Artificial intelligence (AI) is one of the most used
words in the last few years, especially when talking about companies going through
digital transformation, changing their business models, etc. At the same time,
it is also one of the most misused words, but this will not be the topic of
this blog and we won’t go into by-the-book definitions. Also, inside that
digitalization jungle one of the most important topics is: Will “robots”
replace humans in doing their jobs?
On the one hand, this is not only a real problem which requires a complex process of preparing people for different jobs, but also a fun topic that media like to play with and, then, misuse. The idea of this blog is to share our experience of how collaboration between human and AI looks like in some of the industries. It doesn’t cover all potential use cases, but it’s going to give a clear picture of how people interact with new technologies with focus on consumer goods distribution industry.
So, to wrap this intro up, we aren’t going to speak about robotization of industries like automotive, or about digitally native companies that have already disrupted some industries, but instead we’re going to focus on companies and industries which are not digitally native and don’t have gathering and analysing data in the core of their businesses.
Why do human and AI have to live together?
To understand better what this coexistence looks like, let’s discuss the elements that AI usually supports human with.
1. We can take flexibility as the first component.
Production line in a modern automobile industry can support numerous different customizations one after another, because adjustments and optimization processes are run by various algorithms, and, therefore, much more flexible in comparison to humans and how fast they can adapt production lines for different car versions or types.
2. The second component is speed.
Probably the best example would be a real-time detection of fraud transactions. Take any of online bank operators and imagine what fraud detection on that number of transactions would look like with the use of modern algorithms.
3. The third component AI usually helps with is scale.
A great example of this is the AI-based screening algorithm that a company can use while tackling huge number of applicants for some job roles. For example, we can discuss Unilever’s introduction of AI-based system which allowed them to go through 30.000 applicants in just 4 weeks and extract only those applications which were most suitable for a potential job role.
4. The fourth component is decision-making.
Decision-making in general is obvious help that humans can get from AI, but it’s especially visible in the high-risk industries, one of the great examples being GE’s introduction of an AI-based system to detect predictive maintenance on airplane engines.
5. Finally, the last segment visible to all of us as end users is personalization.
Serving a huge number of users on the most popular web shops like eBay in a personal manner would be impossible without the AI-based smart segmentation and targeting algorithms.
Do not fear AI!
It is hard to introduce a systematic approach to such complex topics, for example, the implementation of AI system, but you know us, engineers - we always have to try. 😊
The picture below covers some of the key steps that companies need to take into consideration together with their implementation partner to ensure a project’s success.
The key steps of successful AI project
Now, while some of these components are self-explanatory, the other could have the whole blog just for themselves. So, in this one, we’re going to focus only on the three components marked in pink, because we see them as the most critical part of making sure that humans can live WITH AI after the implementation project, and we’re going to explain those using the example of a stock optimization project in the industry of consumer goods distribution. Also, these are the components that we put heavy focus on in our implementation projects.
Work constantly with end users – All the projects of software implementation have a lot of phases where we work with end users, such as testing, designing, etc. But, introducing a sales-forecasting and stock-optimization solution requires us to shift cooperation with end users to another level. In BE-terna we’re not fans of standardized AI apps which can serve all the needs, implying that we budget and plan a phase called “hypercare”, which basically means that we sit together with end users and analyse sales forecast together, and also discuss the order proposal that algorithms provide. Users have many questions in the first weeks of using a new system, and they have many doubts which is completely normal. In the end, we introduce a system which is supposed to make a more accurate forecast, and a more optimal ordering compared to before, and, of course, those users will be challenged, and they will try to understand where the numbers come from.
Simple self-service BI app to improve confidence – This last sentence brings us to the second really important component of our project. There are two endpoints of our system, the first being the automatic integration of proposed orders into ERP system so that user can just place an order, and the second being the BI application used for interpretability of results. Now, especially in the first weeks and months, it’s important for users to have a clear presentation of forecast trends, how system calculated forecast, what the constraints are, what’s the weather like, what the long-term order proposal is, etc. in order to make a person comfortable with accepting an ordering proposal from the system. Also, it is going to make the feedback from the user come more directly and quickly. During the first months of using a system, there are always many valuable inputs that end user will give to us (to a system) showing once again that this human + AI interaction really goes both ways.
Don’t change your everyday process for 180° – Finally, we come to the last component we’ve found out to be highly important for the success of an AI project in industries that are not digitally-native. The key puzzle-piece for this is component is our native integration with ERP system during which our platform populates, during the night, all orders for the vendors, which should be ordered by a definition in the ordering calendar. This fact significantly helps users in a way that their process is not turned upside down, but they just receive some extra time to focus on those highly specific items, such as tender items, slow-moving items etc.
Let’s wrap it up!
Everything we’ve discussed here will be different for every project, and it’s also much more the change of management than some simple tips and tricks. However, I’m sure that following these few steps will raise the odds for success of your AI project implementation, and make sure your end users really feel comfortable with new technology in their hands.
Now if it feels I’m talking about things that seem like a far future to you and you have no idea where to start, just hang on and wait for our next blog on this topic where we’ll discuss how to start an AI journey, the most typical use cases in FMCG industry, and the importance of knowing your data and the potential lying in them.