While working with our clients, we have seen the transformational effect artificial intelligence (AI) has on customer experience, cost reduction and profitability. Considering the opportunities and advantages that AI delivers, it’s not surprising to witness its growing adoption globally. Results from Algorithmia’s third annual survey, 2021 Enterprise Trends in Machine Learning, showed that 76% of enterprises prioritize AI and machine learning (ML) over other IT initiatives in 2021.
However, we have also seen how AI deployments can run into headwinds. Executives start with many hopes and expectations but eventually struggle to put their models into production or ensure that the end users are actually using the intelligence to drive actions and impact. According to a white paper published by Pactera Technologies in 2019, about 85% of AI projects fail eventually.
Why is AI implementation challenging?
Based on experience, we have realized that most executives start with a “data science lab” approach to launch their AI project. In their minds, AI is about developing some ML models which one of their data analysts or data scientists can easily accomplish in a few months.
However, because this is a “lab” siloed from other key components of the entire AI implementation chain, soon they realize that what started as a few-months-long project with an OPEX budget now stretches into years in some cases and is not delivering actual value despite overshooting the budget.
What do AI plans miss?
Here are a few key observations about why some AI projects face deployment delays and budgetary overruns and fail to meet business goals:
The AI or data science lab approach is flawed.
A successful ML implementation requires all the talent and resources in place. This includes a data scientist to create and deploy models, a data engineer to align the model with IT, a developer to deploy logic, validate and test and a UI developer to present business insights. Then business users need to use the insights and make quick decisions daily.
Most companies do not have the necessary skills in-house. Even when an organization has all the resources, they may be aligned differently, functioning under different teams. Such a fragmented structure may lack coordination, resulting in inferior outcomes, increased costs and deployment delays.
My recommendation is the AI leader must set up a dedicated, multifunctional team to drive the AI project. This team can comprise data scientists, business analysts, an IT engineer, a data engineer and a full-stack UI developer. The team strength can vary based on project scope but inclusion of all these roles creates a synergy that is paramount for success of any AI project.
Enterprises should upskill their in-house talent and build an AI team with a long-term vision. They should expand and diversify their AI talent and explore how they can leverage their vendors.
Turning data into actions is not as straightforward as it looks.
Developing and deploying ML models and ensuring that they drive real actions require a considerable time investment and are an ongoing process. It involves multiple stages and many associated challenges. A few considerations here are:
1. Availability of high-quality training data. Data is often scattered across silos — unlabeled, biased and not consumption-ready. There should be a single repository of data and models should be continuously fed the up-to-date, accurate and labeled data.
2. Integration with legacy infrastructure. Integrating new AI systems with the existing infrastructure can be costly and time-consuming. Here, plug-and-play solutions and cloud-based applications are a big advantage.
3. Translating data into action. Many ML initiatives prove unable to convert data into actionable intelligence, which leads to compromised outcomes. It is vital to establish a clear relationship between data, insights, intelligence and outcome.
AI goals should be based on a deep understanding of business-user needs.
Though AI is about ML and automation, remember the key role of human judgment and human experience. Especially in use cases involving customer retention or customer care, think of augmenting human intelligence (of your front-line reps) with AI because some sets of customer problems are still not ready to be addressed by machines alone.
In those cases, how your front-line reps interact with the customer-level insights and use them to solve real customer problems is key to driving the desired business impact. If the intelligence is not delivered to the agents in a usable format, you won’t see the desired impact on metrics like NPS or customer retention.
AI strategy should benefit everyone and be aligned with long-term business goals.
In a hurry to join the AI bandwagon, businesses may want to choose convenient use cases, whether or not those use cases are aligned with broader organizational goals. Failing to prioritize projects may result in suboptimal resource utilization, significant opportunity costs and low ROI on AI/ML investments.
Another common obstacle is convincing the leadership and stakeholders. Stakeholders include everyone who the new technology will impact in any manner — shareholders, directors, manager or employees. Getting buy-in from everyone is essential and challenging. Even if the leaders and investors believe in the value that the technologies offer, others may be skeptical and resist the change.
Technology leaders must understand that these resistances are natural and educate leaders and shareholders on why the technological leap is required and how it will help. They can list and prioritize all possible AI use cases by their feasibility and business impacts in the short and long term.
It is advisable to start with simpler use cases and scale to more complex ones incrementally, which will ensure the adequate investment of time and resources and accelerate the organization-wide AI adoption.
AI has enormous potential to transform a business by driving efficiency and profitability. I have seen it unlocking millions of dollars in net present value for subscription businesses just from customer life cycle use cases such as retention, satisfaction and win-back. It is, however, a complex undertaking that requires cross-functional skills, a clear road map and alignment with business goals to deliver real business impact.