Artificial Intelligence (AI) is coming of age, quickly moving from the pages of science fiction novels to customer interactions and back office operations of organizations in nearly every industry worldwide.

While it’s clear that company executives recognize AI’s potential, many are still challenged when it comes to effectively integrating it into their organizations. According to a 2019 study from Harvard Business Review and EXL, only 8% of companies have scaled AI deployment, and 27% are at different stages of deployment. 57% are still in the planning to pilot stages of AI implementation.

The reason? Unlike other enabling technologies that humans use to work more efficiently and effectively, AI only delivers tangible value when it orchestrates people and machines into a cohesive team. Successful AI deployment involves much more than launching technology to provide access to users. In addition to changing the way your team works on a daily basis, you also must align their mindset, their management and the way in which their work is measured. The gains that result from this orchestration approach are undisputed.

The insurance industry is already using AI to price, purchase and bind policies in near real time. Moving forward, by harnessing and analyzing the data from connected devices, companies will evolve from a model of “detect and repair” to “predict and prevent,” reducing costs, improving safety and providing a better customer experience.

Retailers are not only applying AI for suggestive selling, based on a customer’s previous purchases, but using physical robots in store to scan shelves for low inventory, or price tags that need replacement.2

When applied to healthcare, AI could be used for quick disease detection. Stanford researchers have already developed an AI algorithm that detect skin cancer from medical scans, with a 91% accuracy rate.3

These examples and myriad others illustrate that organizations that strategically apply, integrate and orchestrate AI in their operations are positioned for the future. Those that don’t will be left behind.

In this paper, we explore the best practices of an AI orchestration—the stages, the considerations and the cultural shift required to ensure humans and machines can work side-by-side as peers.

Approaching AI in Stages

The optimal way to approach AI implementation is to break it into four distinct stages: Envision and Define, Solution Orchestration, Operationalization, and Shaping and Scaling for the Future.

STAGE ONE: ENVISION AND DEFINE

Every AI initiative begins with clearly defining the business problem driving the initiative. Is it a problem within a particular process? Or, are you solving a market problem and creating a digital solution for the industry? Defining the “what,” the “why,” and then quantifying the outcomes is critical.

One of the biggest mistakes organizations make at this stage is trying to do too much, all at once; applying AI to multiple, end-to-end tasks from day one. Because the machines, like humans, have to be taught how to perform their functions, companies will improve outcomes by starting small, with basic functions, and then, as the machine learns, expand the breadth and scope over time. This approach, known as Practical AI, not only improves results, but also gives staff members time to adjust to, and embrace, the change.

For example, email is still a vital customer service channel for many industries. The faster and more accurately these companies respond to the emails, the higher the customer satisfaction rating. However, attempting to train AI to sort all the different types of incoming email at one time will not produce the desired results.

During the Envision and Define phase, organizations wanting to streamline email should identify the area in which faster email response would produce the maximum gains and focus efforts there. For insurance and healthcare organizations, that focus could be using AI to more quickly identify and respond to a claims segment.

Finance and accounting areas could apply AI to supplier management, to monitor any supplier that falls out of contract compliance.

Long-term AI strategies are vital, but, the best results come from narrowing that vision so execution can occur in an iterative, agile manner.

STAGE TWO: SOLUTION ORCHESTRATION

In this stage, stakeholders need to consider the real-world factors impacting their AI vision: infrastructure, data and talent.

Infrastructure

Training deep learning models for image, text, speech or video requires tuning thousands of parameters in an iterative manner. These models demand a lot of computing power from hardware while training. Commodity hardware, such as general purpose central processing units, are rarely sufficient to train such models. Some AI applications require specialized hardware, such as GPU servers, for their development and deployment.

So, organizations have to assess whether they have the digital technologies, infrastructure and computing power in place to support the AI framework they’ve envisioned.

Data

Another important aspect of successful AI solution implementation that is often overlooked is the availability and accessibility of data. The necessary information, like client contracts, health records or sales figures, could be siloed and difficult to access.

So, it’s critical to identify the required data and the sources of that data upfront. Then, partner with the owners of those data sources for access, putting controls in place to ensure adherence to all statutory and regulatory guidelines.

For example, healthcare organizations are benefitting from using AI to summarize medical reports for critical claims processing. This initiative requires a thorough understanding of patient medical records as well as eligible medical benefits, so data must be securely pulled from multiple different sources before being run through the algorithm to generate the summarization. Without data access and the infrastructure to support rapid analysis and throughput, the initiative will not produce results.

Talent

If the company is considering implementing a packaged AI solution, along with partnering with a solution provider, its executives also have to assess whether or not the organization has the AI talent in house to effectively execute the initiative, and take it to maturity.

The reality is, most don’t. Experienced data scientists and machine learning engineers are in high demand but short supply. So, recruiting and retention are both challenges, particularly for organizations at which AI or digital solutions are not a core business.

In a recent study EXL conducted in partnership with Harvard Business Review Analytics Services, 48% of the 800 executives surveyed identified “Difficulty finding analytics talent to build, deploy and maintain AI systems” as their top AI orchestration challenge.

Addressing infrastructure, data and talent issues upfront, along with adopting agile development practices, are key to a successful, sustainable AI initiative.

STAGE THREE: OPERATIONALIZATION

After the solution is developed, it’s time to operationalize, to ensure it solves the business problem or delivers the desired outcomes. A number of factors come into play here:

Determining Method of Execution

One big consideration is determining the best way to roll out the solution. Will it be a big bang or will it be integrated in methodical, controlled iterations?

Although it may be tempting to go all-in at the onset, the best outcomes, particularly if an organization is new to AI, come from taking the “sprint” approach—making small changes, seeing the results, and then using those resources to go forward with the next phase of the initiative. This approach enables companies to accelerate their operationalization of AI solutions and realize business outcomes.

For example, client due diligence is a significant challenge, particularly for larger banks. Before onboarding new account holders, they have to scrutinize their financial health, and identify any suspicious activities that could be related to money laundering or defaults.

One large global bank used AI to automate a significant portion of the due diligence but, instead of a full global rollout, it chose a more strategic, phased approach. After starting with a pilot in the U.S., it slowly expanded to multiple geographies after incorporating local compliance requirements.

This approach enabled the organization to ease users into adoption, document some success stories and tweak the solution before a worldwide rollout, ensuring the best outcomes.

Creating a Change Management Office

Because AI solutions involve many disparate teams, having a regimented change management program is key to not only executing the solution but achieving the desired outcomes. Organizations need a change management program to coordinate between the sponsoring business unit, the internal talent, and the providers developing the AI solution, as well as overseeing the business case and economics of the change. Everything has to be a coordinated cohesive effort with formalized oversight.

Making the Base Solution Reusable for Other Applications

Although organizations can choose to view each AI component as its own entity, starting “fresh” and going through the full development cycle each time, this stifles agility and wastes development resources.

By creating a set of reusable AI components that are replicable and relevant to other parts of the operation, you can maximize the benefits of your initial work in multiple areas across the business unit.

Instead of building out a one-time solution for one specific purpose, the AI transformation organization work on the principles of transfer learning by building a set of templates with algorithms and machine learning components that serve as the foundation for change. So, every time that a business unit needs an AI solution, the team can use this set of components as the foundation and build from there, instead of starting fresh with each iteration.

For example, if a company builds an AI module to read an email, the development team should identify which other functions this same skillset applies to, like reading a claim or customer inquiry. The idea is to augment the machine’s base knowledge and competency, and apply this to other functions to benefit multiple areas of the organization.

If an insurer utilizes AI to streamline policy change requests for automobile policies, those algorithms can be leveraged to other lines of business, as much of the foundational work is already done.

It’s very much like putting an experienced staff member in a new position, as opposed to bringing on a new hire. Because that staff member has a base of knowledge, and understands how the company operates, he or she will typically gain proficiency in the new role much faster than the new hire.

  • Are the machines or algorithms being fed the information and feedback they need to continually become more intelligent and productive?
  • Is the overall solution fulfilling the projections presented in the original business case?

Along with milestone tracking, companies need continuous mechanism for governing and monitoring the output. To this end, some providers have created a digital hub or control center that gives company leaders an ongoing, real-time view of the operation. They can see what the human staff and the digital workforce are each accomplishing, whether the operation is getting the desired throughput, and, if not, identify the root cause of the issue. How effective is the feedback loop for the machine to learn and generate more accurate outputs? Has the machine learned enough to move more work through without human intervention, or are the bulk of the tasks still going through people? Are the algorithms performing? Has the digital worker been involved in the operation or has it been shuffled off to the side while the humans conduct business as usual?

Putting a Governance Mechanism in Place

In addition, organizations should put a separate governance mechanism in place to conduct periodic reviews to ensure the solution continues to deliver value, progresses as per the roadmap, and meet the overall corporate objectives.

STAGE FOUR: SHAPING AND SCALING FOR THE FUTURE

With AI, there is probably never an end point. Organizations should continually evaluate what they want their operations to look like in the future, and how they can leverage their existing AI investment to shape and scale for that vision.

The goal is to create an integrated, orchestrated economy within the company that not only drives high levels of productivity and output, but also higher employee satisfaction. Although companies will see value from individual AI initiatives, the focus should not solely be on the here and now. There always has to be a vision of the future state—where the continued mix of human and digital works can be applied to ultimately transform the business and position it to thrive in an AI-empowered world.

For example, an insurance company with a unique, “pay-per-mile” underwriting model, based policy premiums on auto usage. It uses AI to monitor driver history, driving patterns and using this information, creates customer profiles.

Moving forward, the organization plans to further leverage this collected customer data to calculate each customer’s lifetime value, which helps with claims management and fraud detection. It is using the available technology to create a new way of underwriting and managing insurance.

Effectively Orchestrating the Human Side of AI

It is impossible to discuss AI orchestration in technological terms alone. Human beings, particularly those who will be training and working alongside the virtual workforce, directly impact how well the solution will succeed. As such, companies have to orchestrate the human aspect of AI with as much care and precision as they develop and integrate the technology itself.

Involve Talent at the Onset

In many cases, AI frees its human counterparts to do higher-level work. Supervisors who were previously managing 10 people are now responsible for overseeing 10 people and 400 digital workers, and ensuring that this blended workforce functions as a team to meet the common goal. In both cases, the original job and success criteria of these human workers have fundamentally changed. So, their job titles, their pay grades and the way their work is evaluated should evolve in kind.

As soon as your company creates a business case for AI, start planning to reevaluate the positions of your talent, as well as their roles, responsibilities, and compensation, to help retain good people and effectively embrace change from the disruption.

Start Employee Communication Early

No question, AI comes with a fear factor, not because of the technology itself, but due to the concern that algorithms or machines are going to “take over my job.” Without the right communications and change management strategy, employees charged with training the machines often believe that they are training their replacements—a mindset that not only impacts attrition but the ultimate success of the AI initiative.

The best way to quell any nervousness is to start communicating with employees early. Educate them on the realities of AI, and set up a solution simulation, so they can better understand how AI works—and its impact on their day-to-day tasks. If you’re working with an AI partner, that company may be able to bring in operations staff from other customers to answer employee questions or talk about their experience. This peer-to-peer interaction can do wonders to ease fears.

Most importantly, let your staff know what’s in it for them. Talk to them about how they’ll spend their day, and about how the machine will take over the more boring parts of their job so they can concentrate or higher-level tasks. If their job grade is changing, or their title is changing, let them know up front. Also, start educating supervisors on how they should manage a virtual and human workforce.

The earlier you start the communication, the more likely you are to transform your staff into AI advocates and agents of change.

Rethink the Traditional Approach to Training

Typically, training on other digital solutions is focused on consumption—how the employee uses the technology to do his or her job, and how that change impacts the existing workflow or specific steps in the process.

“With AI, by contrast, employees have to be trained on how to interact with the machine, so it can learn and continually refine its knowledge”

For example, if an insurance company applies AI to its claims processing operation, the machine will start to identify patterns, then checks with its human counterpart to validate whether or not there is a particular action that should happen when that pattern is detected. In other words, does a claim with this name and these set of conditions automatically pass through, or does it require human review?

If there’s solid interaction, and the staff member validates the pattern, the machine learns to pass those claims directly through, which quickly improves output and efficiency in the operation. If the employee isn’t responsive or provide the necessary feedback, nothing will improve. Claims will all be passed to humans for review and the company won’t see the results they expect.

So, traditional “how-to-use” training must be replaced with sessions that emphasize “how to work together with AI.”

Celebrate the Victories

After the solution is live, keep the communication going. Share the successes. Let the staff, and the company, know how the team is performing, and what they’ve been able to accomplish. This approach will keep the staff motivated, and make change management easier when additional instances of AI are integrated into other functions, companywide.

Adapting. Adopting. Competing in a Changed World.

AI has moved from the world of science fiction to the mainstream, combining virtual and human workers, working in tandem, to produce better outcomes than either entity could accomplish on its own.

Companies that can adapt to change, take advantage of the opportunity and capitalize on this model today will be tomorrow’s industry leaders. A well-planned, well-executed AI orchestration makes all the difference.

Contact US