Working with many founders and managers has taught us how important industry knowledge is to the success of AI projects. Unfortunately, experience with mobile apps or web-based SaaS is not enough to successfully run a project. We discuss the most common mistakes below. 

Failing to understand the research nature of an AI project

AI projects tend to be exploratory. Their implementation requires a trial-and-error approach, and no guarantee can be given for the expected results. There are many reasons for this: overly ambitious goals, mistakes during implementation, and incorrect formulation of the problem.

Progress (incrementalism) in a project is the verification of hypotheses and gaining knowledge about which direction to go. Even a negative answer is valuable knowledge that forces a paradigm shift and poses further questions that bring the project closer to completion.

Incorrectly collected data

What's the problem with collecting data? - The backend team writes a crawler that collects data, measures the app, and installs physical sensors. We wait a year and proceed to teach AI. After two months of work, we realize the implementation and the product is ready. Is that right? Of course not!

What went wrong? Well, a great many factors, to name a few: lack of regularity in the control of the data collection system, a bug in the crawler that didn't work as expected for months, a sensor failure. After fixing the failure and potentially controlling the fire, the backend team failed to pull all the relevant data. And the non-technical project manager, unfortunately, was unable to assess it.

Failure to actively engage data analysis stakeholders from the moment of project planning results in errors and an exponential increase in time and cost.

The belief that AI can do anything

Contrary to the pop-culture narrative that often places AI as an almost limitless technology, artificial intelligence is not a one-size-fits-all, always-effective method of solving problems. An AI specialist can verify at the very moment of planning whether the project being undertaken is a routine activity or a research topic being worked on at universities. Assessing the risk of project success will certainly help in making key decisions.

A good advisor will pay attention to whether the use of AI is necessary. For many years, computer science has excelled with classic algorithms. Maybe it is enough to use one of them?

In summary, the three most common mistakes are also key moments for project success. A properly estimated budget, an emphasis on data collection, and a critical approach to the method can determine the effectiveness of an AI project.