How to Scale AI
By Huw Kwon, Vice-President and Anil Eknath Sonavane, Senior Director
Addressing common obstacles — from delivery transformation to human-centric challenges.
This is the third instalment in a series on how organization’s can embrace AI specifically around conversational AI and hyper-personalization. The first issue discusses AI: Leveraging Superior Customer Experiences and the second focuses on Harnessing AI.
AI-driven analytics has briskly grown as a discipline over recent years. The COVID-19 pandemic has clearly sped up this trend, yet few examples where AI has delivered significant value at scale exist. In fact, 40% of AI projects don’t get off the ground: even after a hypothesis validation, most initiatives miss the bar and falter. What needs improving?
We cannot emphasize enough that business must establish clear goals before they ideate hyper-personalization and conversational solutions. What are the key aims, which executives will sponsor initiatives and what sort of budgets are sustainable? AI can be quite disruptive, as it involves change-management processes. So it’s imperative to define measurable goals first, then seek buy-in from the appropriate stakeholders.
Using human talent more efficiently is another area in need of improvement. Data scientists, for example, spend between 50% and 80% of their time scrambling and cleaning up data. As a result, businesses can lag behind on innovating because manual work bogs down their brain power. Data scientists are best at sifting through and making sense of data, and that’s what they should be devoting their time to.
Data scientists often toggle between spreadsheets and disparate tools to manage massive information flows. It isn’t feasible to develop successful algorithms without modernization — a sluggish system cannot handle high-volume, high-velocity data with accuracy. Companies with leading track records in AI credit cloud-based platforms for their achievements. Growth relies on equipping data scientists with the computing power that AI requires.
The field for applying AI is huge, with uses in as many as 500 business cases. Translating this potential into value entails, once again, defining clear aims. Without a data context that correlates with and supports the business goals, it won’t be possible to move forward. Having the right data, not all the data, is what matters.
Pulling everything will lead to paralysis at best and faulty modeling and validation at worst. And since customer needs and behaviors constantly change, inputs cannot be static — what worked this month could be irrelevant and useless next week.
Above all, an actionable AI data strategy requires a human-centric approach. Businesses should regularly consider how AI can: best interact with people, from enabling employees to supporting customers; be designed to register patterns in behavior and expectation; have enough “emotional intelligence” to adjust its decisions according to the people around it.
Realizing how to work with the right data for the right purpose is the bedrock of success. With that said, failing does not automatically spell disaster.
“Failure can be a vital part of the learning process. The trick is to fail and learn quickly to avoid wasting time, money and talent on unfeasible projects,” says Fabian Dupuis, Director of AI and Analytics for Cognizant.
As businesses look to scale AI projects, they need flexibility. Building a multidisciplinary team based on each customer use-case is essential. Besides speed and agility, such a team will focus on sustainability as much as outcome. Should issues pop up at any point, an agile, detail-oriented team can swiftly identify and address them. In the first phase, when validating, a data scientist plays a vital role. A dynamic team should also include a product owner, domain expert and scrum master. Ideally, the team during this phase comprises a data architect, data engineer and business consultant as well. This team make-up guarantees a smooth data-management process.
Each phase of an AI project, from validation and testing to scaling and deploying, demands different sets of expertise. Changing up team dynamics for each phase bakes resiliency into every step, ensuring the best outcome. This graphic illustrates how Cognizant’s “Pods” method works on client projects:
The core “DNA” of the Pods is combining digital expertise, deep knowledge of customers and fundamental delivery excellence. It offers a customizable KPI framework.
White Pod — They move the project from establishing the business case to creating the first algorithm. Their work includes assessing the data ecosystem and the one-off build of analytical records. While not building the final data platform, this pod validates sustainability and feasibility of a future solution.
Black Pod — They create a first MVP, bringing the algorithm to production. They also validate the business value and real-world outcome. Quite often, this pod reworks and refines the algorithm.
Grey Pod — With an MVP success, this pod scales and embeds the solution. They oversee the software engineering of the AI/ML model to make sure the solution is actionable from data ingestion through designing for user experience and adoption.
Addressing Common Challenges
Reaching maturity in AI is undoubtedly a longer-term objective for most businesses. How can companies see greater impact from their AI investments? By acknowledging why their AI spending is not translating into results, then tackling those issues. Here are the three most common challenges businesses face, along with our antidotes.
1. AI Leadership
The multidisciplinary nature of AI requires a broad set of skills to manage different aspects of its application. This brings issues in identifying the right people to lead AI projects. The tendency has been to put this responsibility in the hands of senior IT, digital or data science heads, rather than treat it as a company-wide endeavor.
AI is a cultural change, not an IT investment. The implications of AI span from the back office to customer interactions. These diverse perspectives are valuable for ensuring the application is solving real problems. Sharing the onus between different parts of the business guarantees accountability, and it incorporates necessary viewpoints around need and usability.
Place less emphasis on the tech, and put more on the “why.”
2. Regulatory and Ethics Concerns
Legislation such as the GDPR makes companies more aware of the reputational and financial risks associated with misusing data. Make it easy for staff to use data correctly. Use technical safeguards and checks so that data processing for AI adheres to regulatory requirements and maintains a high threshold for security and privacy.
Improve the overall quality of data on record, at a technical (coded into software) and process (employee training) level. Sidestep negative consequences by confronting the issue of bias in data, especially in areas like human resource and recruitment.
Support data compliance and best practices.
3. Data Infrastructure and Management
Before they can explore advanced data capabilities, companies must modernize their data infrastructure. This means putting hybrid cloud capabilities in place to make systems like neural networks widely accessible.
Cloud transformation is part of the AI process — hybrid cloud capabilities are needed to access the services and resources that sustain AI programs at scale.
Build a robust foundation or your data.
AI technology is most powerful when it collaborates with people, augmenting human activities and decisions. A human-oriented approach is key to ensuring that AI fits into real-world contexts. The following human-centric actions are critical to attaining sustainability with AI.
Grow your understanding of AI and build an AI strategy
Many companies lack a sustainable strategy for integrating AI. Further, the technology is constantly changing. Business leaders must strengthen their grasp of AI so they can develop clear-cut plans to leverage ever-evolving tools and data.
Prioritize company goals alongside responsible behavior
A rigorous AI strategy process starts with emphasizing value-creation and ethical behavior. That may sound like common sense, but many of today’s AI initiatives mistakenly focus more on technological capabilities and algorithms than on impacts and business benefits.
Experiment continually while applying learning to the next stage
There is no one-size-fits-all solution. Each AI challenge needs a unique approach, rather than a sequential process. Balance testing and measurement with risk-taking and innovation. Move on from failures quickly, yet be prepared to rapidly scale winning experiments.
Get your data right, then enrich it
While accurate data is vital, it isn’t enough. Businesses must bring in richer sets and types of data, such as psychographic, geo-spatial and real-time data. Digital maturity demands managing data well and realizing how to make it useful.
Solve the human side of the equation
AI is mostly about people. Hire tech-savvy talent who are aware of business needs. Make sure your team doesn’t stagnate in building models: create solutions. Consider HR hiring and retention plans — avoid disruption by securing the next generation of talent.
Kick off your own skills renaissance
AI requires new roles like big data specialists, process automation experts, security analysts, human-machine interaction designers, robotics engineers and machine learning experts. Besides hiring fresh talent, organizations must upskill their existing talent.
Construct workflows around performance thresholds
Forge the trust that makes human-machine teaming succeed. Prepare staff for profound shifts in how they will work. Without this foundational trust, staff will see machines as a threat to job security rather than a protector of it.
Promote collaboration and learning
Executives across functions — not just in IT — should cultivate a digital culture that motivates employees to use and apply new data services within their roles. AI has the potential to touch many parts of the company, so algorithms must “understand” the larger context in which they operate.
Navigate an ocean of possibilities
Reducing costs is a key benefit of AI. But there’s a vast range of other benefits. For example, AI can help improve product and service quality, reduce cycle time and enhance job satisfaction. See past lowering costs — stay open to exploring AI-enabled possibilities.
When it comes to bringing it all together, the imperative lies in forging the trust that makes human-machine teaming succeed. Share with us your thoughts.
You can download the ebook from here.
About the Authors
Huw Kwon is the Vice President — Global Head of AI & Analytics Strategy and Anil Eknath Sonavane is the Senior Director and European Global Head of Delivery. They work out of Cognizant’s London office.