard when they have an interview invitation to review. The clear calls to action, coupled with clear information layout, enable the candidate to quickly review and accept an interview invitation.

A single pane for monitoring

Research told me that Candidate Users, much like Client Users, wanted to be able to monitor their progress through the interview process in a single pane, or single page. I tested multiple designs on the Candidate Users. The validated design makes use of a single pane, with all pertinent information about the selected job opportunity. I designed a segmented progress indicator to ensure Candidate Users always know where they are within the interview process.

Bringing-To-Life The Shortlist With Strategic Data Collection

The RH Direct product is driven by RH Direct AI's ability to match Candidates to Client's open positions. Research told me Client Users wanted to review a limited, high-quality shortlist of potential candidates. The surveys and user interviews gave me a range of anywhere between 5 to 20 candidates as the proper candidate range for the shortlist. I tested the range through usability testing, & ultimately arrived at the magic number of 10 candidates for the shortlist.

The real world problem evidenced during beta & pilot testing. It became clear, the AI Algorithm, driven by the existing Machine Learning Model, was really good at narrowing the shortlist down to 40-150, or so, potential candidates. In its initial state, the technology wasn't capable of producing the high-precision matching, necessary to achieve the user's ultimate goal of 10.

The match precision challenge

High quality match precision is achieved through high quality data. Robert Half Direct AI takes in massive amounts of raw, free-text, through job descriptions & résumés. This is called unlabeled, or unannotated data. AI & Machine Learning is limited in what it can do with unlabeled data. To achieve higher quality predictions, & thus, high-precision matching, the machines need labeled, or annotated data.

Deploying annotation, or data labeling teams is extremely costly, so much so, that it would collapse the viability of the Robert Half Direct product. Not only is it costly, but it isn't easily scaled.

The solution: user-labeled data

My idea was to influence users to label their own data, by hardwiring the mechanism directly into the Robert Half Direct platform itself. Designed & implemented properly, users aren't cognizant of the fact they are labeling data. This meant the method of user-labeled data needed to be quick, with very low effort required. Influencing users to label their own data is not only zero cost, but it's highly scalable, & also enables the users to consistently & systematically QA the data going into the Machine Learning Model.

To ideate the solution, I started with sketching-out the process flow above, as an initial guide.

Influencing client users to label their data

For Client Users, I built-in to the job posting experience one additional data collection prompt: Top 3 Skills. Based on the job title & job description, Robert Half Direct AI displays common/suggested skills. Alternatively, the user has the ability to write-in their own skill tags. These 3 top skills are weighted the most in the Robert Half Direct AI matching algorithm. Additionally, the user is able to designate one skill as 'Must Have', which further increases the algorithmic weight of the skill.

Collecting additional client data to QA the Machine Learning model

Seen in the screen above, the Client User is given a QA point post-interview. This helps to continuously refine the next batch of matches to populate the shortlist.

Influencing candidate users to label their data

I designed two individual prompts, to act as reminders for Candidate Users to consistently update their profiles by adding their skill tags (labeled data). In order to keep the sign-up flow quick and unobtrusive, the first prompt comes in the logged-in experience, a toast notification suggests the Candidate User update their skills. Secondarily, on occasion, as the user works through the experience, they will be met with the second prompt for user-labeled data, the blue modal suggests the user updates their skills. In both cases, the copy reinforces to the user, the effort to update their skills tags will result in better job recommendations/matches.

Collecting user-labeled candidate data

The Robert Half Direct AI suggests skill tags based on the user's résumé & profile. The Candidate User has the ability to add up to 50 skill tags to enhance match precision. In the broader sense, these skill tags are directly cross-referenced to the Client's top 3 skill tags.

Testing & validation

In both testing, & real-world application, the approach to user-labeled data consistently ensures a shortlist of 10 candidates per batch.

Designing For Graceful Degradation

AI isn't perfect; there will always be scenarios where AI doesn't produce the desired result, or is incapable of producing the desired outcome. In those circumstances, it's important the user still has the ability to engage in a powerful job search experience.

The candidate user solution: searching traditional Robert Half job listings

Solving for a degraded candidate experience was simple; we made the decision to enable candidates to search and browse traditional Robert Half job postings. After creating an account, while the candidate waits for Robert Half Direct invitations, they can engage with Robert Half's traditional service offerings by applying to a job posted by a Robert Half Recruiter (on behalf of the client in a traditional search).

The client user solution: browsing close matches

If there are no matches for a Client User to review, instead of making them wait for the next batch of high-precision matches, Robert Half Direct allows them to review close matches, whom are just outside of the given requirements.

Outcomes

Revenue
$225M
RH Direct has quickly become a $225M+ product for Robert Half.
Daily Active Users
12M
As of 2022, RH Direct serves over 12 million users per day.
Increase In Recaptured Clients
78%
RH Direct increased the client recapture rate by 78% over the previous year.
Cross-Sell Rate
84%
84% of RH Direct's new clients also engaged RH's traditional service offerings.