In Reconfig, we frequently run demos with potential clients. The purpose is to showcase different parts of the tool and discuss how it can be applied to improve their organizations.
A few months ago, we discussed how we could further improve the demos by tailoring them to the people we speak to - without asking them to provide any data from their organizations.
The solution: generate synthetic data.
On most of the screens, we now have a button you can press that automatically populates fields with data. Not just any data, but data that are plausible given the organization you work for. We have integrated Reconfig with ChatGPT to make this happen.
As an example, entering the name and website of a university and pressing “Generate activities” produces a list of 20 activities that are realistic for that type of organization. The list can be fully edited - every activity can be modified, deleted, or supplemented with your own. Armed with synthetic data, we can demo every key feature of the tool using realistic inputs - the system even generates employee lists and organization charts with realistic-sounding names.
Once we had this functionality in place, we started wondering whether it could be used for something more. The data are artificial, so one could argue this is mainly a gimmick to help us run demos.
However, Tore Christiansen, a colleague at Reconfig, believes that synthetic data can provide real insight. In fact, synthetic data may allow entirely new types of organizational analyses in the future.
In a previous blog post, we discussed how clients can use Reconfig to create an Organizational Digital Twin (ODT) - a data-driven model of how an organization actually operates. This remains a core focus. Yet in most cases, it is fairly time-consuming and costly to keep a digital twin up to date.
With synthetic data, we can instead create a “digital sibling.” The goal is not a perfect representation of the organization, but one that is close enough - a representation that decision makers recognize as realistic. They can then use the model to better understand how their organization differs from competitors, or to simulate and test the effects of potential organizational changes. One can easily generate multiple siblings to evaluate various combinations of resources, activities, and organizational units.
As an example, Tore recently generated synthetic data for the world’s 12 largest airlines, going through all the different Reconfig features and calculating the automation savings each airline could achieve. Based on the synthetic data and integration with ChatGPT, Reconfig concluded that one airline has 60% automation potential in its customer service department alone.
The data will never be as precise as actual data from the organizations themselves - but they can be corrected along the way. If you have access to internal data or publicly available sources, you can refine the synthetic model and get an increasingly realistic result.
This capability not only enhances our demo experience but also opens up new possibilities for organizational analysis and decision-making. Synthetic data can be used to generate hypotheses about an organization that one can later confirm or disconfirm by collecting real data. As AI becomes more powerful, we expect synthetic data to become even more realistic and useful as a starting point for organizational design.