Clever clouds: Considerations when outsourcing AI

Outsourcing AI: Is it right for you?

These are all important questions. Most, if not all, of them can only be answered on a case-by-case basis because there is no one-size-fits-all answer for all industries or countries, let alone all companies. For example, some organizations, such as governments and others who store sensitive data, are not allowed to use cloud-based solutions where data centers are located outside national borders. Whereas others, especially less digital businesses, do not have the skills or infrastructure to even consider building AI solutions in-house.

However, since off-the-shelf, cloud-based AI solutions (what we could call "outsourced AI") are more available and accessible than ever before, it is worth highlighting some of the general considerations you should make when assessing these options.

Outsourcing AI to a major cloud services provider is a smart choice in many cases. For example, it is beneficial when your organization is unwilling or unable to create in-house capabilities; where the nature of your business is not digital or technical, such as a clothing retailer; or when you are looking to experiment with AI solutions before investing further. Major providers have ready-to-go AI solutions that can do things like image classification, language recognition or speech generation.

Essentially, cloud-based AI providers have accrued valuable, far-reaching experience in a field that is relatively new and fast-evolving. Many businesses now see these services as a valuable commodity—a "plug-and-play" solution. But there are a couple of limitations to consider as well:

• Limited customization. Cloud providers currently only offer quite basic AI services that support common use cases, but they may not always be able to support a specific business need. For example, while image recognition solutions are ideal for spotting well-known faces or objects, if the client wants the AI to recognize, say, the faces of their own personnel or components of industry-specific machinery, the third-party AI solution is less likely to be useful. This is because the providers’ AI models are trained using only general public data and simply don’t have the depth of knowledge that this client-specific task demands.
• Data governance. Data and its management require strict governance. Where is the data stored? How long is it being stored for? And who can access it? Providers offer contractual assurances, but some organizations (e.g., those that deal with classified or sensitive data, such as medical records) may not have the permissions or governance structures in place to use these services.

Hybrid solutions are often the best strategy

For these two reasons, we often advise clients to opt for a hybrid solution. Along with their plug-and-play AI services, most cloud-based AI providers sell the building blocks of those services. This could be, for example, server images with all required software installed, or template code frameworks. These can accelerate the building of customized AI solutions because data science teams in your organization can take advantage of elements of proven, large-scale AI offerings from a partner (e.g., advanced image classification), while building and integrating custom solutions much faster than if you were to start from scratch.

If your organization does not have data science teams in-house, you can consider outsourcing that side of things too, via a partnership with a suitable organization. For example, an accounting firm approached us to digitize scanned invoices issued in Italian and extract the list of issuing vendors. To do this, our team of data scientists used an AI-driven, image-to-text API from a cloud-based provider. This digitized the invoices, but the team then also developed a custom natural language processing code to extract the data of interest from the digitized text. The time needed to prepare the infrastructure and install the software was minimized by using a cloud-based server, and so the customized solution was built in a very short time using template code frameworks (also from the same provider).

Realistic expectations and the value of customization

Before using cloud-based AI, or outsourcing your AI build, it is important to have realistic expectations around what the technology can do, especially when it is new to your organization. “The biggest pitfall in AI, and therefore outsourcing AI capabilities, is the assumption that it will solve everything,” says my colleague, Vinod Patel, Managing Director of Accenture Operations. “Organizations need to apply a level of discovery to their current and, importantly, their future business to determine the applicability of AI. The prioritization of AI will help business leaders determine where to start and the journey to increase AI—and in turn this will frame the strategy for outsourcing AI.”

Another point to emphasize is that some of the best AI solutions are the most customized solutions. Often this means, at the very least, using hybrids of off-the-shelf solutions and customized elements. Off-the-shelf components are an invaluable component, but in our experience, it is only by fully adapting AI to suit an organization's specific context that it can deliver the most value.