#10: Pay attention! It's all connected.

Alright, no bulls**t. I will give it to you straight. Something to quote in your next talk or pitch to the c-level.

Your digital HR processes are creating the data you need to integrate, to provide the learning material needed by AI models to predict personalized results. There is no magic pill for it. It's all connected.

Within that statement, all problems of innovation-seeking HR departments become obvious. Because, we are, again, getting ahead of ourselves. I think it's generally fair to say, that, before we ever perfected our digital HR processes, we jumped to "being data driven". And before we perfected our HR data, we now jump to being "AI augmented".

It is this hastiness, that kills our reputation at the c-level. We keep saying, we want a seat at the table, but all we bring is a children's high-chair.

You don't think so?

Then, take the time and answer these questions:

Q1: On a scale from 1 to 10, how well-defined are your digital HR processes (10 = absolutely perfect)?

The following characteristics of your digital HR processes would mean a perfect score here.

First, all of your processes are as short as legally needed, and as long as the end user wants them to be. Nobody needs a 10-click-process just to book paid time off.

Second, all processes need to be accessible for all people. This means that a blind person should be able to finalise the process with little to no more effort than a fully able-bodied person.

Third, all of your processes should be unambiguous. Your HR business partners cannot be left in doubt where to enter the cost center information or the preferred name of the candidate. In the best of all cases, the processes are intuitive and hardly need any explanation.

Fourth, all HR people know all the processes they need to absolutely rock their daily work. Thus, you need tutorials, trainings, in-process mouseovers, and automated checks for correctness. Why does my bank know that I did not input a possible IBAN code, while your digital HR process still takes the wrong cost center structure, if entered?

Fifth, your digital HR processes are fully integrated. If you start anywhere in your tool landscape, you can jump across all tools needed to complete the respective process. Yes, without media breaks.

Q2: On a scale from 1 to 10, how well-integrated are the data streaming out of your HR processes across all HR tools (10 = absolutely perfect)?

There are a couple of characteristics here also, that would mean you can score a 10. If the following things are complete news to you, please take some time to research "extract -transform-load" procedures, first. The next few paragraphs are going to be a bit more technical than usual, but they're with it. Bare with me.

First, based on your digital HR processes, you collect the data created by the users within these processes ("EXTRACT" phase of ETL procedures). A person applies to one of your jobs, and a timestamp is created with all the data entered at the time. From there all data created by the recruiter, the hiring team, the candidate, and the automated tools (e.g., background checks) are collected and integrated in a centralised HR database (or data lake house, if you like fancy new words).

Second, external data (e.g., labour market data) is also integrated into the same data base. These data are very important to give context to your analytical models and AI-based prediction machines. When a hiring process is considered "fast", this may mean very different things depending on the target group you're trying to recruit. Labour market data will give this context, as you can qualify your KPIs against information like the number of graduates in certain majors or job groups.

Third, you data is transformed (automatically) so that it can be readily used for training AI models or applied statistical models developed by your people analytics team (aka "TRANSFORM" phase of ETL procedures). Yes, you guessed it right, data are usually not perfectly clean, when they are extracted from their sources. Also, analytical models or machine leaning algorithms may need the data to be in a specific format. Beyond that, AI models are usually trained, tested, and validated on different randomized sub-selections of the data. So, you may need to create so called "training sets", "test sets", and "validation sets" from all basic data sets in your database.

Fourth, your transformed data is accessible by your people analytics experts (aka "LOAD" phase of ETL procedures). The data is useless, if you're not analyzing it. And you will not be analyzing to the fullest, if your best analysts are not accessing the data. This also means that you have to have full documentation on the general characteristics of the data. They may even be sectioned by topics into different "data marts" or "data tables", which then can be connected for analytical purposes.

Fifth, all of the above happens in a compliant and secure way. This means you're not wildly integrating every data point, but only the data needed for your models to work. Also, it means that you either anonymise the data, or you have been given consent to use them by the people, whose data you integrate. This also means that the data base is secure and well-protected from malicious actors. Visit your nearest IT security and legal expert for details on this matter.

Clean the processes, bring the data, train AI

I know. You may say: "Dan, AI models come pre-trained, and deliver results for us.", and you'd be correct. Most AI-based tools have been trained with vast amounts of data, tested with another vast amount of data, and (hopefully) have been validated with new data coming in every day.

You can use these models to predict your best match in recruiting, your best future leader, the people with high exit likelihood, you star-performers, and many other things.

But the important assumption is, that the data used during the training and testing of the AI-models used within your tools, are representative of your employee population.

If there's doubt about this assumption, you're predicting something for sure. But you're not predicting something useful for your context and company-specific setting.

The good news?

It's not too late, because almost all of the HR departments around the globe did not pay attention to this connectedness of processes, data, and AI tools. That's why they are currently underwhelmed by the results they are seeing. Ask your peers. They will likely proof my point.

So, you can still get ahead of the competition. By simply starting tomorrow.

Be the data you want machines to learn from.

Peace out, Daniel

Liked this piece? Keep on separating fiction from facts with my German newsletter: HRDL News or my English Newsletter: Humanly Analytic .

Over 6600 people already follow these newsletters, and that humbles me. Thank you for your support.

This article reflects my personal views only and is not necessarily the view of the companies I am associated with.