Over the years, our firm has had many discussions with employers on the eve of a new talent analytics project. Often, it is the firm’s first deep-dive look at employee data. Sometimes we act as a strategic sounding board, and sometimes we can help them move directly forward into predictive analytics. It is always interesting.
Typically we see one of two analytics approaches and it is worth considering why you might want to begin with one over another. And though we are speaking about employee analytics, the same logic applies to any kind of business analytics.
Fishing Expedition – Let’s Analyse Human Resources Data
Basic business awareness leads most firms to the conclusion that (a) employees are their largest expense, and (b) employee behaviour is a significant driver of corporate success or failure. Other human behaviour domains (customer analytics, patient analytics, voter analytics) continue to demonstrate clear ROI from using predictive analytics, and it is natural for employers to ask how the practice can be applied to employees and job candidates for similar ROI.
This potential often results in a decree to “look for something interesting in the HR data.” We often speak with the professionals who have been given this vague task: to explore HR’s system of record — hires, terminations, payroll, home information, demographics, performance reviews. Sometimes, the agenda can also include unstructured data — internal email, chats, and social media in or outside of the company.
Often the effort is scoped to include a wide swath of employees in diverse roles: all managers, all professional roles, high and low potentials. The objective is almost always a vague “something interesting” rather than any specific objective or business outcome.
Inevitably, with enough data, enough time, and an expensive visualisation package, something will be found. Let it be noted that you can almost always find something of interest in any dataset. Whether it is actionable or not is another story. For example, analysis of HR data may find that:
- Employees from certain schools sell more;
- Employees with short commutes get higher raises;
- Employees from expensive zip codes have higher influence scores;
- Different office locations work longer hours;
- Younger users use corporate chat more;
- Longer-tenured male workers shop online from work.
The problem is, of course, that these types of projects are fishing in a vast sea of input variables, with no known business outcomes. The problem is that the HR data is essentially metadata of the employee’s life at a company, absent the results of their work.
The interesting, high value data — the work, the performance — is generated on the job, in their department after they’ve been hired and on-boarded. HR doesn’t typically track this data; it is tracked by the line of business.
The real business outcomes — the KPIs, the reasons the business hired the employee — are lived out and documented at each line of business for each role: in the sales department, or in the call centre, or in the bank branches, or in software development.
The problem with the “fishing expedition” type of analytics effort is that it wasn’t framed correctly to begin with. An effective talent analytics effort simply cannot span all lines of business, across every single job description. It needs to be focused and framed.
Without measurement of actual business performance from the line of business itself, the analysis is fated to discover trivial relationships between input variables. Even with performance data, such an effort can’t hope to understand and normalize the subtleties of Key Performance Indicators (KPIs) in each role, line of business and location.
Our experience with these fishing expeditions is that the results are unlikely to yield patterns that are actionable. If you want greater sales, do you limit hiring to a small number of schools? No — this limits your sourcing options. If you want higher influence scores, do you simply start hiring from more expensive zip codes? No. And is there any proof yet that influence is tied to any kind of performance, in this role, in this company? No.
The result of a fishing expedition is descriptive analytics — glorified reporting — not predictive analytics. The only thing that is predicable is that employees may be less than comfortable finding that their employer has been reading their email and social media profiles. It seems obvious, but only by considering outcomes in the analysis can one predict outcomes.
However beautiful interactive dashboards and visualisations of these patterns may be, fishing expedition projects are unlikely to receive follow-up executive sponsorship. And, sadly, the reputation of talent analytics projects at the firm becomes trivialised or suspect.
The Targeted Win — Let’s Find Business Pain and Solve It
A more narrowly focused “business win” that can be implemented to save hundreds of thousands of dollars a year will win over a vague, ambitious effort just about every time. These more focused wins are indicative of millions that can be saved when implemented at full scale.
Targeted wins can be led by either HR or line-of-business management (or together). They identify a business “pain” in a particular role, such as:
- Sales reps that aren’t making their quota;
- Insurance claims adjusters who make too many errors;
- Bank tellers who quit in less than six months;
- Call centre agents who can’t pass training.
The successful talent analyst goes beyond HR’s carefully manicured “system of record” and merges actual performance data from the line of business. Business relevance is here. Actual outcomes are being tracked in the sales office, where the sales are being made.
The data are often quite structured. Even so, performance and KPIs can be confusing and contradictory. It can take effort to untangle signals to devise simple data experiments that deliver results, measured in line-of-business terms (not HR terms). It takes a data scientist with keen control over the data to tease patterns apart to find the truth and present it in a way that can be understood by the business.
As in all analytics efforts, a targeted effort with a targeted group is needed for the effort to yield strategic results — not just interesting results.
Cost Measurements are Essential
In marketing analytics, the costs of acquiring and maintaining customers are a vital component of a predictive decision. Likewise for employees, the costs of acquiring, on-boarding, and maintaining staff in a role matter.
We don’t understand why every line of business or HR department doesn’t already know and publish these cost curves, at least for high-volume roles. Most managers know that attrition is expensive, but still only count employee costs in terms of compensation and perhaps the cost of the job advertisement.
To understand attrition cost, one must factor in training, productivity, and break-even as outlined in the CFO’s Guide to Employee Attrition. The cost information provides a vital link to place a value on employee turnover, prioritise efforts, streamline improvement, and to tune predictive models. Managers who care about performance or attrition need this information.
This cost model is an excellent high-value deliverable that lays the ground for smarter operational decisions as well as bigger savings from predictive modeling.
- HR data alone is rarely sufficient for strategic, actionable insight;
- Line-of-business leaders hold critical performance data in their systems;
- Analyses that tie to actual business problems lead to sponsorship, funding of more analytics projects and strategic business success;
- Purely descriptive analytics of secondary issues can’t provide actionable insights, even with pretty dashboards and visualisations;
- Cost analysis should be fundamental to any analytics effort (and in fact every business effort).
Analysts, HR and line-of-business executives need to obsess over finding incremental fact based solutions to business problems, rather than grandiose patterns seen on a dashboard. HR and analytics teams need to earn their way into bigger budgets and bigger toys. Employee analytics projects should focus on a targeted win. Cost modeling is an example of a targeted win and is the ideal way to get started.