
15 UX Tools that make my work stand out.
A top designer like me relies on his tools to create true craftmanship in UX. Due to massive demand by the universal design community, I will let you in on my 15 top tools & tips.
It’s every kid’s fantasy to be a star. My eldest son of 10 wants to be the next Tony Hawk, the iconic skating legend of the 90’s. ‘Dream big’ we tell them, but ‘start small’ we advise. It’s something of this rashness that still keeps us dreaming big in business today. Yet we realize we can only reach the big if we respect the small. Artificial Intelligence is like dreaming big, really big actually, hoping one day it delivers on our wildest expectations. But how many iterations does it take to get it right, even in small scopes! This article is about how we addressed this challenge for NxtPeople, an upcoming tech product in HR.
Before we set out on the path of intelligence and how to design for it, it’s wise to point out that intelligence comes in many forms. I have quite a big knowledge of random facts but my wife is much smarter on a social level. In fact, she’s much smarter, period! In a similar sense, process automation is not the same as artificial intelligence. Where the first relies on structured data to provide predictable outputs, the latter feeds on algorithmic patterns and gives us back insights. That’s about as deep as I’ll go for now. If you feel like diving in, I invite you to read some clarifying guide about it. One final word on Artificial Intelligence though: AI is a loaded term that too easily hints possibilities that aren’t always accessible yet. Furthermore AI means different things to different people. For matters of safety, I’ll just frame it using the word intelligence instead.
Now I’m no preacher of evil. I won’t tread on slippery ice and declare bold predictions on where AI will take us because I just don’t know. I only have an opinion. For one, AI will not eat our creative jobs as UX designers, not yet for a long time though. I do believe that it will support us in our job however, help us rule out the more repetitive tasks and focus on designing for better usability. But it’ll take time. One inexorable law I learned as a UX Designer: if your product isn’t easy enough to use, it won’t be used. Period! Same goes for anything involving AI or any other innovative touch for that matter. A promise is one thing, but it means nothing if it doesn’t add real value.
I’m currently running a project as UX architect for NxtPeople, a leading HR company in Belgium building a SaaS product to track applicants (in what is generally understood as ATS or applicant tracking system). NxtPeople is a digital platform with a set of tools, each focused on one specific task within the ATS process and perfectly aligned with the other applications so the process flows smoothly. The power of the platform really proves itself for larger operations with many employees working in changing shifts but any single application can also be used in isolation by any type of business. Although the applications are designed to onboard the user intuitively, the nature of the ATS process requires some applications to come with a certain learning curve for any consultant using these tools. Furthermore, all tools have a slightly different usability framework.
Dealing with a ton of candidates and/or employees means there’s a huge amount of data passing through. A candidate’s personal specifics and preferences, an employer’s need for a workforce count during seasonal periods, a consultant’s inclination to work with this or that candidate… to name just but a few. It goes without saying there’s a huge win to be made by automating manual tasks and improving effectivity in the process.
If we want to start applying intelligence to our digital ecosystem however, it’s imperative we take a step back and assess how it could benefit the human process. After all, any process starts with people, not technology. To start with, who are the people using it and what data do they really need in which context? Then, which data is already available and how can that data be leveraged to reach the solution we wish to define? So let’s take a look at the most common challenges a HR recruiting organization faces.
When we think about making an application intelligent, we tend to think about complex algorithms and futuristic scenarios of sensors communicating with actuators and receptors. Yet the most obvious intelligence lies within the people using the product and its most common features we use day-to-day. So improving thereupon should be our first quick win.
Taking a look at HR, the core of what the business does is getting the right people to the right job and keeping everybody happy in the process. One frequent challenge arises when we look at what type of jobs our candidates get proposed. For all the open positions available, the actual proposition is almost always too narrow because it focuses on what the resume says (if at all) or it misses a crucial understanding of the candidate’s skillset or preferences. To make it even more dicey, it often (if not always) depends on the skills and appetite of the consultant dealing with the candidate. The result is like playing roulette, hoping the ball ends up in the right slot. Yet behind the apparent facade of a candidate’s resume, there is a whole array of possibilities, if you care to look in the right places and make the right connections. It’s not something you expect the consultant to do, let alone the candidate. Either one just wants to get the ball in the right slot as soon as possible. Everything in between is just unwelcome but seemingly inevitable friction. With the right intelligence, it shouldn’t be. An intelligent engine powering the process functions as a superfast broker by scraping all relevant data on the candidate and matching her with many more recommended jobs.
When we looked even further, data showed us that the average age of the candidates proposed is often at odds with the average age of candidates that started at any given company. When you notice there’s a gap in age (for example, the candidates starting are on average 5 years younger) you might ask yourself why the proposed candidates are older? Is it because the recruiter fixates too much on skillset or proven experience? Maybe if you loosen the requirements or bring in more relevant data to the process you could more easily match a younger person. It’s a bias AI doesn’t have to deal with.
Fortunately there’s intelligence to help our matching achieve a better rate of effectivity. We’re using an analytics engine to help us search any kind of document by mainly using text alone and letting the software interpret words and concepts to its intended meaning. Think of it as a server that processes requests you make and gives you back the data you need (both in JSON). Say a consultant is looking for the perfect job for a candidate, who wishes to drive a clark in a warehouse storing fruit. The consultant enters the terms ‘clark’, ‘warehouse’ and ‘fruit’ and the engine maps these words to the documents they occur in, using what’s called distributed inverted indices. The engine then gives back the relevant documents (such as job descriptions) for the consultant to work with. It tries to complement the way a consultant works using her natural language, either relatively directing her input to the right document in an index or conceptually understanding what she’s looking for.
On top of this intelligence, we need an interface that also understands how a consultant handles her process in a natural way. This means displaying the way she tries to match candidates and jobs in a logical manner, mimicking the natural process of a conversation. If in any physical encounter, she has a candidate before her and that conversation tends to flow into many different paths, the UI should provide her with ways to search for the data easily, within one screen. A tool like NxtMatch keeps track of every possible scenario, providing all relevant information in one screen. It’s a bit of a challenge to cram many features in one screen while keeping it user-friendly but these kinds of challenges remind me every day why I became a UX designer.
Systems that are intelligent rely on feedback loops to refine their algorithms. Feedback can come in many forms and we need to assess when we ask for it in what way. Agreed, you could ask direct feedback from the user, but let’s ask ourselves just how much feedback we need and at which touchpoint the user comes into contact with the product. Again, when you look at the data, there is already a lot you can deduce from a user’s preferences. Intelligence should be something that works for itself mainly, instead of reinforcing it with unnecessary feedback which might frustrate the user or leave him completely indifferent. I can tell Spotify which song I like, but I rarely do. I just use the platform to listen and Spotify recommends based on what I listen, which is pretty accurate most of the time. It helps when I click the heart icon but it doesn’t necessarily need to know. Why wouldn’t any other digital application, say a HR tool, work that way? After all, it’s what we’ve grown used to.
In that regard AI is about getting the biggest hit ratio, knowing full well that you’ll never get a perfect ratio. It’s a balance between how far you want people targeted in a segment and how flexible you let the system recommend other suggestions. Remember that we’re dealing with humans with changing preferences. If, for example, you notice that you get a 80% hit ratio with only 20 profiles, work on these instead of trying to perfect the other profiles as well. That’s how you’ll produce volume, which is an important factor, especially when dealing with temporary resources in HR.
Next to the various intelligence solutions we have yet discussed and implemented at NxtPeople, we also set out to see how we could introduce intelligence in more precarious tasks, like planning forecasting for example, where planners need to accomplish the task of finding the right people to fill a job, based on the employer’s needs for time and resource allocation. Planners need to do this on a daily basis and they usually plan ahead for a week or two. Today, this process is mainly done manual, with a few twists of acumen added to it.
NxtPlan is the tool within the NxtPeople family that helps consultants easily plan employees. It’s a relatively intensive job that requires a lot of human process knowledge to get the plan board right.
Getting started wasn’t easy, but the easy thing about getting started is to just get started, so we got started. Before we even set out to invite external partners we wanted to know how we could work the data ourselves and make it predict planning needs based on various datasets we have at our disposal. We knew how some clients had a history of planning a certain amount of candidates within a given timeframe and how climate types affected that need. Furthermore we know which (types of) candidates some clients prefer and in which location they prefer to look. So if you think logically, the answer to apparently challenging questions can be quite simple if you care to connect the right dots in your database and combine it with relevant external datamodels. As much as we could build the algorithms ourselves (any team with the right kind of developers can), we felt it would be wise to also look outside and see how external partners could hook their expertise into our product. The success of any platform relies on the strength of its partner network and the openness to integrations. We invited Tangent Works, an AI company that specializes in time series forecasting. TIM, which stands for Tangent Information Modeler, is a custom tool they built that uses machine learning algorithms to build a mathematical model of sample data, known as training data, to perform a task (such as planning forecasting) without using explicit instructions, relying on patterns of inference instead. These patterns are models using data at certain interval points in time, extracting data from these points from different parameters, making the tool able to give mathematical models and expressing various factors influencing manpower needs in the form of a formula.
There are several types of models, e.g., regression, optimization models, probabilistic models, etc. These are complex and suitable only for organizations dealing with large amounts of data, like NxtPeople. Examples of time series are heights of ocean tides, electricity prices and well, staffing needs.
Under the hood, the tool lets planners make use of ratio analysis to forecast staffing needs. This method helps to calculate the ratios on the basis of past data. First, it calculates the future ratios on the basis of the time series analysis/extrapolation, after making allowances for the changes in the organization, method and jobs, if any. Extrapolation is a mathematical extension of past data into the future time period. So-called moving averages and exponential smoothing can help for projections. As such, the model estimates the demand for human resources on the basis of ratios.
This makes it a perfect tool for HR planning forecasting, the process of predicting demand and supply in human resources. Or, in other words, the ability to see into the future and make educated predictions about any number of sourcing requirements an organization may have. This may include the number of employees or types of skills that are needed and available to get the job done, a requirement key to the benefit of a planning tool like NxtPlan.
In order to successfully make forecasts based on time series data, it is important to understand where the differences come from and how they can be exploited to create a fully automated modeling engine, which is what Tangent Works does.
To make matters like this tangible right into the interface, I needed to step into the realm of engineering and get a grasp of what’s under the hood, just enough to get a conceptual model of it. Which is what we’ve done at NxtPeople. To better envision where a designer’s role is ultimately heading to, you can read my article on ‘the end of UX’.
Because no smart product is smart without proper metrics, we use PowerBI as our main analytics tool. Power BI is a business intelligence tool by Microsoft that connects to the various data resources the organization has at its disposal (like excel sheets or data warehouses) and merges everything in a dashboard screen to display actionable insights. Dashboards play a crucial role of congregating key insights in one place and as such — if done right — they act straight on the behaviour of the user. It’s very much the control center of your process. At NxtPeople, we conducted a kind of layer-by-layer data architecture analysis to convey what insights to transfer to the user? The crucial question thereby is: who exactly is the user? Is she a consultant, a district manager or a payroller looking for insights in a particular client? In any case, building an analytics dashboard should be global, allowing granularity with interactions at request. Futhermore we analyzed what focus we needed to attend to. What needed the attention of the user in which context? What does he need to remember, prioritize or execute? Having mapped the right data layout, next it was important to decide how to visualize that data. Showing the client’s dropout rate of employees during a periodical interval is better conveyed with a line chart than a pie chart clearly.
The above are parameters of configuration PowerBI helps us a great deal with. However, design-wise it’s a bit of a curse implementing an iFrame of a PowerBI dashboard into an application that’s visually shaped in a very different way. The question is how much effort we should put into designing and building our own dashboards when all the data is already at hand within PowerBI. Sadly it doesn’t look very sexy or it doesn’t allow us to model the interface like a fancy Dribbble experiment. Yet, apart from those myriad Dribbble doodles that never see the light of day, almost all analysis tools look the same. For a designer, they’re just shitty to work with. But for the end user, they do the job.
Providing a user-friendly interface is not only about the information architecture, the usage of adequate patterns like autocomplete and progressive enhancement or even the correct in-context positioning of UI elements. It’s also about designing a flow that allows backend intelligence to enrich the user experience when relevant. At NxtPeople we learned a lot about how to make the process run intelligent and user-friendly. It took us years and it will take us more years, but we’re strongly dedicated to building the best possible experience for consultants and candidates in the HR process and becoming the number 1 tool in ATS.
A top designer like me relies on his tools to create true craftmanship in UX. Due to massive demand by the universal design community, I will let you in on my 15 top tools & tips.
How is it possible that for an interaction billions of people engage in daily, there are only a few modes of communicating? Why can’t the honk be designed to communicate different messages? Read the story of Angry Adam.
The most extensive guide ever written, featuring 25 fundamental laws!
Designing a website isn’t as simple as you always thought. There’s a lot to cover. Fortunately, Pimp has written the most extensive guide ever written, with 25 fundamental laws he invented. A must read for anyone with the intention to ever design a website, or look at one.