CIOs and business leaders know they are sitting in the gold mines of business data. Additionally, traditional tools such as business intelligence platforms and statistical analytics software can effectively express insights from matched data resources, but real-time, rapid, and rapid, is an unresolved challenge.
Enterprise AI can turn these bottlenecks into opportunities when responsibly deployed at large scale. Acting data quickly, and even “live” (while customer interactions), is one of the capabilities of technology, as well as scalability. AI can process large amounts of information from heterogeneous sources as easily as it can summarise a single-page spreadsheet.
However, deploying AI solutions to modern companies is not easy. You need structure, trust and proper talent. In addition to the real-world implementation challenges, AI poses unique challenges, including data governance, the need to impose guardrails on AI responses and training data, and the problems of sustained staffing.
We met with PWC principal and Technology Managed Services (AI, Data Analytics, Insights) Rani Radhakrishnan to talk about what is working and what is holding back the CIOs on our AI journey. We were prior to her speaking engagement at Techex AI & Big Data Expo North America at the Santa Clara Convention Center on June 4th and 5th.
Rani has been particularly adapted to some of the governance, data privacy and sovereignty issues facing businesses, and has worked with numerous clients in the health sector for many years. This is the area where privacy, data monitoring and, above all, the accuracy of the problem is the make or break aspect of technology deployment.
“It’s not enough to have a quick engineer or Python developer alone. …We need people in the loop to curate the right training dataset, check and address output bias.” – Rani Radhakrishnan, pwc
From support to strategy: A shift in AI expectations
Rani said there is growing enthusiasm from PWC clients, both AI-driven managed services that can provide business insights in all sectors, and technology that can be used more aggressively, with so-called agent roles that allow agents to act independently of data and user input. If an autonomous AI agent can take actions based on human interaction, access to data resources, and automation.
For example, PWC’s agent OS is a modular AI platform that scales the system into workflows and connects intelligent agents to the workflow at more speeds than traditional computing methods. This is an example of how PWC responds to AI demand from clients, many of which see the potential of this new technology, but internal expertise and staff do not act on their own needs.
Depending on the sector of your organization, interest in AI can come from different locations in your business. Active surveillance of physical or digital systems. Predictive manufacturing or engineering maintenance. Or, just a few examples of cost-efficiency gained through automation in complex, customer-oriented environments.
However, no matter where AI can bring value, most companies still don’t have the skills and people within the company that are needed to deploy effective AI.
“It’s not enough to have just a quick engineer or Python developer,” Rani said. “We need to put all of this together in a very structured way, but we need people in the loop to curate the right training dataset, check and address output bias.”
Cleaning House: Data Challenges Behind AI
Rani said effective AI implementation combines technical skills such as data engineering, data science and rapid engineering in conjunction with organizational domain expertise. Internal domain expertise can define the correct outcomes, and technical staff can cover responsible AI practices such as data matching and governance, ensuring that AI systems work responsibly and within company guidelines.
“To get the most value from AI, organizations need to get the underlying data right,” she said. “I don’t know of a single company that says the data is very shaped. You need to enter it into the right structure and get it right, so you can query, analyze, annotate and identify new trends.”
Some working companies need to put in for effective AI use. This is observation and correction of biases both in the output of the AI system and in both the analysis of potential biases specific to the training and operational data.
As part of the underlying architecture of AI systems, it is important for teams to apply stringent data sanitization, normalization, and data annotation processes. The latter requires “many human effort,” said Rani, and the necessary skilled personnel are a new kind of data expert that is beginning to emerge.
If data and HR challenges can be overcome, the feedback loop makes possible results from the generated AI very valuable, Rani said. “Now there’s an opportunity for AI prompts to go back and improve the answers you get. That’s what makes it very unique and valuable to train the model to answer questions in the way you want them to answer.”
For CIOs, shifting is not just about technology realization. It involves integrating AI into an enterprise architecture, collaborating with business strategy and managing governance risks associated with scale. CIOs are becoming AI stewards. It builds not only systems, but trust and change.
Conclusion
With AI only a few years since it was born from the roots of academic computer science research, it is understandable that today’s corporate organisations feel to some extent to achieve the possibilities of AI.
However, a new playbook is coming. This is what helps CIOs access the value held in their data reserves, business strategy, operational improvements, customer-oriented experiences, and more areas of business.
As a company immersed in experience with clients large and small around the world, PWC is one of the key choices for decision makers to begin, streamline and direct their existing AI journey.
Find out how PWC helps CIOs embed AI into core operations and see Rani’s latest insights at June Techex AI & Big Data Expo North America.
(Image Source: “Network Rack” by one individual is licensed under CC BY-SA 2.0.)
