Synthesis is integrating several pieces of research and evidence sources to form one insight or set of explanations. Researchers have been doing synthesis for a long time. It is called “literature review” in academia and, in some businesses, it’s called “joining the dots”.

With synthesis, we combine different evidence points. Often, these are different metrics, qualitative and quantitative, and from different paradigms. Joining the dots is not necessarily an easy thing to do because different sources can seem incomparable, especially if the sources were not designed to work together.

Synthesis is critical for complex business challenges

You will need to use synthesis when you are looking at complex business problems, e.g. why are sales declining or increasing? Why is NPS declining or improving? Why is conversion decreasing? Will people adopt this new feature? After a while, you might start to use synthesis all the time. It’s a question of how much to use synthesis – not whether to use it or not. You might be familiar with people saying, “let’s start with what we already know”. That is a signal there are sources you need to review!

Why is synthesis important?

The amount of information in your company and category is expanding rapidly In the banking business/category, there are approximately three times more sources than there were 10 years ago. There are more choices of syndicate sources and more complexity in different internal customer feedback programs (relationship, episodic, and interaction), social media, and app sources.

As a researcher, your value is determined by your ability to answer questions and solve problems. However, you are dealing with information asymmetry. The more senior your stakeholders are, the more information they have. In contrast, a researcher typically focuses on a specific part of the business and likely has limited data sources. Oftentimes, the stakeholders will know more than you! Your observations and recommendations will be incomplete if you don’t consider everything your firm collectively knows.

Synthesis overcomes another challenge, one caused by the fact that different stakeholders often have different preferences for sources of evidence. Traditionally, one looks to achieve “alignment” on business problems and opportunities. But different sources, answers or attention on the wrong things, results in tension. And, it’s not unusual to observe different stakeholders coming together with different versions of the truth based on different sources of data.

How do you synthesize data?

Applying synthesis is a three-step process. It starts with the creation of a conceptual map of the constructs, which becomes your ultimate overview and digest of the business issue. You can apply this to both quant and qual sources. And, youj might need to iterate the process as you drill in on specific areas.

There are some rules for this we will cover later, but first, the three steps:

1. Build a conceptual map of the constructs. This could be a fishbone diagram, a horizon tree diagram, or even a couple of boxes with lines on the back of an envelope! Most importantly, there will be an outcome, e.g. sales. Secondly, the constructs which drive the outcomes, e.g. marketing investment, creative strength, proposition strength, distribution (online and offline).
2. Identify and plot the different variables used to measure the constructs, so in the concept diagram you will have constructs and measures. This distinction between constructs and measures helps you integrate different types of research at the construct level, where at the measure level this is harder. This is what enables the dot joining!
3. Assemble the relevant pieces of data that validate why the constructs are in the diagram and the extent to which they are changing.

An example of data synthesis

Imagine you have been asked to explain a decline in NPS over the last six months. You might know that the research team has a driver model of NPS. Let’s use a fictitious example that resembles reality. The driver model might say that overall NPS is determined by:

1. Customer service – 20%
2. Trust – 20%
3. Product – 20%
4. Price – 20%
5. Ease of use – 20%

Now assemble the performance of each of these constructs based on the relevant measures. Utilise the NPS or satisfaction score changes over the period for the channels including digital and people channels, product, price, and ease of performance from internal and syndicated sources.

It’s likely that this picture will indicate where (at least the first why) the issues might be. In our example, we see trust and customer service have declined.

To drill in further on the issue, you can draw another concept diagram of the drivers of customer service. You find that service results are driven by:

1. The contact center NPS
2. The app NPS
3. Quality of issue resolution

You have noticed the measures for each of these items have declined. A review of the survey verbatims indicates a 20% increase in wait times to answer calls. Reports from the contact center indicate wait times have increased. The feedback survey from the app reveals that the decline in NPS was due to the loss of some features that people customers felt were critical and some changes users are not yet familiar with.

Tips for conducting a powerful meta-analysis

In order to utilise synthesis successfully, you need to keep a few tips and rules in mind as guard rails to your evidence-based approach.

1. Be comfortable if the data comes from different sources. That is ok. The measures that you are comparing will be consistent over time. You are joining the dots however with the constructs, not the measures! By doing this you will learn how different measures reflect the constructs.

2. Prioritize the most robust evidence in the conceptual map. It’s all right to have different levels of confidence on data points. Use a hierarchy of evidence to be comfortable with the information you are using. Larger sample sizes, and reputable sources are important, and directional replications of conclusions across different sources are what you are looking for.

3. Ensure that the observations about changes are statistically significant at the measures level. Even though NPS is a calculation, you can still do confident tests on the underlying question distributions. Don’t think about statistical significance as on-off. Be a Bayesian!

4. Measures of every construct are testable by their nature. Look to ensure multiple sources are directionally consistent. This replication of the same results adds to the strength of your explanation. The overall why answer you are seeking is the best explanation with the least inputs.

5. Keep asking why to find the drivers behind the drivers.

6. Be aware of the differences between cause and effect. Be aware of whether it matters in your business problem. While this can be a big topic, there is a practical approach to this. In our example above, operational reports from the contact provided proof the surveys were accurately solving the causality problem.

7. Beliefs or theories are never immune from revision based on new data. Your conceptual diagram should constantly be reviewed and updated as new data and information become available. It is a journey, not a destination. As consumer behavior changes over time, there should be changes in your conceptual diagram.