Over the past six months, I’ve had many personal meetings with potential and current clients and attended many IT exhibitions and conferences, particularly in Germany and the UK. I can’t recall a single booth that didn’t claim to be developing or implementing AI solutions.
But when I tried to dig deeper and understand what exactly they offer or have implemented, the answers were often vague and generic. In many cases, it all came down to simply using the API from OpenAI to work with ChatGPT.
If your product doesn’t mention AI today, does it make you a failure? History will decide.
Yesterday, I attended a startup pitch night where a fellow angel investor said something I agree with: “Stop bombarding me with AI; show me the product!”
To me, this reflects the current situation of the industry. Instead of developing the necessary functionality for their clients, which will drive revenue, many companies think about what AI feature to implement just to stay on trend…
This often leads to misunderstandings, particularly in our field. We frequently meet a client who wants to implement AI because “it is a must, and everyone is talking about it”, yet they do not fully believe that their investment in development will ever pay off.
With this article, I want to start a series of materials showcasing real-world cases of practical AI implementations. My hope is that these examples will inspire readers to develop AI features that truly add value to their products.
AI – though I feel like I’m developing an allergy to the word and acronym – is neither magic nor something entirely new.
These algorithms were developed long ago but have become so popular today thanks to advances in computing power (servers) and, of course, the rise of ChatGPT. In 99% of cases, AI implementation comes down to either using third-party solutions or training and developing proprietary DeepML models.
So, what truly useful AI-driven features can we create for our products? Of course, each product is unique, but there are common patterns and implementations. I’m not aiming to provide an exhaustive list – nor could I – but I hope this will spark ideas and be useful. I will list the software types and domains we have encountered at Softacom.
Office software such as CRM, ERP, database applications, desktop software, SaaS, and more.
- Initial filling and customization of the customer profile using data generated by a generative model trained on a customer database based on previous manual customizations or configurations by an expert system. In this case, it is better to use your own trained model.
- Summarizing texts or documents to get a summary, highlighting the main points and core meaning. API solutions from OpenAI, Google, Microsoft, and others can be used for this purpose.
- Filling out questionnaires, provision forms, and other documents with AI-generated content, followed by human verification.
- Creating a personalized message for a lead for marketing purposes, using data about a potential customer (lead) from your database or from third-party sources.
- Generating text for messages and other content using local generative models or services like ChatGPT.
- Generating images or photos based on specified criteria or training your own models with your custom image datasets, e.g., creating images in your own style for photo libraries.
- Processing images and photos according to predefined templates.
Software with video analytics and video processing functionality or any other solutions for computer vision and pattern recognition.
- Training a local neural network, such as YOLO, to detect specific objects in a video stream or photo and relay the information to a decision-making system. One example is detecting a drone in an airport area.
- Conversely, detecting objects that fail to meet the required criteria, such as the absence of a helmet on a worker at a construction site.
- Training a neural network to detect specified items in X-ray images, for example, to identify drug capsules in prisoners’ bodies or search for prohibited items in passengers’ luggage at an airport.
- Quality control of machinery parts by analyzing photographs on a production line to identify defects, cracks, or damage.
Security software:
- Everything described above is for computer vision solutions, including detecting people, weapons, drones, vehicles, and other objects.
- Face recognition and license plate recognition.
- Fire detection from video and photo, along with smoke detection.
- Detection of crowd formation.
- Analysis of virtual line crossings and control zone monitoring.
- Training a local neural network in real time to recognize standard employee behavior and detect non-standard actions. Such implementations are also applicable to banking and other financial software.
The most interesting thing is that almost all of this functionality could have been fully implemented without relying on what is now commonly referred to as AI. Expert systems and custom software, which implemented the necessary mathematical algorithms, were perfectly capable of handling the tasks at hand. But neural networks, machine learning, and third-party cloud services have significantly expanded our capabilities and provided new approaches to solving the aforementioned tasks.
We will periodically update this article and add to the list.
This article was written by humans, for humans, and reflects the experience of our company. ChatGPT or DeepSeek could likely offer more ideas if you share details about your product with them.
