• Chi Tran

Why Most People Analytics Initiatives Fail

Updated: Nov 17



People Analytics is a general term for various measures that improve employee happiness by making better personnel decisions by utilizing all data related to employees and thereby promoting the improvement of management effects.


According to PwC survey results as of 2017, 79% of HR professionals said they were interested in People Analytics, but still thought it was "important but not urgent" and "did not know where to start". I often hear people say, "I don't know if it's good," or "I'm trying various things, but I'm struggling."


Therefore, in this series, I would like to discuss the following two points in the first and second parts in promoting People Analytics within the organization.


Part 1: Why hasn't People Analytics worked (so far)?

Part 2: Points to overcome technical and awareness issues


In addition, I address a fundamental issue, "How to make People Analytics a useful weapon for management and the field" in areas such as business development, corporate planning, and marketing at companies drawing from my previous experiences at BCG (Boston Consulting Group), Recruit, and Google. We will also consider reflections and learning when using analytics in areas other than human resources in this article.

Why did People Analytics not work? Cases caused by awareness issues


Analyst teams and human resources departments tend to be regarded as "cost centers", and management is often reluctant to invest in systems and increase personnel. Regarding People Analytics, we often hear people say, "I'm interested, but I don't need to work on it yet" and "(in case of emergency) existing methods and tools can be used." However, I think we should change our awareness of the problem by looking at the following three points.


a. See People Analytics as a risk-handling tool


This is not limited to the human resources area, but we all know of numerous cases in which “after a problem occurs or becomes apparent, a response policy is formulated and improvement measures are implemented”.


However, for example, even if an increase in the turnover rate becomes apparent and a project team is formed to investigate the cause and implement countermeasures, it will take 6 to 9 months just to investigate the cause as described above if the analysis base is not prepared in advance. In the meantime, the wave of turnover will not stop. Shouldn't we build a lighthouse and breakwater in advance in preparation for the waves that inevitably come as the organization expands and transforms?


It is often said that the management resources of a company are concentrated in "people, goods, money (data)". "Financial statements" can be cited as a "tool" that captures the movement and soundness of "money" in real time, evaluate it with objective indicators, and gives suggestions to management policies such as future financial strategies. With regard to "things," CRM (customer relationship management) and SFA (sales support) exist as "tools" that capture customer movements and soundness (signs of separation, etc.). Moreover, these are not proprietary patents for the finance manager or sales manager, but now serve as a catalyst that promotes management and the field to become a monolith and make decisions from the same perspective and values. Similarly, for humans, there should be a "tool" that universally captures the movement and soundness of organizations and employees by personnel, management, and workplaces (* 1). It may be a little exaggerated to say so, but People Analytics should be regarded as "financial statements for human resources".


I don't think many companies exist now that correct their PL (income statement) for the first time after a deficit is suspected. Similarly, and especially in companies that advocate the utilization of human resources, it is considered that it is well worth investing in advance as "insurance" rather than sunk cost so that organizational issues such as turnover will not become apparent and People Analytics will not be started at last.


(* 1) As a side, the churn (= churn rate) of goods (products) is frequently raised on the agenda at management meetings and investor briefings, but the churn (= turnover rate) of humans is not raised as frequently. The situation will change if it becomes easier to acquire and visualize indicators that objectively and universally evaluate human movements and conditions.


b. Surpass existing HR tools


The value provided by tools and services called "HR Tech" can be broadly divided into operational efficiency (process management) and decision-making efficiency/optimization (data management). In addition, the scope of coverage may include part or all of the employee life cycle, such as recruitment and human resource assessment (talent assessment). Among them, People Analytics is positioned as "supporting the efficiency of decision making throughout the employee life cycle."


What is currently dominating the HR tech market will be "partial optimization tools" that digitally streamline some of the work of the employee life cycle. It is thought that this is because the results of reducing unnecessary man-hours and improving productivity are visualized relatively quickly. As an extension of such efforts, we often hear people say, "I will try People Analytics with the analysis function of existing tools for the time being", but it is possible to visualize the indicators that the organization should really grasp with those tools alone. It should be carefully evaluated whether it is highly reproducible and contributes to improving the quality of decision making.


c. Forcing analysts to struggle alone


Recently, there are some companies that have "assigned (appointed) several People Analytics analysts for the time being". I think it is a very meaningful first step to establish a People Analytics-dedicated role/position within the organization. However, in order to promote People Analytics in the true sense, its scope (business) should not be underestimated as mere analysis.


This isn’t limited to People Analytics, and the value of data analysis can be divided into three categories:


(1) Description (understanding/visualization of the current situation),

(2) Prediction (suggestion/prediction for the future based on the grasped data),

(3) Prescription (concrete action).


Especially in (3), whether the final beneficiary, that is, whether it was useful for the management or the field in the company, and whether the quality of decision-making has improved (= existing inefficient work and irrational practices are corrected to more valuable

ones.) is the watershed of success or failure of data analysis.


To that end, it is necessary to deal with tasks that cannot be covered by desk analysis work alone, such as working with management and discovering on-site needs and insights. It is very difficult for several analysts to take on all the responsibilities (1) to (3).


Why did People Analytics not work? Cases caused by technical issues


Compared to other data groups owned by companies such as POS (point of sale), CRM (customer relationship management), SFA (sales activity), finance, etc., there are multiple unique challenging in analyzing and utilizing employee data.


a. Input data is particularly dirty


Compared to other areas, employee data has a weaker sense of purpose to "analyze later and utilize it for decision making", and the data input rules are different and there are many missing values. Among them, data related to recruitment has various recruitment sources and the input timing tends to be irregular (the announcement date and the effective date do not match the system input date, etc.), so data cleansing (duplicates, errors, correction of notation fluctuation etc.) is particularly difficult. A good example is data representing "Linkedin", can be represented in various notations such as "Linkedin", "Linkedin", "LinkedIn", "Linked in", and "LI" exist at the same time.


b. Not centralized


Employee data covers a wide range of areas such as recruitment, training, internal transfers, attendance, sales performance, evaluation, communication, engagement, salary increases, and turnover, but in many cases, different departments manage them separately. Connecting these factors into one will shape the company life cycle of employees. Since each data has inherent continuity and causal relationships, there are many suggestions and findings that can be highlighted by connecting them (* 2).


For example, it has been confirmed in some cases that the experience and evaluation of candidates at the time of hiring have a certain correlation not only with the acceptance rate of job offers but also with the performance and retention rate after joining the company. To achieve a consistent analysis, candidates at the time of hiring and employees after joining the company must be “identified”. For that purpose, the "candidate ID" of the recruitment data and the "employee ID" of the employee data must be linked in the business flow. However, in a situation where data management and utilization status are siloed, even such seemingly simple business design and operation become extremely difficult.


(* 2) In the world of digital marketing, centralized management of customer contact points has become easier with the evolution of data acquisition and analysis technology, and from about 5 to 6 years ago, the degree of contribution by lifetime value (LTV) and measures. There is an active movement to optimize the overall budget and activities through analysis and calculation. There are many frameworks and methods that PA can also apply from digital marketing.


c. Complicated privacy design


Employee data usually belongs to the company when the employment contract is signed, but it must be handled very carefully. For example, most companies in the Human Resources department limit the members who can access salary and evaluation data according to department or job title. In addition, when comparing and analyzing the "average salary" for each department, it is necessary to consider the risk that the actual value of individual salary can be almost identified by back-calculation as a result of various filtering. Masking is a method that does not display the calculated value when the population falls below a certain number, and it is necessary to design to avoid risks by such means.


Furthermore, similar to privacy design, "hierarchy design" in a general BI tool (i.e. business intelligence, to collect and analyze a large amount of data accumulated in a company to help quick decision making.) is extremely difficult.


For example, the only data population that is meaningful to a division manager A is the data of the members linked to the division manager A, so if the data of the members of the B division manager is mixed, its usefulness will be diminished. However, since organizational change events such as reassignment, announcements of concurrent posts, consolidation of departments, etc. occur frequently every quarter to half a year, it takes a lot of effort to accurately grasp and analyze the hierarchical structure of the organization in real-time. In fact, when an organizational change of hundreds of people is made once every six months, there are some cases where it takes more than a few weeks just to put a new organizational chart into Excel.


If you do not deal with these risks and scrape off all of them, the number of data sets that can be safely used for analysis will decrease, and the true value of People Analytics, which is to "connect points with lines and obtain a bird's-eye view," cannot be demonstrated.


Technologies and tools that solve technical issues such as a-c are gradually emerging, and the number of data scientists and analysts who specialize in People Analytics is also gradually increasing (* 3). I hear that it will take 6 to 9 months at present for several People Analytics analysts to develop the People Analytics analysis infrastructure and system, but in the future, the technical hurdles will be lowered, and it will be realized with less time and resources.


(* 3) Harvard Business Review described that "data scientists will be the sexiest profession in the 21st century" in October 2012, but the demand for data scientists in the labor market is overwhelmingly predominant in the marketing domain. I have the impression that it has not yet fully penetrated the HR area. Therefore, it seems necessary to further promote the unique appeal of employee data. A future article will look at the appeal of EX (employee experience: HR area), which is not found in CX (customer experience: marketing area).