Start with a Question or Hypothesis

There is an insidious problem that pervades many attempts to analyze business progress. Somewhat counterintuitively, this problem grows as the availability of data increases. The impact is dramatic. Analysis stalls and initiatives can fail because the insight is not there. The dynamic is the draw to dive straight into the data and begin the analysis. Large Language Models like Claude and ChatGPT make doing so even easier than in the recent past, but these attempts will rarely succeed. Their failure comes from a lack of direction or purpose in the analysis. To remedy this lack, those preparing to analyze data should start with a question or a hypothesis.  Having that question or hypothesis provides the goal for the analysis. Going straight to data analysis without a question or hypothesis is like a driver jumping into a car without a destination. There will be movement, but to what end?

The importance of well-formed questions in measurement isn't unique to change management—it's a fundamental principle of scientific inquiry. Research on organizational change measurement emphasizes that metrics must align with clear theoretical constructs before data collection begins. In change management specifically, practitioners must identify which dimensions of change actually matter: readiness, adoption behaviors, or sustained usage. An excellent foundational paper on this is Armenakis, A. A., & Bedeian, A. G. (1999). Organizational change: A review of theory and research in the 1990s. Journal of Management, 25(3), 293-315. Without this clarity, analysis becomes what researchers call measurement without theory, meaning analysis that describes but does nt explain.

How to form a question or a hypothesis

Although forming a question or hypothesis is important, the quality matters as well. A question needs sufficient specificity, meaning it requires specific data and pointed analysis to answer. It cannot be a general question. “Is the team ready for the change?” is far too general, but “Have the machine operators demonstrated sufficient interest and expertise to operate the new manufacturing process successfully?” gets at something that can be measured and analyzed. Similarly,  a well-formed hypothesis, which means it’s falsifiable, sets up the analysis to answer the hypothesis (technically, we reject or fail to reject the hypothesis). This bit of rigor has a dramatic impact on the value of the analysis because it directs what analysis should be done. 

A question is specific for data analysis when it clearly defines what needs to be measured, for whom, and over what time or context, in a way that data can realistically answer.

To make the question specific, do the following:

1. Focus on one issue. that targets a single problem or a tightly related idea, rather than “How is the business doing?” which is too broad. 
2. Clearly define the population. Stating who or what the subject of the analysis is eliminates the parts of the population not under consideration.   “New customers in the Northeast region” instead of “customers” is an example. 
3. Use measurable concepts that are time-bound.  Defining the range of time and using terms you can calculate from the data (e.g., “time spent in the test environment in the last month, or courses completed since January”, not “engagement” without definition are better formed.
4. Action-oriented drives decision-making. Not yes/no, not trivial, the question should invite analysis, comparison, or explanation, not just a simple fact lookup. The answer can inform a decision, policy, or next step (e.g., “Which marketing channel drove the most qualified leads last month?”). 

Well-formed hypotheses will benefit from all these, but should also be formed such that it is clear when they are false.  Whether a question or a hypothesis, a good way to check their overall usefulness is to conduct a Premortem. Imagine all of the ways a change may fail. Answer questions that help to understand how to prevent those different possibilities, and the analysis will deliver a result that will facilitate decision-making to increase the likelihood that the change will be successful.

Useful versus Interesting

After ensuring that the question or hypothesis is well-formed, the next step is to make sure it will guide toward useful and not just interesting data. Many avenues of inquiry will produce interesting data. Afterall, humans, by nature, are creatures of curiosity. We like to know. But curiosity does not always lead to useful. It often ends at interesting, which may be satisfying intellectually but is not helpful unless it can be acted upon.

The test for interesting versus useful is conceptually straightforward. Simply answer this question: “If the question or hypothesis is answered, can you then make a decision?” If you can answer “yes”,  then it is useful.  Otherwise, the answer to the question or hypothesis may be merely interesting.

This distinction reflects what organizational scholars call predictive versus confirmatory measurement. Predictive measures, such as readiness assessments or early adoption behaviors, enable course correction during implementation. Confirmatory measures, such as final adoption rates or return on investment, validate what already occurred but arrive too late to influence outcomes. This description comes from another illuminating paper that is a literature review of change analytics. Useful questions generate predictive indicators that support real-time decision-making. 

Setting up the Data-Driven Change Process

A well-formed question or hypothesis sets up the data-driven change management process to be more likely to succeed. Each of the four steps in the process benefits from doing this.

1. Choose the Data: You know what data to gather because you know the variables needed to answer the question or hypothesis.
2. Collect the Data: You onlyspend time and effort on what you need, thereby saving resources for all the other work that must be done.
3. Analyze the Data: You know what analysis to perform because you know what you're answering
4. Present the Data: The decision-making process is sharpened because it proceeds directly from the answer to the question or hypothesis.

These four steps of the data-driven change management process are powerful but only valuable when directed toward answering a useful question or hypothesis.

A Gesture of Respect

Large Language Models (LLMs) make it even more tempting to skip the formation of a question or a hypothesis because they can process data so fast.  Not only is this a mistake technically (the analytical result is worse), but also a mistake philosophically.  One of our Core Four Philosophies is: 

“Leading change intentionally is simply a gesture of respect.”

That includes the change practitioner doing the data-driven change management process. Ensuring that you have a question or hypothesis is an expression of respect for the value of the data-driven change management process. Forming the question or hypothesis may be slowing down (at least at the beginning), but the process will go better, if not also faster, in the end.

Broadly, we should think of AI as an engine or accelerator, not a self-contained analysis tool. AI-driven analysis without a question or hypothesis is just an "autonomous car without a destination". It is n better than a driver hoping in a car without a destination. AI just gets you nowhere with less work. The role of the human in this process is to do the thinking and questioning. The AI’s job is to do the calculating.

Research across six decades of organizational change studies reveals a consistent pattern: change initiatives fail not because organizations lack data, but because they lack clarity about what the data should reveal. Success comes not from measuring more, but from measuring better—from starting with a clear question rooted in theoretical understanding of how change happens. That question-first approach is what separates analysis that drives decisions from analysis that simply generates reports. What's the question driving your next change analysis?