Artificial intelligence (AI) and digitization promise a world of optimized productivity, greater efficiency, and improved profitability. AI can handle tasks at a pace and scope that people cannot match. Plus, once these repetitive, mundane tasks are handed over to AI platforms, valuable human resources can take care of more high-value tasks. So much so that - according to a recent Deloitte survey, 82% of early AI adopters cited a significant financial return from their AI investments.
Most sectors and industries benefit from AI adoption and projects, with a recent McKinsey report showing respondents in the manufacturing and risk management sectors, in particular, seeing high-value returns in AI adoption.
Figure 1. Reported AI benefits by sector. Information sourced from McKinsey & Company.
Why Collaboration is a Critical Part of Ensuring AI Projects Succeed
“We have many years of experience working with organizations who want to use AI for greater business insights,” says Anthony Damian, Director of Solverminds - an international AI and data analytics solutions provider. “In my experience, organizations that struggle to get the most out of their AI deployment, or worse, fail to launch their AI initiative entirely, demonstrate these similarities.”
1. Challenge: Lack of strategy
When an AI project is first planned at a strategic level, it may not address the core business objectives and ultimately struggles to find a place in the organization’s broader or more comprehensive plans for growth and business development. “For an AI project to work,” says Anthony, “it needs buy-in and input from the uppermost levels of leadership first.” Thus, an AI project is an opportunity to collaborate with all key stakeholders, guided by a clear strategy from leadership.
Solution: Work with the end goal in mind
What is the main objective for deploying an AI project? Articulating this is the only way you will know if it has been a success or not. For instance, do you need AI to help with customer-facing processes and tasks or non-customer-facing processes and tasks? AI is best deployed for repeatable, non-customer-facing work.
“The best way to keep the AI project on track and drive it to completion is by insisting on clearly defined goals from the start. These goals could be to use the data, AI, and data scientists to minimize risks, lower operational expenses, and / or highlight opportunities that improve profitability,” says Anthony.
2. Challenge: Unsure of what AI can do or why it is needed
Many leaders feel the pressure to adopt AI and start AI projects for the clear benefits it offers. However, very seldom is any meaningful time or resources given to first research and explore how these AI platforms work and apply to their real-world, practical business operational needs.
“Leaders know what their challenges are,” says Anthony. “It is imperative that business leaders know if and how AI can solve their challenges, so that realistic targets are set, and measurable outcomes are defined at the onset.”
Solution: Source the required skills
This can be an outsourced resource or employee, but data and AI tools are best put in the hands of data scientists and data analysts. Data scientists are currently a very rare resource, and they can be difficult to find (and keep) and are expensive as a team. Outsourcing this to specialists can be a more affordable, successful solution.
Figure 2. Data scientists are currently a very rare resource and can be difficult to find…
3. Feeding it the erroneous data
It has been said before that AI does not exist in isolation, and this is true. AI is data-driven, and it needs plenty of it. “Not knowing what data to use, where to find it, how to clean it, or analyze it is a major pitfall,” says Anthony. But, more than that, data omissions are as impactful as erroneous data.
“We were approached to carry out an AI project for a large international concern,” recalls Anthony, “intending to improve approval rates for their operations. The data scientists were unaware of all the nuances and influencing factors specific to the industry that would result in a higher approval rate and were working solely with the data provided. Only once we started asking for this niched data did the projections start adding real business value.”
Solution: Get the right data and the data right
AI needs data like we need air. While the volume of data is important, the quality of data present will reflect the value of insights and predictions your AI tool provides. Ensure there is a data strategy in place to extract, clean, and analyze data at any point. It is also important to have a sufficiently sized data set to train the AI’s machine learning.
Utilizing a resource who knows the industry, knows what data is required, knows what questions need answering, knows what influencing factors are at play, is as important as choosing the right platform.
4. Challenge: Building when you can rent
Organizations often ‘go for broke’ and try to launch their own AI solution to meet their needs. ‘Renting’ your AI solution is a more affordable, faster, and more accessible option than building AI into your organization from scratch. AI tools like SEDGE work on a subscription model that allow you to use it for as long and as often as your organization needs for a fraction of the cost.
Solution: Find the best technology for your business needs
Finding the right AI tool and platform are critical. Building a bespoke tech environment can be tremendously expensive, where there are flexible, customizable AI solutions available to take care of the bulk of business data requirements. This reduces time to market and cost of entry.
5. Challenge: Data protection issues
Security risks, such as personal data protection and data security, are critical factors for all organizations, including those who seek to use vast amounts of data with their AI project.
Solution: Embedded privacy and security compliance
It goes without saying that the AI solution you use must be GDPR compliant to support your data privacy commitment. It is important to be mindful of the use of personal data in the project. For instance, does the data being uploaded have any special category fields, such as data revealing gender, race, ethnic origin, political opinions, religious, genetic, biometric data processed for the purpose of uniquely identifying a natural person etc.? Be aware of the people accessing the data, and collaboration with team members and also have clarity on how long the data will be stored. To secure the personal data, the data processing AI tools should have the ability to pseudonymize the sensitive fields, or if these fields are not relevant, then ignore or exclude them from the model building.
6. Challenge: Leaving it alone
“It is important to monitor the model in the production environment, as models need to be refreshed from time to time to prevent model drift,” says Anthony. “Leaving an AI model in production unmonitored is not recommended.” If left unchecked, the model may face model drifts and the predictions can slowly lose their accuracy overtime. This will produce incorrect outcomes, which may be due to data drift or concept drift.
Solution: Monitoring the model drift
Monitor the models in production for model drift. The actual outcomes versus the model predicted outcomes need to be compared and accuracy identified on a regular basis. If the accuracy is waning off, then the model needs to be rerun on fresh data, and the new model accuracy needs to be determined. Then, if the accuracy is good, the new model should be deployed back in production and the model monitoring should recommence.
7. Challenge: Lack of Project Management
At the onset, most AI projects start with a clear focus on the customer requirement. However, as the project moves ahead, due to challenges in extracting data, or due to challenges in the quality of data, the original project charter is set aside and the project veers off course. This results in the original purpose of the AI project being lost somewhere along the way.
Solution: Applying Project management best practices
“A project is a project,” says Anthony. “Whether it is building a house or deploying AI, these fundamental project management skills are absolutely necessary to manage an AI project. This includes managing customer requirements, understanding the project success criteria, understanding domain and data, building a detailed project plan, executing that project plan, monitoring the team, watching deadlines, overseeing workloads, executing effective change management, watching costs, effective communication, training, and management reporting, if the project is to be a success.”
8. Challenge: Getting it to work
Implementing AI is more than just deploying AI technology in a silo and hoping the organization is now ‘digitized’. It requires a team with industry knowledge, in-depth AI acumen, project management skills, as well as specialist skills such as data science.
Most organizations simply cannot afford such a highly skilled team and abandon the project without exploring viable, affordable outsourced options.
Solution: Domain knowledge is non-negotiable
You can have the best AI tool and outsource top data science skills, but you will get the most from both only if there is a domain expert on the team. “We don’t know what we don’t know,” says Anthony. “A domain expert will ask the right questions, insist on data segments that others would simply not know to ask about, and direct the data scientists towards solving the business problem that the data and AI can work towards.”
Get Your AI Project on the Right Track with SEDGE
Whether you have a complex organizational challenge that needs AI and data scientists to solve, or you need AI to give you immediate insights into your data, SEDGE has the skills, the domain knowledge, and the AI solution that gives you the competitive edge.