What’s your approach to HR analytics projects? Are you prepared with a plan?
HR analytics involves some complex interactions between data scientists and HR professionals. In a business environment, a plan helps to manage your time commitment and the necessary management control.
It’s also important to be able to communicate clearly and deliver exactly what the client or colleague is looking for, in a format that is useful to them. You can spend a lot of time going down data rabbit holes, only to find it’s not what they wanted!
Here is my take on tackling HR analytics projects with a plan:
HR analytics – a vast subject
In my experience, it’s only beginners who start a project without a plan. The most successful HR analysts I know all commit a plan to paper (or an electronic means) before embarking on a project. A robust plan helps you to focus your efforts and maximise your time.
The field of HR analytics is such a vast subject, you can go down many potential roads. From my earliest ventures in HR analytics, I’ve advocated choosing to concentrate on a clear subject, preferably one you are very interested in! Look for something you can analyse or improve – often the “low hanging fruits” are the best for getting your HR analytics expertise going.
You might start with an area of concern that the business has that relates to a core goal. This way, your data analysis can be focused on anything that impacts that key metric.
Establish clear communication
Any successful plan needs to have clear two-way communication and a documented strategy. This is especially true when you are undertaking analytical work for a colleague or client.
Matt Dancho tells the story of working all weekend on a short-order project for one of his first HR analytics clients:
“We received the data on Friday, and were to present findings on Monday so the executives could integrate into a presentation for Wednesday of the following week. I contracted an HR Specialist out of Houston. He and I worked non-stop from Friday through to Monday analyzing their data, extracting insights, and creating a detailed report that documented our entire analysis.
We actually created a pretty cool algorithm that detected 13 or so employees that were not currently being targeted for the “executive track” but should have been based on their data.
But this report and algorithm was not what was wanted. Our contacts were expecting something different than what we interpreted from the conversations (still I do not know what they wanted).”
This highlights the importance of starting out with a clear goal in mind. How will you know that you’re delivering what is really needed? It’s also important to consider how you need to present the data, so that it is in a format that is understandable and useful to others.
Business Science Problem Framework
Following that unsuccessful project, Dancho took what he could learn from the experience and saw the importance of having a good plan. Any sort of data science needs to have defined parameters and expectations if you want to truly get value from it.
Dancho developed a “Business Science Problem Framework” (BSPF), which outlines a systematic process for an HR analytics project, including ensuring that expectations are clearly set with any stakeholders.
You can see an outline of the framework in the image below:
As you can see, the BSPF provides a ready-made plan for any sort of data science project. It also shows that a major research project needs longer than a weekend to get truly meaningful results.
Importantly, if you’re taking on a project for clients, colleagues or your boss, the framework helps to clearly define what is required, along with expectations of the project steps and timeline. It helps to lay out the scope at the beginning of the project and ensure that any ROI is aligned with what the project sponsor or person who requested it wanted.
As Dancho points out, this framework also helps to bring client questions up at the front-end of the project, while instilling a sense of confidence that you know what you are doing! In his own case, he says client satisfaction rocketed after they adopted this model.
You might not want to follow this exact approach, but the point is a structured plan will give you better results. It helps you to set parameters for yourself and communicate clearly with others. No one wants to be in the situation where you’ve put in a whole lot of work, but haven’t delivered whatever it was someone else was expecting!
Research your chosen area
With a plan structure in place, you need to read and gather as much information as possible on your chosen metric. Given that HR data metrics are often interconnected with something else, this isn’t always the easiest task. I try to map out any metrics that I think are related and bear those in mind when analysing results.
If you’re a beginner in HR analytics, you might find quite quickly that there are some hard boundaries that cannot be easily overcome. I would not call it a failure if you do discover some hard limits. There are various reasons that you can’t use some forms of data, such as data protection laws or ethical issues that mark data as best left alone. I would call that “positive experience acquisition.” As you gather more of this sort of knowledge, your data projects become better informed.
Give appropriate time to analysis
A key mistake that many people make is to jump too quickly to conclusions about data. It’s easy to make assumptions or to quickly think that data backs up a preconceived notion you had, but you are in danger of missing something entirely different.
For this reason, I ensure that I devote a good portion of time to a root cause analysis. You can never assume that you got to the bottom of something the first time.
The main objective here is to move away from a state of mediocrity in how HR functions are maintained and decisions are made. No more ignoring the existence of data or showing a preference for intuition over solid metrics. Yes, intuition has its place in the initial phase of research (it can be helpful for devising a hypothesis then narrowing down the data you need), but the aim is to have data-driven final decisions.
That being said, you need to be prepared that it may not be a smooth ride. Real life HR data is always many times messier than you may have imagined. Often data is incomplete due to lack of compliance or perhaps even a lack of cooperation from staff. This is an area to tread carefully – maybe staff were not convinced enough of where the advantages were for them in sharing data.
Should you give up and use only intuition just because of that? Certainly not, in my view. Should a CEO be unable to make conclusions, because only 70% answered to an internal survey? Certainly not. Conclusions are possible and are even recommended, in my view.
Final thoughts
A recent experience of mine highlights how persistent you have to be in the HR analytics field. I went to a very enjoyable datadive (a sort of hackathon) organised by a charity called Datakind. The datadive went on for two days. After signing a very detailed one-page non-disclosure agreement I was given some real-life business data.
Having the option to analyse data from four different charities, I decided to go with real business data from Foodshare, which is a charity redistributing excess food supplies to those in need. Although the dataset had already been cleaned by my data ambassadors it was still a bit messy and still required further cleaning to be able to extract some sensible information. Although I dream about Artificial Intelligence and Machine Learning algorithms, I still tend to be a data cleaner most of the time.
I spent a lot of time thinking about what data science tool I should apply. I also spent a lot of time considering which aspect of the data I should analyse first, given that it was all a bit messy and there were many options to go for. Real data are just like that, but that’s not an excuse to give in to mediocrity. The point is I started with a plan and worked my way through the data from there. It was a very enjoyable weekend doing hard work, even if it was slightly stressful.