How to succeed with your AI project

The threshold for starting to use AI solutions has been significantly lowered since the launch of ChatGPT from Open AI. But if you as an organisation want to integrate AI into your existing solutions and systems and work with your own data to improve your business, how do you do it?

"AI is so much more than ChatGPT," says Stella Riad at RISE. "Organisations often do not understand what data they have, what value it has and how they can make better use of their data.

Data can be anything from logs from the web, machines or devices to emails, support messages, chats or documents. It can be used to free up time, streamline and improve processes - and even create new business.

To get your first AI project off to the right start, and to minimise the risk of wasting money, you need an effective strategy.

"Address a real challenge in your organisation so that the first AI project tackles a real problem," says Stella Riad. "Then it will be worthwhile, and it will be more natural to give the project a budget. Ask questions like: "What are the pain points in our business?", "What is important to solve?", "Where can we create real value?". She gives some examples:

• Repetitive routine tasks. Every week an employee has to spend three hours digging up the right information to move forward.
• Product uncertainty. Which product should we produce next? Which one has the best chance of success?
• Predictive maintenance, making sure that interventions are more accurate, thus saving time and money.
• Dynamic manuals that are updated based on data from your systems.

"Give each pain point an approximate price tag. What does it cost in terms of money, labour and lost potential opportunities?

Select a limited project

Once the homework of identifying pain points is done, the work of finding a suitable first AI project can begin. This is where many companies turn to RISE for help.

"Together, we look at the list of pain points and try to find something that is simple and limited enough to be carried out in 4-12 months," says Stella Riad. "We want to avoid spending 24 months on a big project that doesn't turn out well and has cost millions. It's better to start with something small, to minimise risk and to prepare the organisation for new technologies. Even in a small project, you learn what you can do with your data, while building trust from both sides for larger projects later on. Smaller projects also make it easier to change direction, if necessary. We want the project to result in something concrete, and then the delimitation is crucial.

Does useful data exist or does it need to be generated?

It's not just the scoping that takes some time at the beginning of a new project. An important variable that affects the nature of the initial project is what data is available, related to the pain point you want to address. AI needs organised and annotated data, both to train its intelligence, draw the right kind of conclusions and generate relevant texts, graphs and perform different types of tasks. It is also important to have data that can answer the questions you want to ask:

"What data exists, and is it in a format that can be analysed? Should more data be logged? Or a different type of data?" says Stella Riad.

These initial discussions also include more in-depth exploration of different aspects of the problem to be solved.

"The problem space discussions tend to be valuable in themselves for companies. They are forced to explore important issues that accelerate their business development."

The discussions also lead companies to ask again and again: "Does it work?", "Is it better to do A or B?".

"The goal, of course, is to find a way forward that strikes a good balance between benefit and effort."

Implementing AI requires work

Using AI also means a commitment to provide the computer with ongoing feedback. When does the computer get it right? When does it get it wrong?

"The AI develops by getting feedback on what it delivers," says Stella Riad.  "It is an investment that needs to be maintained."

For example, unexpected distortions in the data may begin to emerge that cause the AI to make strange conclusions. A train derailment in Gothenburg may lead to many incoming emails suddenly mentioning this, but that in itself does not necessarily mean that Gothenburg is more interesting to customers in general.

"When you start working with AI, you need to take into account that it may require some effort as well. We want to help companies understand what they can and cannot do with AI and their data. There is no better way to boost your AI expertise than to start with a first, small, concrete AI project."

"Most often, we start by optimising an existing flow. It is during this process that companies discover how they can start thinking more AI. They are forced to think carefully about the collection and management of data, and become better clients and buyers in the process. They begin to realise that AI could redefine the way they work, or even disrupt their niche or industry, and create new types of offerings. The important thing is to get started."