Many companies need and want to gain experience with generative AI. However, tools from the internet quickly reach their limits – they are too general and not sufficiently adapted to the context and challenges of the company. Nevertheless, the tools available online are ideal for gaining initial experience in the field of AI or generative AI. However, this blog article will focus on how to successfully combine in-house expertise with generative AI – within a secure and GDPR-compliant framework – and what to look out for. The introduction of generative AI is not as complex as it first appears.

Understanding the added value of generative AI

Generative AI can save a lot of time, particularly in the formulation of texts (e.g. in marketing or sales) but also in customer service (answering questions) and – if used correctly – deliver comparable quality. Another use case is the management of internal company know-how, which is usually stored in various data silos. The correct use of generative AI can ensure that employees can access the relevant expertise quickly and efficiently, regardless of where it is stored. 40% of knowledge workers spend more than 30 minutes a day on internal research. In addition, the time spent on digital searches has increased by 77% in the last few years alone. Whether it’s about making information accessible during onboarding or during long-term work, or whether texts are to be generated based on internal expertise, with amberSearch we create a central location where employees can solve these and other use cases.

In one of our older blogs, we once calculated the business case for an enterprise search. With the help of generative AI, further use cases are now being added.

What fundamentals are relevant for the introduction of generative AI?

In recent years, we have realised that the most successful projects are those in which the project partners have a certain understanding of the technology. These project partners have the most realistic expectations of the software and can therefore identify where the potential lies. In the blog article linked here, we would like to give a rough overview of the various basics. We have already written more in-depth blog articles on some of the topics, and there will be more articles on other topics in the future. You should also have defined the right use cases. We have outlined what this can look like here using an example.

What requirements do most companies have?

Regardless of the technical basis, most companies have very similar requirements:

  • Access rights: The most important requirement, especially from a GDPR perspective, is the consideration of access rights. Every employee should be granted access, ideally controlled via an existing access rights management system (e.g. Active Directory or Single Sign On).
  • Security & GDPR compliance: A fundamental prerequisite for the integration of such systems is an appropriate security architecture and compliance with, for example, the GDPR. At amberSearch, we take all these points into account.
  • Scalability: Another requirement is scalability, especially in terms of performance. One thing is to have a stable system in a small test group. Another is that this system also works stably with large numbers of users.
  • Independence: The OpenAI scandal in November 2023 showed that dependence on other providers can be problematic. At amberSearch, we have the expertise to make changes to our algorithm ourselves at every point and do not rely on third-party providers such as OpenAI & Co. as standard.
  • Flexibility: A major point that sets us apart from other providers at amberSearch is our flexibility, which goes hand in hand with the independence described above. AI models evolve very quickly, which means they need to be interchangeable so that the solution remains state of the art. We can therefore also replace our AI models with more powerful ones at will. Incidentally, this is another reason why you should not train company-specific AI models for most use cases.
  • Traceability: When an AI model starts to formulate answers, these should be traceable. This is why we at amberSearch reference the source file from which we have quoted information. This allows the employee to directly understand where the information comes from if they suspect hallucinations.
  • Integrations: A system like our solution will only be successful if it is smartly integrated. We have therefore developed various integrations (both systems that we can connect and how we make our solution available [e.g. Teamsapp, desktop app, iframe, mobile, …]), which we make available to our customers. Another point is the integration of AI into existing data sources such as drives, Teams, Outlook, etc…
  • Up-to-dateness: A company’s expertise is constantly evolving. A system like ours must therefore not be based on a rigid data set, but must always take the latest findings into account. Of course, we also take this into account.
  • Hosting options: Should the software be hosted as a managed service, in the private cloud or on-premise? Due to the relatively high hardware requirements, a managed service deployment is usually an option for smaller companies in particular – regardless of the provider. At amberSearch, however, we support all variants.

What answers should an interested company have when talking to providers?

Of course, many things can be worked out together with the provider. At the end of the day, the following questions must always be answered and the better prepared you are, the more successful the project will be:

  1. Which use cases should be solved in the company?

In order to implement a successful project, you need to know what added value you expect from the tool and what challenge you want to solve. Generative AI can be used in a variety of ways and in different departments, but you usually start with one use case. This helps to build up a deeper understanding of the technology.

  1. How many users/departments should have access to the system and by whom should it be used and how often?

In order to design software accordingly, it is helpful to know how many users are expected to use the software. In most cases, the number of users or frequency of use is already determined by the use case.

  1. How should the software be integrated?

A key point is the integration of the software into the existing IT landscape. This must be considered from both sides:

  • Which of the existing IT systems should be connected to the AI?

At amberSearch, we have a whole range of different standard integrations. Self-developed systems can also be connected via an open interface, so that from a technical point of view there is always the option of connecting a system to us.

  • Where should employees access the software from?

Software like ours will only be used appropriately by users if it is accessible to them. That’s why our solution is where the user is. We offer integrations such as a team app, an iframe solution, mobile optimisation, Chrome plug-in and more alongside the classic web app so that users can easily access amberSearch at any time.

  1. How should the software be hosted?

If generative AI is to be used to achieve comparable results to ChatGPT, for example, in combination with internal company data, then in most cases it comes down to a managed service or private cloud variant. If you have smaller licence quantities (<300 units), you will have to rely exclusively on a managed service variant, as it is difficult to calculate a business case in advance compared to the necessary hardware requirements. From a technical point of view, we can provide amberSearch as an on-premise system due to our independence, but this is more common in exceptional cases.

Tips for the tool selection

There are currently a large number of tools on the market. The added value that is promised is usually quite similar. However, the differences lie in the details, especially in the maturity and quality of the information generated. On the one hand, care should be taken to ensure that the solution fits the use case. On the other hand, many providers only offer a certain part of the processes mentioned here (e.g. either the search or the generation of answers). Providers that focus exclusively on search usually do not have the expertise to connect their own generative AI, but usually have to rely on third-party providers such as OpenAI, Google Bard or Aleph Alpha. Providers that focus exclusively on generating answers have difficulties selecting the right data in the first step.

The integration of amberSearch was actually as easy as originally discussed and enabled us to quickly activate the solution for users. After the quick set-up, we were particularly impressed by the speed at which user feedback was incorporated into the amberSearch software. You can tell that the team is passionate about what they do and takes customer requests seriously.

Nader el Azabi, Head of IT at R-Biopharm

At amberSearch, we offer such a solution from a single source, i.e. a complete end-to-end integration takes place directly. This means that even companies with limited IT resources can realise the added value of modern technologies such as generative AI. The technical integration effort (in the standard case) for our solution is half a day.

As we have built up a lot of expertise in this area over the last few years, we have also written a blog article on selecting AI software, which is primarily about the questions that should ideally be asked of a provider if the answers are not known. However, our blog article Checklist for introducing AI software is more about how a company that uses AI should approach such a project.

We will be happy to contact you for further discussions. If you already know us, please fill in the following fields.