Every AI or Machine Learning project is unique: diverse data sets with different variables, integrations in or with existing software or hardware and different expectations and goals to be pursued. The decision as to how a business case should be set up technically in practice is an important factor in its ultimate success. In this article a number of important points of attention are summarized to take into account when choosing a suitable AI software.
1. Consider data preparation
The quality of the available data will be essential for the quality of the final solution. “Garbage in, garbage out” applies here as well. Data – however qualitative – is rarely suited to start building an AI or auto ML model right away. Often the required data is spread through various databases, software systems or data files. So make sure you choose an AI software platform where you can not only generate an AI or auto ML model but also offers to do the essential data transformations within the same environment. Look out for tools that can partly automate data transformation and enrichment, which makes starting an AI project a lot easier.
Also think about other important steps for your application. For example, if you want to build an image recognition application, data, in this case photo or video, will have to be annotated or labeled to train the model. Therefore, check whether all this is possible within one and the same AI software platform so that you do not need another system just to be able to do the preparations. You are probably already using a wide range of apps or software systems, so managing everything within 1 environment is advisable.
2. Model updates are crucial
Every AI or ML model is based on data. Often the data is dynamic in nature, seasonality e.g., resulting in a need for adapted models in predictive maintenance applications for machines. A model may be the best solution for you today, but it is not necessary and even unlikely that this model will also perform best within 3, 6 or let alone 12 months.
To keep performance high, keeping the built models and applications up to date is crucial. If all your models have to be maintained manually, this requires a continuous allocation of time and resources, which in the long term has a negative impact on the ROI of your solution. To mitigate this, choose a software solution where model updates are automated. This way you always have the guarantee of the best performing model for your business case without having to worry about extra work.
Just like your data, your business processes are not invariable. You might want to integrate an extra data source after a while, or want to include or delete certain variables in your AI model. Or you simply want to experiment with a variant of the application. Make sure that you are in control and can easily make adjustments yourself without having to seek the help of an external partner. This way you retain maximum cost control.
Keep yourself well informed about offered training and ensure that the software is simple enough. Always ask for a live demo of the software.
4. Scalability is key
The search for a suitable AI software platform usually starts from the need to realize a specific business case. However, it is very likely that other opportunities will be discovered over time, whereby artificial intelligence or machine learning can be of added value. So make sure you choose a tool where you can manage multiple AI use cases or machine learning flows from one and the same environment.
For every application that you want to build, a connection with a data source will likely be needed, which is an important point of attention when assessing the scalability of an application. In other words; make sure that connections with other software or databases can easily be reused.
5. Easy to integrate into your existing way of working / ecosystem
To analyze data you obviously need access to that data. Integrations with databases or software in which this data must be retrieved can be time-consuming and expensive. Therefore choose software that has an open API architecture.
Some ai software platforms have a marketplace where all kinds of plugins have already been built for commonly used software systems such as Microsoft Dynamics / Navision, SAP and others. You only need to enter administrator credentials to make the connection. No custom development needed, so you can get started quickly without budgetary implications.
6. Customized support
The right support is a must, especially when getting to know a new AI software. Test the help desk, will you get someone to speak to you immediately? Great! Nothing is more annoying than being sent endlessly around and struggling from one ticket to another. If you can be further assisted by a help desk employee in your native language, this is an extra asset.
7. Transparent pricing
Everyone has experienced projects where development costs go wild. Make sure you clearly map all requirements at the outset and that you also include any expected future costs, such as model updates, adding data sources and the like.
Be sure to inquire about included support. Nothing is more convenient than being able to rely on expert knowledge without having to pay a price tag.