From the idea to an intelligent solution with Tallence

How we can support you in creating sustainable value through Artificial Intelligence

Tech // Feb 3, 2020
Csm Tallence Website Header afterwork new 702cff6aa9

In the previous article, we gave an overview of the field of Artificial Intelligence and outlined important use cases. The article describes exemplarily the enormous potentials and the various fields of application we see for the use of AI-based algorithms in different companies and industries. Through the targeted use of AI, business activities can be made more efficient and profitable. In addition, we have addressed factors that currently still restrict the widespread use of Artificial Intelligence in companies. Typical reasons for this are a lack of know-how, an unsuitable data basis or the lack of imagination for potential use cases within the own company. At Tallence, we can draw on a profound knowledge of machine learning as well as data science and engineering. We use this knowledge to accompany our customers on their journey to AI-based solutions as a competent partner.

With our support, AI solutions in your company will no longer remain just a vision, they will be brought to life and help you to make your everyday business more efficient and innovative. As experts for individual and enterprise software, we develop customised machine learning solutions for you and integrate them into your existing system landscape. For projects in this context, it is essential that development and consulting go hand in hand to create long-term value. We can rely on our 20 years of expertise in software development and digital transformation consulting and accompany you as a full-service partner throughout the entire process. In addition, the change towards the benefit-maximising use of Artificial Intelligence often requires a modification of existing data structures. These may also have to be rebuilt. We support you just as professionally as we do in setting up cloud systems and IT system architectures that enable you to collect, manage and analyse the large volumes of data that may be required for your machine learning solution.

Transparent communication with you and continuous feedback are very important to us throughout the entire process and enable us to provide you with a customised, sustainable solution. The process is individually adapted to the customer. We choose between an agile, classic or hybrid approach in order to make the development as efficient and target-oriented as possible. In the following, we would like to explain our process step by step:

Darstellung der 10 Prozesse für die Integration von Machine Learning

1. Analysis of the business processes as well as the data basis and IT infrastructure

In the first step we analyse the status quo. We look at your processes in detail and identify digitalisation and optimisation potentials in close communication with your specialists. We outline the identified potentials in first use cases. We also check your data. It is the basis for Artificial Intelligence and largely determines how well the later algorithm performs its tasks. It is therefore essential to carry out an in-depth analysis of all data generated in the course of everyday business in your company. In this way, we can evaluate how we can use your data even more effectively in the future for the intelligent automation of your processes. In addition to the data, we also analyse the existing IT infrastructure of your company. This is necessary because the solutions to be developed must be seamlessly integrated into the current structures in order to generate the desired value.

2. Evaluation and prioritisation of use cases

The insights gained from the analysis phase form the basis for the evaluation and prioritisation of the outlined use cases. In this step, we evaluate together with you where in your company Artificial Intelligence can add significant value and quantify the impact of possible solutions on the current business. In this way, we want to ensure that the cost and return on investment for the implementation of the AI solution are in a healthy ratio and that the business value is maximised. We then prioritise the use cases in terms of their potential and feasibility. At the end of this phase, a selection of prioritised use cases is determined, which are then targeted for realisation.

3. Conception of the solution

While use cases were outlined in the previous step, the task now is to work out a detailed concept for the realisation of the selected solutions. At the beginning of this phase stands the requirements analysis. In requirements workshops with the relevant stakeholders, we determine which functions the new software should offer and which quality standards are to be applied. The requirements are refined and reconciled over the course of the workshops and result in a specification. The requirements also provide information about which data is relevant for the new system and in what form it must be available for optimal use. An initial proof of concept, which we will develop with your support, already gives an outlook on the future solution. We hereby test the general feasibility of the envisaged solution by implementing and validating selected requirements. The knowledge gained from this serves as an input for the solution concept, which forms the starting point for the subsequent preparation of the data basis and development of the solution.

4. Preparation of the data basis

Even if there are already many data pots in your company, it is possible that they do not yet have the form and quality to implement the use cases optimally. In this phase, among other things, different data sources are brought into a consistent format and incorrect or inconsistent data is corrected. This also involves identifying necessary interfaces to existing systems. Of course, we are also happy to support you in bringing your data storage systems up to a more modern standard and to optimise them, for example by creating a data lake. A concept for the selection and generation of suitable current training and test data is also one of the end products of this phase. In this process, we always pay special attention to compliance with the highest data protection standards.

5. Model development and optimisation

This step consists of two closely linked phases, which alternate in a cycle until the final model has been developed. In the first phase (5a), a basic model is developed, or an existing model is improved. Then, in the next phase (5b), the model is evaluated, and its weaknesses are examined. The insights gained then lead to a continuation of phase 5a. This feedback loop between the two phases runs until the model provides adequate results that meet the requirements. At the same time, the training data basis is continuously optimised to support the process in the feedback loop.

a) Development and optimisation of the models plus training

In this step, we build to a large extent on our preliminary work and use our proof of concept and the prepared data. Based on the concepts, the machine learning models are developed precisely and ready for productive operation. In contrast to the proof of concept, the focus is now much more on the performance and quality of the solution. The initial training with already existing training data enables the first steps on the way to the solution and an evaluation of the models with regard to the requirements of the use case (see step 5b). Depending on the use case, it is important to keep in mind that the models can also be retrained with daily updated data or even continuously improved.

b) Evaluation of the models

The evaluation of the models includes not only the evaluation of various standard quality features (e.g. F1 score, accuracy, data throughput), but also testing for typical problems from the machine learning context, such as overfitting (the trained model only explains the training data but not a general case) or bias (there is a systematic error that leads to large deviations). In addition to standard quality assurance procedures, we work hand in hand with your experts to check the plausibility of the results. We will provide you with the necessary tools to better investigate even black box algorithms. In this way we can increase the interpretability and explainability of the models.

6. Integration of the machine learning solution into productive systems

After developing and evaluating the models, the new solution must be integrated into a productive system. It depends on the nature of the solution and the existing IT infrastructure how exactly this step is implemented.

We build independent solutions as well as modules that are deeply integrated into your system landscape and use new or existing interfaces. Our expertise in front- and back-end as well as app development allows us to provide you with optimal support in both scenarios and to ensure that the machine learning models can be used optimally for the respective application. If desired, we ensure that the models used can be continuously trained and improved with new data.

7. Testing of the overall system

Especially when integrating into complex existing systems, both new and old components must be tested intensively with regard to their interaction.

In addition to classic integration tests, the performance of the system must also be ensured. Security aspects such as data protection and protection against DDoS attacks must be taken into account. If requested, our testing team will provide you with full support in all matters during this phase and will work out a suitable test concept with you. In cooperation with our developers and data scientists, the team will ensure that at the end of this phase you receive an optimally balanced system for productive operation.

8. Piloting and acceptance

Before the new solution is finally integrated into productive operations, we include a pilot phase in order to give a larger group of employees the opportunity to get to know the solution and to carry out final quality optimisations and error corrections on the basis of productive data. At the end of this phase, the final solution will be accepted by you, the customer. The solution is now ready for the Go-Live.

9. Go-Live

With the Go-Live, the solution is made available to all users and integrated into your productive business processes. Depending on your needs and the complexity of the system, we support you intensively during the first weeks of the Go-Live with tailored training courses on the new solution and are always available as a contact partner in case of problems.

10. Continuous development and onboarding of new projects

Our support does not end with the implementation of the new software. Throughout the entire software life cycle, we continue to be available for you as a competent partner when it comes to operating, further developing and adapting the software to changing market or company conditions. We can also actively tackle the identification of further AI potential, the targeted development of solutions through to their realisation and integration with existing software in subsequent projects.

Parallel to the actual development and implementation of AI-based solutions, we support you with a change management programme tailored to your organisation. The path to Artificial Intelligence can cause uncertainty and fear on many levels of the company, as its implementation can bring about a number of changes and is often associated with the reduction of jobs. With a holistic approach, which we develop in close cooperation with you, we want to counter negative feelings in a systematic way. We rely on regular, cross-departmental communication that is tailored to the information needs of individual stakeholders. The basis for this is our well-founded expertise, which we have been able to build up during the implementation of numerous digitalisation projects in various industries. Our goal is to strengthen the acceptance of your employees towards Artificial Intelligence and to ensure a seamless integration of the developed solutions into everyday operations. In addition, we are happy to support you in adapting your existing processes, which are affected by the new technology, to future conditions and thus to fully exploit the potential created.

Article by: Dr. Bastian Eggert, Timo Fuhrmann and Dr. Kai Matzutt