Everyone is talking about generative AI. Now that the first companies have launched their first projects, some companies are becoming disillusioned. Avoidable mistakes have been made. We explain what these mistakes are and how to avoid them.
These are the most common reasons why generative AI projects fail
In order to use generative AI effectively in a company, it is important to find the right mix of willingness to implement and planning. You certainly shouldn’t blindly ‘go for it’ without knowing where you want to go. But there is also no point in just watching from the sidelines while all the other companies are already gaining initial experience and realising added value.
To ensure that companies enter the playing field with realistic expectations, we have summarised the main reasons why AI projects in companies today are not as successful as they could be:
1. Unrealistic expectations
The technology is either not properly understood by decision-makers or users. Expectations are either completely exaggerated (AI is the jack of all trades) or downplayed (AI is stupid, it’s just marketing blah blah blah…).
2. No problem solution fit
(Generative) AI is only a means to an end. Never introduce AI to introduce AI. Be aware of the problem you want to solve with AI. Only then will you be able to find a meaningful and, above all, sustainable solution.
3. User training/awareness
AI only works if the user understands how the technology must/can be used. If we look around our own social bubble, we get the feeling that we are surrounded exclusively by AI experts. If you look deeper into the economy, there are more than enough companies where AI is not yet a real topic and the necessary knowledge is far from being widely available.
Quality of the data
One point that is repeatedly mentioned as a reason why AI projects fail is the quality of the data. However, this needs to be considered in a differentiated way – it depends on the use case:
Application example 1:
An image recognition AI is supposed to recognise whether a produced part is faulty or not. Of course, this requires a clean data set that contains error-free and faulty images and can be used as a reference for an AI.
Application example 2:
Internal expertise is to be made more accessible with the help of generative AI. Information is stored in various data silos and it is becoming increasingly difficult for employees to find the right information. A classic use case for generative AI. Many companies have already digitised far more knowledge than they realise. A well-implemented solution can get much more out of this than you might think. This is precisely the use case we solve with amberAI. amberAI is an in-house chatbot – based on an enterprise search.
Depending on the use case, data does not have to be prepared in a complex way, but can be used directly as a plug’n’play solution via standard connectors.
How do you prepare for a successful introduction of generative AI?
With over 60 customers, we have learnt that the most important thing is that the use case is right and that the users are involved.
The use case
There are thousands of problems in companies that can be solved from a technical perspective. But not every problem is worth solving. We have explained here how best to identify and prioritise a use case. Above all, it is about the business impact and the effort required to implement this use case.
The user
Especially for the use case of generative AI, success stands and falls with the acceptance of the user. This point must therefore be analysed from two perspectives:
- Training the user
Users of AI software will only be able to get the most out of the software if they understand how it works. The first step is not about the architecture and functions of the software, but rather about the functionality and technical limitations of the technology.
If you are considering introducing generative AI in your company, you should ensure that you have a realistic set of expectations.
- UX of the software solution
There are dozens of software solutions in companies today. It therefore goes without saying that not everyone can be fully trained on every software solution. It is therefore all the more important that the software is designed in such a way that it meets the expectations of the user. Only if it is intuitive to use will it be accepted accordingly. This is why we also advise against in-house development. Although it is easy to develop an initial demonstrator, this does not mean that you have a sustainable solution that is accepted by the user.
Software selection for the introduction of generative AI
If you want to select software, you should ask the right questions. What these are in detail naturally depends on the specific use case. Nevertheless, we have written down an overview of the most important questions for selecting AI software here.
The areas of questions include the following questions:
- User-friendliness
- Functionalities
- Data security
- Operation
- Training of the AI
- …
Introduction of the AI software
In order to successfully introduce the software, there are also a few things to consider. We have written these down here.
These include best practices such as involving a works council (if available) at an early stage, but also things that are easily overlooked. For example: Which systems should be connected in the first place? How will the data grow over the next few years? …
Testing, testing, testing….
To further increase the chances of a successful implementation, the software should of course be tested beforehand if possible. That’s why we at amberSearch have published a free online demo, for example, in which we have stored a demo data set of over 200,000 documents. This allows interested companies to get an initial feel for how such a solution works. We are taking it a step further. Anyone who would like to try out our solution with their own data is welcome to do so. Simply get in touch with us via the contact form and we will take all the next steps!