As AI capabilities continue to expand, emulating and augmenting human skills, they tend to materialize in organizations in the form of a virtual workforce, which is becoming an increasingly sought-after option—especially for organizations looking to complement their existing human workforce and harness additional, virtual, resources capable of handling part of their business operations.

To put a virtual workforce into practice, however, requires a number of processes, such as product selection, implementing, integrating and testing different components, not to mention parameters around security, data privacy, scalability and resilience. For this reason, renting a virtual workforce is becoming an increasingly popular alternative.

Cyrille Bataller, managing director for Intelligent Automation at Accenture, explains why enterprises should consider renting not building their virtual workforce, and what they need to consider when they do.

You can think of the virtual workforce like a car. Let’s say you need to travel to many different places and therefore need a car. Your immediate thought wouldn’t be to build the car yourself by assembling a wheel, frame and engine. You’ll more likely consider buying a car that fits your needs.

However, what if you know you will only need a car on occasion, like on weekends? Or you have different errands or tasks throughout your week that will require a different type of vehicle? What if you don’t want the responsibility of constant car maintenance? The benefit of renting a car - whether it’s the car itself or the ability to get from point A to B using a sharing service - is that you have the flexibility to do whatever suits you at any given time.

The same can apply for renting a virtual workforce. Renting enables you to have that flexibility and trade in your current virtual workforce for a new or different model. Maintenance and upgrades are taken care of, freeing you up to focus on what matters.

The virtual workforce is a complex thing—with many different components, options and traditional IT questions that need to be answered for security, scalability, encryption, performance and response time. This can often make renting the most suitable approach for organizations when deciding to leverage the virtual workforce.

There are several benefits to renting as well. For starters, renting is an operating expense (OPEX) rather than a capital expenditure (CAPEX), meaning the cost is spread out which may be more suitable in certain cases.

Renting allows organizations access to the best skills on the market, much like outsourcing to a team of skilled freelancers. While organizations tend to outsource only non-critical parts of the business to optimize costs, with the virtual workforce they can harness all the skills they need to outsource a variety of tasks. This not only can yield more flexibility but also save on the costs of training, management and turnover.

When renting, organizations can also utilize different types of skills depending on their needs. If their business is quite seasonal, for instance, they can rely on a virtual workforce to absorb peaks in volume. This can then be scaled up and down with ease and simplicity.

For example, take a university in France that registers 30,000 students every summer. When these students register, they provide between five and eight pieces of documentation to support their registration, such as identification documents, previous transcripts or diplomas.

To facilitate this process, a solution was put in place to automate document assessment and decide whether they could be validated. Only if there was a problem would a human worker need to look at them. This is an example of a seasonal peak. For the rest of the year, the university would not need that workforce, meaning that renting a team of virtual workers could be the best option.

Yes, the risk is lower—but the risks are also different. When renting, the investment is less because it was made by the company that created that virtual workforce. It's less risky and economically more interesting than building because it also means you can be in business tomorrow and not in a year or two.

There are a lot of misconceptions about what AI can do, and a lot of resistance to implementing it at scale as a result. Renting a virtual workforce is a good way to trial it on a business-centric use case to separate hype from reality and show what it actually does, what its limits are, and how reliable and accurate it is.

Renting also helps enable organizations to trial the concept of using a human plus machine approach, taking only days or weeks to prepare, rather than having to commit large investments without knowing how the enterprise will accept the change.

With a virtual workforce, it’s about building trust in the same way as you would with a new employee or an intern. First, you’ll train them on what they need to do, then you’ll give them some simple activities and you’ll audit to make sure that they do this right. You won’t put them in front of clients to start with. You’ll have them work in a shadowing mode and scale progressively, building confidence that they are doing what they are supposed to do. Eventually, you might try throwing more complex cases at them to see how they handle those and understand the limits in complexity that your workforce can handle.

To test the virtual workforce, it’s important to utilize real data to make sure that it does what it is supposed to be doing. There’s a lot of hype around AI capabilities, and many claims of what AI, machine learning, natural language processing and computer vision can do, but actual abilities vary widely from one AI bot to another. There’s nothing like a test or hands-on experimentation to ensure your virtual workforce can meet your business needs.

There are a number of questions to consider:

• Investment: Is the investment the organization wants to make opex or capex? Are there resources to invest in building a platform that's going to take a year or two before it can be leveraged at scale and with benefits?
• Skills: Does the organization have the right skills in-house to build these capabilities themselves? And if not, is it safer to rent from the market?
• Priorities: What use cases should be prioritized for the virtual workforce? Businesses should normally start with a use case that is internal facing and non-critical but impactful in scale. This is a good way to test the virtual workforce as well as to train the human workforce on what virtual workers can do and how they team with people.
• Seasonality: Is there much seasonality in the business? If there are seasonal peaks, renting will allow a business to scale up or down as needed, so that the investment reflects spending patterns and business needs.

In the same way that DevOps shortens the processes between software development and production teams, enterprises looking to scale their virtual workforce adoption should consider BotOps. Put simply, organizations need to have an HR team for their bots, closing the gap between their human and virtual workforces.

In most cases, BotOps would take the shape of an orchestration platform that would ensure there is a consistent view of AI adoption and the type of virtual workforce that is being leveraged across the enterprise. This platform would also be responsible for a number of different things, ranging from provisioning, training and investing in the development sandbox, to deploying as a pilot, production and monitoring, operating and handling errors.

A virtual worker is like a human worker in that it is going to make mistakes occasionally and therefore there needs to be a mechanism for escalation, redress and correction. This orchestration platform would also constantly train and tune the digital workers in the same way that organizations constantly train and tune human workers.

Above all, making the decision to rent is about enabling an organization to stay in control of its virtual workforce, allowing it to accelerate the pipeline from idea to production to outcome, and harnessing the power of applied intelligence at scale as the business needs it.