The AI introduction checklist helps companies to successfully organise AI projects. In most cases, AI projects either arise from the needs of a specialist department or are driven by the IT department or management. Since 2022, the topic of generative AI in particular has been on everyone’s lips and also offers numerous use cases in companies. This is why more and more companies are looking into this topic.
To ensure the success of the introduction of a generative AI application, three key factors are crucial: the data, the requirements and the users. Regardless of the underlying technological basis of the search engine, the number of data sources or the size of the index – without the users, the project is not feasible. Even if the AI product offers outstanding functions, the project will inevitably fail if the users are not adequately supported in their specific requirements or the usability of the software is not intuitive. It is therefore advisable to consider the following tips.
The user at the centre
The user and their needs must take centre stage for the AI project to be a success. It is not enough to simply convince the line manager to make budget available for the search application. Before such a project is considered, the individual departments and users should be taken into account.
The better you understand the users and their daily work, the more precise the requirements and the benefits that can be generated.
Users are used to a very high user experience (UX) from their private applications (Instagram, TikTok & Co). B2C software solutions are designed to be very lean and user-friendly, something that is often neglected in enterprise software. At the same time, users have to familiarise themselves with more and more software solutions. In order not to overwhelm the user, care should be taken to ensure that the software has a very good UX that is easy to understand even for inexperienced users. This also reduces the amount of training required.
Consideration of different scenarios
In contrast to other software solutions, where the aim is to strictly map a rigid process, the introduction of generative AI is all about having flexible software that can cover as many use cases as possible. Nevertheless, you should not look for the perfect solution, but rather focus on a few main use cases or departments.
There will be so-called power or key users who use the search every day and, for example, set up an AI tool such as amberAI as their personal start page in the browser. In contrast, there will also be users who rarely use the software. And there will be many users who integrate the AI software into their daily work routine, just as an internet search engine, for example, is used with varying degrees of frequency.
The data
For years, many companies have been told that a software solution only works well if the metadata, tags etc. are correct. Thanks to new breakthroughs in artificial intelligence, this is no longer the case. There is no need for time-consuming manual maintenance of metadata – most employees have only done this half-heartedly in the past.
Instead, new AI solutions based on intelligent algorithms can find and process information based on semantic content. This means that data – at least if you want to solve a kind of internal company ChatGPT or similar use cases – no longer needs to be processed in a time-consuming manner. The AI now does this independently. Nevertheless, there are still other use cases (e.g. if you want to train your own AI model) where this is still necessary. However, this is not relevant for the application of generative AI.
Access rights
An important aspect of preparing an AI project with internal company content is checking the rights on the file servers and other systems. Copying or moving directories can lead to the problem that directory and underlying document rights are not inherited correctly. Whether with or without AI, this is a challenge that must be resolved for compliance and data protection reasons alone. The introduction of an AI should therefore be seen as an opportunity to quickly uncover such security gaps. Documents can then be found using the search function, even though the user may not have read access to a folder above them. A good AI solution displays the relevant directories as part of the initial clean-up work, making it easier to correct the rights violations.
Requirements – the AI solution should be scalable
In our blog post Introducing generative AI, we have already discussed various requirements for solutions in the field of generative AI. However, it is important that it should be easy to connect new systems and that the system is scalable. After all, the amount of data in companies is growing faster and faster. The solution should be able to “grow” with them.
As such systems almost always rely on a retrieval augmented generation system, an index is created first. In principle, such an architecture can only process information that is also in the index. Indexing should therefore be restarted as often as possible.
Different languages play a major role for companies that are represented in different countries: In which language are the documents available? Which language is spoken by the employees? Which language should the search application support? The different locations must also be taken into account.
In order for interested companies to be able to ask the right questions, they should at least have a basic technical understanding of the technology they want to use.
Our tip: Start small for maximum success!
In order to gain experience and be able to share these learnings with colleagues during the roll-out, it is worth taking a step-by-step approach for larger companies. The lessons learnt can be used to gain further experience for the next group during onboarding.
It is therefore advisable to initially index only a manageable amount of data, for example parts of the file server (departmental drives), and not yet allow the entire team to use the solution. If everything works as desired, additional data sources can be connected at any time and the solution rolled out to all colleagues. As soon as the AI solution is fully productive, the entire college can be informed via an internal marketing campaign! Only if colleagues know which practical tool is available to them will they use it.
Staff and works council
Staff and works councils should be informed right from the start, even if it means extra work initially. As projects in larger companies start with a pilot project, you should try to keep the bureaucratic hurdles as low as possible in the initial phase. Only when the pilot project is running successfully should you try to go through the necessary processes to enable a long-term roll-out.
In principle, it does no harm to convince these bodies/committees and thus ensure early backing.
The AI introduction checklist
In this section, we have set out the most important points to consider when introducing generative AI.
Data
The data section is about having as complete an overview as possible of the relevant data so that as few adjustments as possible need to be made to the target image during the introduction.
⬜ Which data sources should be connected to the AI solution?
⬜ How much data (or giga/terabyte) will be in the index (ideally broken down by system)?
⬜ Should it be a stand-alone system or should the AI solution be integrated into other systems (Teamsapp, Iframe, …)?
⬜ Which languages should the AI solution support?
⬜ Is there an estimate of data growth over the next few years?
Users
The better the future users are known, the better providers can respond to their wishes. It is therefore important to think about the use cases and the resulting users at an early stage.
⬜ How large is the user group? Which departments do the users come from?
⬜ Should the AI solution be rolled out to different (national) companies?
⬜ Is there any specific, sensitive or personal data that should not be included and should not be found?
Requirements
Last but not least, there are the technical requirements. In order to better categorise the technical requirements, you should also gain a certain basic understanding of the technology. We have written this blog post about this.
Business and organisational requirements
How should the AI solution be used?
⬜ Which stakeholders need to be involved? (Works council, IT security, data protection)
⬜ What are the requirements in terms of server/cloud, support and licences?
⬜ How often should the index be updated?
⬜ What are the requirements for the front end? Does it need to be adapted to the corporate design?
Why should the AI solution be used?
⬜ What benefits should be achieved with the solution?
⬜ Which departments will use the AI solution most frequently?
⬜ In which scenario will the AI solution be used? Knowledge Worker on PC, mobile or both?
When will the AI application be ready for productive use?
⬜ Who do I need to contact in order to obtain the necessary technical users for indexing?
⬜ What is the timetable for implementation? What deadlines need to be met?
⬜ Should the AI solution be implemented by the IT department itself or by an external service provider?
Technical requirements
⬜ Is the solution able to map all access rights? (single sign-on if necessary)
⬜ How is security and GDPR compliance guaranteed?
⬜ Is the software scalable and can it also support larger data volumes/users?
⬜ What is the AI provider based on? Is it based on its own models or is it dependent on other companies?
⬜ How flexible is the provider’s architecture if stronger models are published by other providers?
⬜ How comprehensible are the results displayed to the user?
⬜ What integrations are available?
⬜ How up-to-date are the results displayed by the AI solution?
⬜ How is the software hosted? (Cloud, on-premise, private cloud, …)
What should software with generative AI offer the company?
Regardless of whether employees need better access to existing expertise or customer service needs to be improved, or whether you are forced to thoroughly clean up your database as part of a migration to a different IT infrastructure – intelligent AI software is suitable for many scenarios. It helps employees to search for and find information and, with deep insights into the database, also enables further applications in the area of document analysis. It is an important component of the digital workplace and a fundamental building block of effective knowledge management. When selecting software, pay attention to the following features and functions:
Easy implementation: It must be ensured that the solution is enterprise-ready and has a high cold-start capability during installation. This means that no complex IT project needs to be planned for the introduction.
Comprehensive search: It must be possible to search in several hundred data sources and file formats without any problems. Most intelligent systems require a search in the background anyway, and this should also be made available to users for certain use cases.
Search on current information: Information should not be out of date. The index should therefore be updated regularly.
Secure search: It is important to ensure that the relevant user authorisations are taken directly from the Active Directory. This ensures that users only find the documents for which they are authorised.
Multi-client capable: The AI solution should be able to be rolled out to multiple clients. This means that subsidiaries and departments can easily access different indices as part of a single installation.
Simple user interfaces: Whether knowledge workers, mobile service technicians or occasional users – the solution should provide the right user interface for every requirement.
The benefits for your own company
Greater efficiency: Finding information faster enables efficient working. Employees no longer waste valuable time searching, but can have information prepared directly by generative AI.
Faster response times: The targeted finding and dissemination of information enables faster response times and answers, thereby increasing customer satisfaction.
More security: Breaches in authorisation concepts can be detected and rectified in the course of implementation.
More knowledge: Information is not lost in drives or other systems, but can be found centrally.
Less risk: A comprehensive AI solution minimises the risk of overlooking important information. This means that employees work on the best overall information instead of the first best information and have a much better basis for decision-making.
Lower costs: Efficient working means that resources can be optimally utilised.
More fun at the digital workplace: New employees can be familiarised more quickly and benefit from existing knowledge. They get a search engine with all the AI functionalities they are used to in their private lives – only better.
Customised hit lists: You should ensure that the software delivers customised hit lists based on boosting factors, in which the relevant search hits are displayed at the top.
Linguistics: Questions should be able to be asked in natural language to make it as easy as possible for the user.
No metadata tagging: Data should not have to be elaborately prepared first. Nowadays, AI is able to deal very well with data that is not perfectly tagged, as the content is primarily used as the tags.