### Increasing efficiency through data analysis

Dragonfly is a product that helps increase efficiency of supply chain operations. It can be used for warehouse optimisation via inbound and outbound volume forecasting, simulation of different “what-if” scenarios and optimisation of product location, range, inventory and resources.

The importance of data in managing supply chain and logistics operations is widely recognised. From cost reduction to risk mitigation and supplier relationship management, data fuels a wide range of critical activities across businesses.

Indeed, we have witnessed this in our contact with different customers within and outside Unipart Group. In particular, we realised that, even though the underlying operations were vastly different, all the customers we spoke to had two commonalities:

First, the customer seeks to predict the future: What is the volume of operations for next week? How much demand will I see for a particular product next month? How many products of a certain category will I need to process tomorrow?

Once a forecast is available, the customer is then interested in using it to estimate the impact on the business. For instance, the customer may be interested in controlling operational costs, managing risk, or improving other relevant KPIs.

Finally, the customer aims to adjust their operations to improve these KPIs: How can I best allocate resources next week? Do I need to be more conservative with my inventory policy?

Contrary to what many believe, we found that our customers have a clear vision and understand the value behind a data-driven approach. However, we observed that they were all using inadequate tools to interpret their data.

As an example, consider a customer who is interested in optimising a particular aspect of their operations. Typically, the necessary data is stored in a database to which the business analyst might not have direct access to. Instead, the analyst is given access to a large export of data in a spreadsheet. This spreadsheet will have to be loaded, cleaned and transformed by the analyst before further analysis can be performed. The whole process of getting data from the source database to a state where it is ready for analysis might take several days and consume valuable resources.

The next problem is concerned with the actual data analysis process. Spreadsheets are slow and are limited in their capabilities or even can crash when too much data are loaded. A talented analyst can eventually work around its limitations and manage to account for all data available, but that might take a very long time and will certainly result in a sub-optimal solution in terms of computational efficiency. Moreover, developing advanced machine learning methods using spreadsheets is impractical, so the analyst is limited to basic methods.

Altogether, by the time the analyst completes the analysis process, the data may be out-of-date and the resulting insights may be of limited use. In practice, the entire process is too slow, which hinders its frequent use and, in some cases, even leads to abandoning data analysis efforts.

### The building blocks of Dragonfly consist of:

- Advanced Machine Learning forecasting algorithms – Sophisticated, interpretable forecasting algorithms for uncertain environments. Example application: Daily forecasts of inbound and outbound workload.
- High-performance process simulators – Process simulators for decision support and “what if” scenario testing. Example application: Estimate effect of process change on KPIs.
- Inventory and warehouse optimisation – Powerful, automated optimisation tools for process improvement. Example application: Range and location management across multi-tier networks.

Dragonfly is highly modular, meaning that the solutions above can be deployed flexibly to best fit your needs. We can deploy Dragonfly as a standalone web application or as a plug-in service to enhance other tools, such as UIS, UDES or TMT. We have designed Dragonfly such that these modules interact naturally and can deal with large amounts of data and can integrate with other platforms through standardised interfacing. This holistic, high-performance solution represents a step-change relative to the tools commonly used by most of our customers (such as slow and impractical spreadsheets) and allows you to bring data-driven decisions to the core of your business.

### Forecast

Forecasting is an indispensable tool in supply chain and logistics. Businesses rely on forecasts to make important decisions such as determining how much quantity should be stocked at a particular warehouse and how many operators are needed to perform a particular set of tasks. As a result, inadequate forecasts can have serious implications on the effectiveness of the processes that rely on them.

Even though there are many forecasting tools on the market, traditional forecasting methods tend to overlook a critical aspect of making predictions: uncertainty. Intuitively, uncertainty plays an important role in any decision making the process. For instance, imagine that a particular forecasting algorithm tells you that you should expect to process 500 order lines tomorrow. That provides you with a rough idea of what tomorrow will look like, but how confident can you be? What if the forecasting model told you that it is 95% confident that you should expect between 450 and 550 order lines tomorrow? In contrast, what if the forecasting model told you that it is 95% confident that you should expect between 250 and 750 order lines tomorrow? These scenarios are different and would, therefore, require different approaches to deal with them. This demonstrates that considering uncertainty is a critical aspect of optimal decision making.

We develop advanced machine learning techniques that are specially tailored to deal with uncertainty. These methods generate accurate and interpretable forecasts for a variety of data types. For instance, you can use Dragonfly to forecast future inventory demand or operations workloads.

### Simulate

Human decision making is often very accurate, but also highly subjective. Using discrete event simulation, Dragonfly assists you in predicting the impact of a decision on your KPIs, thereby supporting you in making data-driven decisions for your operations.

Suppose you are responsible for managing a multi-stage device screening process at a warehouse and you need to decide on how many resources should be allocated to each screening stage. On the one hand, if you allocate too many resources to a screening stage that has a modest flow of devices to process, you are effectively wasting resources. On the other hand, not allocating enough resources to a given screening stage may cause bottlenecks which decrease productivity.

This is where Dragonfly can help. By building a simulator of the device screening process, we can help you estimate the impact of your resource allocation strategy on your KPIs.

### Optimise

We have now seen how Dragonfly can help you simulate your operations, but what if we could develop a tool that also optimises them? This is why we provide optimisation solutions as part of Dragonfly. Provided that we can simulate your process, we can build optimisation algorithms that adjust your operations such that the best possible outcome is achieved.

Let us consider the problem of allocating the correct level of staffing across the different workstreams of a warehouse. The forecasting solution provides you with an estimate of how many operations need to be performed for the different workstreams next week. Given this forecast and a particular staffing level, the process simulator allows you to estimate your KPIs for next week. The goal of the optimiser is then to find the staffing level for which the simulator predicts the best set of KPIs. Given a set of process parameters (e.g. staffing levels) and an objective defined by the customer (e.g. reduce cost), the optimiser uses the process simulator to search through staffing combinations and find the process parameters that produce the best KPIs.

Together, the **Forecasting**, **Optimisation** and **Simulation** modules come together to help you optimise your operations. We have designed Dragonfly such that these modules interact naturally and can deal with large amounts of data and can integrate with other platforms through standardised interfacing. This holistic, high-performance solution represents a step-change relative to the tools commonly used by most of our customers (such as slow and impractical spreadsheets) and allows you to bring data-driven decisions to the core of your business.