Machine Learning is still a new technology for many businesses, especially for those that are planning to move their company and process to a more AI, ML -oriented organization.
Companies might not always be ready to have artificial intelligence solutions incorporated into their core processes without diligent preparation or careful planning. Moreover, while business executives are knowledgeable in devising strategic goals and managing the business, the understanding of artificial intelligence and its capacity from a technological standpoint might not be their strong point.
In order to help businesses adopt artificial intelligence projects and plan for them efficiently and make them work for your business, this blog post addresses how to effectively plan for an AI / ML project. This is an important topic because the outcomes of an AI project heavily depend on due diligence, proper planning, and assessment before the project starts.
The first area to be assessed is the existing infrastructure that your company has (including technology infrastructure, platforms, data infrastructure).
Your company might have the whole technology infrastructure on the cloud, on-premise, or hybrid; and depending on the specific problem, the AI solution for your use case might require little computer resources. But some AI solutions require more powerful servers with many CPU cores, memory, GPU/ TPU power. Questions to be asked can be: “What main programming languages are used for your tech stack and what does the tech stack consists of? What kind of API you can accept to integrate the ML API into the core technology? How much budget and resources you can invest in the AI initiative?…”. You may utilize and strongly prefer certain platforms and that will influence the choice of technology, platforms, approaches, and solutions for your AI project.
Out-of-the-box, low code AI platforms can help with low-code problems, usually common, rather simple data analytics / predictive analytics problems. Bigger problems will require skills beyond the limits of low-code platforms, more elaborate approaches, actual data science people planning, researching, experimenting, and implementing the customized solutions.
The answers to these questions will help identify what kind of solution you can build.
The second area is data. Your data infrastructure plays a huge role. Some AI solutions need big data, some only require small data. But the common denominator is Data Quality. Getting quality data is one of the biggest challenges in AI. Data Quality means the data should not be BIASED, it should contain predictive quality, actionable insights, it might need to be consistently formatted; because having unstructured data in heterogeneous formats (from pdfs, text files, SQL and NoSQL, image, logs, etc.) is not ideal.
Aim to have your data properly stored and processed (ideally in structured, at least searchable databases), have the data properly annotated if required; and sometimes, you might need to exceed the minimum data volume required for the AI models.
If you are not sure, try a POC project as the first step.
The third area is the people. You need the right talents. Inexperienced or good enough “data scientists/data/AI engineers” often lead to false starts, bad models that look good, and lots of wasted time and money. Apart from project managers, product people, you need AI engineers familiar with all the life cycles of a data science project: data engineering, data processing, building ML models, MLops, and fluent in building production-level applications. It is essential to hire the right people or AI vendors for your project. The right data scientist, AI architect with their expertise can help avoid the majority of wrong approaches, which can easily cost companies months or years of their investment. And AI investment is very expensive. Please remember a person’s qualifications and degrees have a low correlation with whether that person can deliver on an actual, commercial AI project. I learn that from my experience.
The plan and strategy itself
The fourth area is the plan and the strategy itself. Often strategy is created without a thorough understanding of AI capabilities against the organization’s current infrastructure. That usually leads to choosing the wrong, the infeasible initiatives that cannot be achieved; or the overestimation of AI / ML potential can lead to throwing-dart-in-the-dark approach, which leads to unclear business objectives, undefined success metrics, lack of clarity, and eventually frictions and project abandonment due to the lack of support from senior management. To avoid that, it is worth it to bring the tech and business people together, narrow the knowledge gaps and aim for more clarity in business outcomes and their connection with product features and technical capacity. Have experts review and supervision!
The final area is Risks. Risks in AI projects contain financial and timeline risks – so the project might go over the budget, cause delay, and extend the time to market. There are also the people and process risks, for example, lack of domain experts’ reviews and supervisions, lack of technical skills, overdependent on third-party platforms leading to no real ownership, vendor lock-in, and lack of control over the AI systems. Not to mention Trust, Ethical and Legal & Compliance risks with regards to the accuracy of the model, the bias of the results against minority groups, the explainability of the AI models, and the degradation of model performance over time or with unseen datasets, etc.
How to identify and handle those kinds of risks, I will address in the subsequent videos and blog posts of this series. So please keep an eye on our next videos. And in the meantime, if you need help with AI projects, contact us at Aigenexpert.com.