Businesses have long been effective reporting on what has already happened, but much less effective when it comes to reporting data to better predict what is likely to happen next.
Employee engagement surveys have long been a part of the phenomenon. What can your organization do to help this important data source become predictive? The real answer is, it takes time (at least two survey cycles) – and if you follow these steps below you’ll be on the path to success.
1. Start with a quality survey
Your instrument must gather good data. There is not a “magic bullet” rules for a perfect engagement survey. However, beginning with the end in mind we would want survey results that are 1) representative and full (high completion rates among our population) and 2) accurate data (from well-designed survey items). In order to achieve these results, there are a few tenets to focus on.
· Keep your survey short – target a completion time of 7 minutes or less to increase completion rates
· Limit demographic questions to provide a veil of anonymity (even in the most positive corporate cultures, engagement surveys are often met with a degree of suspicion)
· Ensure each survey question asks for a response about a single defined item (avoid questions with ands, buts, or ors – language that suggests multiple concepts in a single question and often pollutes responses as it is unclear what is being rated)
· Use concrete language to limit interpretation
· Provide coverage across the range of potential engagement areas (e.g. trust, vision/strategy, career development, compensation, organizational tools and processes, quality of supervision, challenging work, etc.)
2. Define reporting groups properly
Large rater groups (e.g. departments of several hundred) are often so large that their true components will have little in common regarding sentiment. (Water and latex paint are both liquids, but mixed together a result is produced that loses the properties that either had individually.)
Rater groups that are too small will have high volatility and are more likely to pierce the veil of anonymity.
Ideally, the smallest reporting unit should be 15 responses minimum, and these units should have common ties (department, leadership, location) that link them together. On the larger side, the lowest survey reporting level should never have more than 200 responses. Engagement (and engagement interventions) happens on the ‘local’ level versus the corporate level, and reporting levels should reflect that.
3. Link survey results to business outcomes (focus on deltas (changes over time), not absolute values)
Select several (five or less) key performance indicators linked retention or engagement/performance that you plan to track. (Side note that retention and engagement are related items, but not interchangeable. Retained people continue to be employed (whether or not they are engaged) and engaged people generally perform at a higher level than their peers. Yes, engaged people are more likely to stay, but only somewhat so.)
Until you have a second data set from a second cycle, you are unable to establish any sort of predictive model. This is not a fast process to do correctly. Rather than taking an engagement snapshot (one time) and comparing it to a historical datapoint (trailing twelve months) – compare year-over-year change in engagement survey items with your year-over-year change in KPI data, and use this as the basis for your factor and regression analysis.
It can be difficult to think of a year needing to pass before the data is as actionable as we’d like, but in the interim you do have the stopgap ability to compare population groups and their business results for your year one results.
That said, you need to keep in mind:
- Certain business results represent data that includes individuals who are no longer part of the survey group (e.g. looking at turnover data over the last twelve months, those who turned over are not included in the survey sample as they are no longer with the company.) So a linkage of a current snapshot to past data is not not valid.
- High is not always high and low is not always low. Certain survey items will tend towards higher results (e.g. do I trust my coworkers) vs. others (am I compensated well for my work). Simply looking at correlations on one year of data will yield erroneous results.
- Correlation is not causation. Just because two variables move together does not mean there is a cause and effect relationship between them.
- Unobserved factors (questions not asked) may actually be the real drivers, but since they were not surveyed they cannot be seen.
- Completion rates are key to limit the impact of margin of error.
With patience and a solid approach, you can begin using your employee engagement survey data as a predictive instrument as soon as your second cycle.