Do we bite the hand that feeds us?
We make our own services obsolete. From an entrepreneurial point of view, this sounds suboptimal at first. But for our customers, it is the key advantage.
Our solutions for data preparation with artificial intelligence create transparency and opportunities to exploit the potential behind a company’s own data. We receive existing data, prepare it and hand over a harmonized and cleansed data set. The company is satisfied and so are we.
After 3 months, the latest figures from the last quarter shall be considered for evaluations in order to check whether the initiated measures are already showing results.
Error. The new figures are not comparable because they are neither enriched nor cleansed.
Now the company can go back to kiresult for preparation. A new project, restarting the activities, execution and handover of the new data set.
This recurring loop and the potentially never-ending source of follow-up projects sounds good at first, but does not correspond in any way to kiresult’s overall objective.
We provide solutions for companies and users to maximize the benefits of limited and often wasted resources.
Enabling an efficient use of resources is the main objective of kiresult.
The process described above requires too many resources and has one main disadvantage: We correct mistakes made, but we do not help our customers to solve the underlying problem. Poor data quality arises when the data is created. So if we correct wrongly assigned data of a transaction subsequently, new data records will still show incorrect assignments.
This means that if an incorrect category is selected during order creation, a false spend allocation will be noticed at the latest during the analysis.
There are many reasons for incorrect category assignments. From our evaluation of existing spend data prior to our preparation, we find that either a too complex category tree across several levels leads to an overly specific and tedious category selection process. Another reason is the naming of the categories, which sometimes contain less intuitive terms or even only numbers. In these cases, particularly small demands are grouped together under “Animals”, for example, because this category is displayed as the first selection option in the system, or under “Other” because the correct commodity group cannot be found directly.
To ensure that the purchase requisition is entered correctly, free text requisitions in particular have to pass the purchasing department at least once in order to allocate, among other things, the correct category for the underlying demand. The buyers know the commodity tree and are annoyed over wrong preselections. Especially as the correct category is not in their own responsibility and must be forwarded to the appropriate colleague. Although these manual intermediate steps ensure that the data quality (category matches the demand) is correct, this process is in fact not streamlined. We are aware of technological possibilities to automate this manual step and thus disburden not only requesters but also buyers.
The algorithms and models of kiresult are trained to analyze texts from requisitions, orders and invoices and to assign them to suitable categories. We already use these basic functions for one-time preparation of historical spend data.
However, since this approach only enables us to look in the rear-view mirror, we do not meet the requirements of modern purchasing departments to have ad hoc access to real-time data in order to make fact-based decisions. Therefore, the use of our AI in the ongoing process is more effective.
We want to provide our solutions to the users already at the data entry stage to support them within business processes. When describing the demand and entering required data fields, our algorithms interpret the underlying data and show suitable options for the field “category”.
In this way we simplify data entry for our users on the one hand and ensure good data quality on the other.
In a nutshell, we provide data interpretation for effective decision making.
We have decided to focus on the process. kiresult aims to support data-driven processes in a systematic way and to solve the problem of faulty data creation.
To make this possible we want to offer solutions to make our algorithms directly available in the operative system of our clients, e.g. in ERP.
Integrated in the process and continuously applied, instead of recurring and manual.
For SAP, we can address our models via interface when creating orders and return suggestions for a suitable material group.
The buyer, alternatively requester when creating a BANF, can accept or reject the suggested material group and enter it manually.
The advantage of machine learning is that the addressed kiresult models are continuously trained and thus become better. With prediction accuracies >95%, the input can also be completely automated to make the process even more efficient, as no selection is required.
In practice, eProcurement systems are increasingly implemented and integrated with ERP systems for a more user-friendly interface and with helpful functionalities for the process. In these cases, the actual data creation does not take place in the ERP, but in the applied pre-systems. We also want to meet this requirement and make our solution available for potentially all applications via interface (API).
Why this is important: The users should not leave their familiar system environment, because with every system break, running processes are interrupted.
To make our algorithms potentially available to all cloud applications, we are working on flexible browser plug-ins that we have successfully tested in a first prototype. This opens up room for many more possibilities, such as
- Derivation of the right cost objects (G/L accounts)
- Identification and account assignment of requests to overhead cost centers
Potentially, by using our AI, all data fields can be pre-filled based on existing information as long as reference data is available. In our experience, a lot of historical data is available unused in every company.
All transactions cannot be directly automated with AI. First-time requirements for which there is no reference data can contain new combinations of parameters (supplier, cost objects, category, etc.). For these cases, data fields must be filled in manually at the beginning. With kiresult we see the potential to offer valid options for a smaller selection of possible data entries even for first-time requirements.
This is the key to automate operational activities, but also to support strategic tactical initiatives. Especially with large amounts of data across multiple systems, memorized rules and principals in the minds of employees eventually reach their limits. At this point, at the latest, we create individually trained models to efficiently consider lived rules even for large numbers of processes.
At kiresult we do everything to make the most efficient use of available resources and therefore the time of our customers and to release them from administrative burden.
For our service portfolio this means that the preparation of historical data can only be the first step for our customers. With the trained model of initially prepared data the way is prepared to support the running process. And thus to ensure an efficient use of resources
Are you interested in automatic data creation and category allocation in your processes?
Please leave your contact details and we will get back to you with further information.