Project management in AI-centric projects | by Aydın Fevzi Özçekiç | May 2022

Project management is a critical process for project success. Big Data, ML and Business Intelligence projects have different priorities and challenges.

Even simple machine learning projects can be complex and managing these projects in a real business is harder than most people realize; that’s why VentureBeat claims 87% of machine learning products never make it to productionand Harvard Business Review says that “The first wave of enterprise AI is doomed.”

FOCUS ON THE PRODUCT NOT ON THE AI

AI is trending and when you use “xxxx powered by AI”, people are more interested in your project. The popularity of AI is causing people to develop useless products.

The product is more important than the AI. Understand your customers’ needs and develop the easiest solution for them.

Deterministic to probabilistic approach

AI systems differ from traditional software in many ways, but the biggest difference is that machine learning shifts engineering from a deterministic to a probabilistic process.

With AI, you often don’t know what’s going to happen until you try it. It is not uncommon to spend weeks or even months before finding something that works and improves model accuracy by 70% to 74%.

The main difference between classical and ML programming is in the inputs and outputs. So you should consider which approach is most applicable to your problem.

Is it possible to create an MVP without ML?

Don’t be afraid to launch a product without machine learning.

Machine learning needs data and the result will be probabilistic. So, at the MVP stage, starting with small data and low accuracy can create dissatisfied users.

My advice starts with rule-based solutions and develops step-by-step ML in the background. When you believe ML results are better than traditional rule-based systems, you can turn the solution into ML. On the other hand, to reach more data, you need a functional and attractive MVP. So, try to use more mature technologies to get results.

If your project is part of a group described below, do not start with ML.

My advice

I worked on ML projects for several companies. My suggestions:

1- Experience: You use different ML models to solve a problem. Therefore, you should experiment. Unfortunately, sometimes you cannot achieve an acceptable result. So, in AI-powered projects, you need to develop a detailed risk mitigation plan, especially for data and ML issues.

2- Focus on customer needs: Don’t get obsessed with using ML. Be practical and develop the most useful solution. Sometimes it can be a simple function instead of complex ML works.

3- Seek advice from ML Experts: Take advice in the planning phase of the project. Sometimes you can create high expectations regarding the implementation of AI. If you are not aware of the limitations of technology, you may experience problems in project management. Remember that theory and practice are really different in ML projects.

4- Use Cloud Infrastructure for ML model development: The cloud is a big advantage for AI projects. You can use prepopulated ML models, compute power, and data pipeline without requiring maintenance. So, select a cloud provider to implement your data project.

5- Work on the data, digest your data and make a data management plan: ML projects start with data. If you don’t have a pipeline of data and business intelligence projects, it won’t be easy to develop a unique ML-powered solution.

6- Do not use ML without a proper plan: In ML projects, establish a clear outline of your workflow. Otherwise, your project begins to be experimental work and you cannot manage your resources.

7- Do not start with insufficient data: To finish on time, don’t launch your product with a small set of data. Then your ML model makes drastic errors and your project loses reliability

8- Use ML to help with data collection: Data collection will be your biggest problem in ML projects. So use pre-built ML models for better data collection. Please look for ML solutions such as image markup, prediction of missing values ​​in a dataset, natural language processing, and anomaly detection to enrich your dataset.

Botmore Technology

We support companies in ML project plans with really affordable prices. Please email [email protected] for details.