We at amberSearch have been working on the topic of AI since 2020. Whereas in the early years of our company, no one in our target group knew what AI was or how this technology works, AI has now mutated into an absolute buzzword that has reached the masses.
Time for us to take a look back at what we have learned and what companies that are now starting to integrate AI into their processes can learn from our experience.
AI is hype – but AI also creates added value
Since the beginning of our company, AI has always been a means to an end. There was a challenge (in our case, the difficulty of finding information across different data silos) and to solve this challenge, AI was the best technical solution. Especially at the beginning of our time, “AI” was never a reason to buy, it was never in the foreground because very few people knew what to do with it. This changed with the introduction of ChatGPT in November 2022 – suddenly everyone was talking about AI and anyone who didn’t take care of AI now would be lost tomorrow.
Suddenly people were buying solutions because it said “AI” and they wanted to be part of it. In the meantime, this type of software purchase has at least reduced somewhat, but the trend also showed that very few companies really knew what and why they were buying such solutions. They often bought just to be there. And many providers – even today – still have difficulties defining clear use cases that deliver the promised added value and return on investment.
First of all – AI is so much more than just chatting with data. There are various AI technologies, but only large language models have arrived on a broad scale, as these enable chatting with data. That’s why we’ll limit this blog article to AI best practices that are based on this technology.
In this blog article, we will therefore look back on some experiences and best practices that we highly recommend to any company that wants to get started with AI integration.
First steps with AI in companies
Especially in 2023, there was hope – I introduce AI and everything will be better. What exactly AI should be in this case and where exactly it should be introduced was never specified in more detail (we will explain what a sensible AI roadmap is in the course of this blog post). The fact is, however, that before companies integrate AI deeply into their processes, they must ensure that the company’s own workforce has a certain understanding of how AI works. Only then will employees and departments be able to formulate realistic expectations of a solution. If this knowledge does not exist, one half will expect far too much from a solution and the other half far too little. Neither is good for successful AI integration.
In reality, it subsequently turned out that AI is introduced in many small steps and not in a few large ones. Companies need to start taking these steps today.
Horizontal vs. vertical AI solutions
In order to spread knowledge across the workforce, companies should start with an AI chatbot that is available to all employees. Ideally, however, this chatbot should not only contain general knowledge, but also internal knowledge – for example, by combining it with an enterprise search that makes the internal knowledge accessible – taking access rights into account, of course.
Such horizontal solutions allow employees to gain initial experience and build up an understanding of the technology. With this knowledge, companies can then invest in vertical solutions to optimize their processes. However, with the prior knowledge gained from an initial horizontal AI project, they have already created a stable knowledge base on which realistic expectations can be formulated.
Start with everyday challenges – not solutions
Almost every software provider currently writes “AI” on their product. And companies are buying it. In reality, however, software solutions – regardless of whether they are based on AI or not – only become successful if they solve problems. And this applies not only to the technical solutions, but also to the projects within the company. At amberSearch, we therefore make sure that we have a meaningful use case before we start working with a company. This ensures long-term success for both sides when working together.
The best way for companies to proceed is as follows:
- Identify pain points in everyday life – what processes and information do you want to optimize?
- Researching solutions – what options are there to solve these challenges?
- Decide on the best solution, taking into account the investment and complexity of the solution
If companies do not do this, they will fail because of the following:
- Frustrated employees: Every employee has more and more to do. There is no time for irrelevant side issues. If a new solution is introduced, it must improve the day-to-day work of employees. Only then will employees support it.
- Business impact: The budget holder will only support such a project in the long term if there is a foreseeable ROI and business impact. And the business impact usually determines the priority. If the challenge to be solved is not sufficiently prioritized, support will also fall away.
- Solution selection based on buzzwords: All of a sudden, companies start buying solutions because it sounds good and the provider tells a good story. However, solutions are then no longer purchased because of their business impact. Disappointment follows at the latest when it becomes clear that the promised added value cannot be realized.
AI is not an IT project
It is already clear from the points mentioned so far: AI is not an IT project and is not driven by IT. We often see that managing directors make the decision “We have to do something with AI now”. IT is then supposed to implement it – without knowing what the success criteria are and what the medium-term plan is. That’s why companies should have a digital strategy and/or even an AI-strategy.
In principle, however, IT departments are only service providers and enablers for the other business functions. The need for a solution should come from the specialist departments themselves. Otherwise, you are once again introducing AI without having a challenge and failing because of the things mentioned in the previous section.
A glimpse of what’s to come…
Many companies still lack the creativity to realize what is possible with AI. They have too little understanding of the technology and its hidden potential. So what will our world look like in a few years’ time?
Chatbots
Everyone starts with chatbots – partly in the form of ChatGPT, partly with internal company knowledge.
AI assistants
In the next expansion stage, companies rely on AI assistants, i.e. chatbots with specialized knowledge on a specific subject area. These assistants can then be asked for support in the same way as a secretary. However, AI assistants always act reactively based on knowledge that has already been defined in advance. This already works very well in 2024.
AI agents
AI agents will become relevant in the next expansion stage. In contrast to AI assistants, AI agents act completely autonomously and can make decisions independently – in predefined areas. This means they can perform tasks without a predefined process. Instead, AI agents are able to “derive” the process themselves. AI agents will certainly not act completely autonomously overnight and will initially act with the support of humans (e.g. by asking questions via a chat) – but AI agents will be useful in the medium term. Some of the AI agents will also act completely autonomously, while others will automate certain manual process steps.
Multi-agent systems
Where one AI agent reaches its limits, in future several AI agents will have to work together in so-called multi-agent systems, in which the respective predefined competencies complement each other. We have tried to record what such a future could look like in the following video:
Restrictions
However, it will still take some time before such use cases can be deployed in a scalable manner within the company. In principle, such AI agents always require internal “knowledge” in order to operate successfully. That is why an enterprise search is a mandatory prerequisite for the success of such systems in the future as a “knowledge base” for such AI agents. Modern enterprise search solutions such as amberSearch therefore offer 2 functions:
- A chat/search function for employees
- The knowledge base for AI agents in the medium term
We do not believe it is possible to have a meaningful 5 or 10-year vision – the market and technology are developing too quickly for that. However, we think it makes sense to set goals for the next 2-3 years based on the points mentioned here.
Establishing an AI-first mindset in companies
But how do you manage to not only recognize the technical possibilities, but also make use of them? To do this, companies need to work on their mindset. Companies need to think “AI-first”. This does not mean that everything should be implemented with AI by force (see previous chapter). But it does mean that every process must be constantly scrutinized and consideration must be given to whether AI is a possible, sensible alternative. And this thinking does not only have to be related to AI in the company, but can of course also be related to every individual.
AI-First is exemplified from the top
If you want AI-First, you have to set an example. And that starts with the management. If management is not convinced that AI is the future, then employees will not fully embrace it.
For AI-First to be successful, employees need to start thinking entrepreneurially. This often starts with the organizational structures: In many companies, entrepreneurial courage is not sufficiently incentivized by employees. If new solutions are proposed and they don’t work, the annual meeting tends to be unpleasant. If the solution works, employees are not sufficiently incentivized, which is only detrimental from the employee’s point of view. Therefore, it is not only the technical mindset that is important, but also the organizational framework created by the company.
Excuses & conclusion on AI integration
In Germany in particular, we often get lost in bureaucracy and miss out on sensible business decisions because someone says “it can’t be done”. A standard here that takes a critical view of things, a function there that doesn’t work and then there are all the costs… If you want, you can find excuses. For AI to work, however, you need to build up expertise in dealing with the technology. In this blog article, we have tried to provide as good and tried-and-tested a scheme as possible for integrating AI into companies and hope that interested companies can learn a lot here. Ultimately, AI only works with a combination of people and technology.
Last but not least – the more efficient AI becomes, the more we will use it. Where relatively few use cases for AI currently make sense, there will be more and more in the future. Just as companies are using more and more software solutions, they will also use more and more AI solutions. AI is getting better every day and it is up to companies to build up the relevant knowledge and use the added value of the technology sensibly.