RESEARCH REPORT

In brief

In brief

  • Many companies today are building AI solutions from scratch when they could be taking advantage of off-the-shelf models, algorithms, APIs and tools.
  • There is a fast-growing set of options available in the marketplace, developed, refined and proven by world-leading vendors.
  • There are also many AI partners with the resources to accelerate projects and limit the risk involved with building alone.
  • Find out when building your own AI solution makes most sense—and how to maximize the value your business can get from AI.


Across industries and geographies, organizations are clear that AI should be a critical element in their competitive strategy. Most believe it will change the nature of competition in their industry, and will also create new categories of products, business models and markets. Pressure is building to avoid falling behind, and multiple functions, units and divisions within organizations are diving head-first into building their own AI solutions. Yet in the absence of clearly-defined business goals and strategic AI goals, the likely result is AI solutions that are siloed or ineffective. Additionally, teams often end up building internal AI systems and capabilities from scratch when more efficient options—like buying off-the-shelf solutions or partnering with specialists—are available.

For some businesses, this AI sprawl leads to internal confusion, duplication of resources and lack of alignment around core organizational goals. The result can be numerous small AI solutions that have limited business impact. Our research shows that less than half of businesses are using AI strategically throughout the organization. Implementing AI in silos or disparate projects will not deliver companies the desired value.

50%

According to Accenture research, less than half of businesses surveyed are using AI strategically throughout the organization.

AI strategy options

To help curb AI sprawl and make smarter decisions about implementation, organizations should devote more thought to defining the value they are hoping to get from implementing AI solutions. An implementation strategy comes down to three broad options (and the hybrids between them), with each best suited to different scenarios:

  • Build. Building means creating the core AI solution using in-house capabilities. For instance, a bank might decide to develop a tool for predicting currency demand. To develop it, the bank will need to hire (or assign existing) AI and data specialists to work with in-house currency experts on a predictive model that can be continuously improved over time.
  • Buy. Major car manufacturers don’t make every component themselves. Instead they focus on areas where they can add the most value and differentiation, while procuring other components from specialists. Businesses can often work similarly with AI, buying software, APIs, or using open-source code. All of these can be integrated into the business and engineered so that they can upgrade to another solution when technology inevitably advances.

  • Partner. Many organizations don’t have the expertise required to develop an effective AI solution. Top skills are difficult to find and often there is reluctance to invest in expert talent before the value of AI has been demonstrated within the business. Other firms have meanwhile built the relevant capabilities, or even services, themselves; partnering with them—or an organization with the knowledge and expertise to work with a company’s domain experts—can be an accelerator of the project.

With AI, teams often lean first toward the build option, believing it gives them the most control over what will hopefully prove to be a competitive differentiator. More often than not, however, buy or partner makes more sense. Small businesses, for example, usually use APIs to access data from companies such as Visa™ on credit card transactions, data that can be used in AI solutions. If the needed data or expertise do not exist internally, companies need to determine who they can partner with to obtain them relatively quickly. These inputs may not be critical to the entire solution, but they can speed its execution along considerably.

There are five main considerations when deciding whether to build, buy or partner, as detailed below and in the decision tree. Some questions in the tree may sound obvious, but it’s not uncommon for companies to forget about the basics. Covering all bases, this decision tree can help diagnose whether teams choose to build, buy or partner.

VIEW FULL INFOGRAPHIC

Question #1: Feasibility

Is the feasibility proven with an agile pilot or case study research?

A quick and efficient pilot will confirm the suitability and viability of the AI project and whether it can scale—and companies recognizing the value of AI are doing just that. We recently surveyed businesses currently experimenting with new technologies and found a quarter are currently piloting AI solutions; another quarter are evaluating or planning to pilot. In the case where companies may not be able to conduct pilots, case study research and professional consultation could also help demonstrate feasibility.

An innovation lab or AI center of excellence should be able to find and deliver use cases that demonstrate project viability within 3-6 months. These use cases should be collected from different parts of the business, quickly evaluated and prioritized—or, if necessary, dismissed. And in today’s fail-fast innovation culture, the ability to discard quickly a business case that is not aligned with the business objectives is the mark of a winning company.

In fact, failing to truly understand the business case up front is likely to result in substantially more time and cost being spent on a project than anticipated and ultimately disillusionment with the technology. And an inability to take decisions quickly can clog up the pipeline of AI initiatives, strangling potentially good ones and leading to others being pursued at the edges with potential for duplication of effort and lack of reusability.

Lastly, feasibility research should also help determine the availability of off-the-shelf solutions or services that can deliver the desired depth of functionality at considerably less cost than building the capabilities internally.

An innovation lab or AI center of excellence should be able to find and deliver use cases that demonstrate project viability within 3-6 months.

Question #2: Organizational fit

Is the project genuinely needed, and is it funded?

Any AI project should be demand-driven. The value it will deliver should be clear, as should the nature of the benefit, whether it is marginal or transformational. The project objectives should be fully aligned with those of the business, and the stakeholders need to be committed and supportive. Commitment is clearly shown by funding decisions, but it also manifests itself in internal resource allocation, internal communication and flexibility. Whatever the project, successful companies are those in which senior leadership is well informed and the project objectives are fully aligned and clearly communicated to the delivery teams.

Question #3: Data

Is the data required for the intended purposes of the project available? Is that data unique?

A project using unique data generated internally as part of the business model—for example, customer interaction data—has the chance of becoming extremely valuable and strategic. This is especially true if the data or algorithms can be monetized (i.e., made available to third parties for a fee). In such cases, building the AI product or service in-house makes sense to retain intellectual property rights.

However, if the data for the project is widely available in the market—for example, data from Twitter™ or other social media—then the company should consider buying its AI solution externally. Top-notch services exist today that collect social media, mobile device or other data and offer analysis on top. Considering there are many options and prices are relatively low, companies can use the solution that is “good enough” rather than try to gather their own data and slow down the AI implementation process. Our research shows that limited internal data availability and quality are the most common reasons why companies lag in AI deployment.

A project using unique data generated internally as part of the business model—for example, customer interaction data—has the chance of becoming extremely valuable and strategic.

Question #4: Strategic impact

Is the project strategic?

If it is determined that a project will deliver a competitive advantage or have transformative impact, it is clearly of strategic importance. If the objectives are narrower—for example, keeping pace with rivals or meeting customer expectations—the impact of the project is more likely to be tactical. An additional consideration is whether the project is relevant only to the unit or function implementing it, or if it will also be useful for other parts of the business. A solution or tool that can be easily adapted and re-used by multiple divisions, functions or units may be deemed strategic, even if the team developing it is pursuing tactical business objectives.

An AI project deemed to be of strategic significance may very well merit a decision to build internally, depending on the existence of internal capabilities (see Question #5). A project of tactical importance should almost always entail a buy or partner approach, given the lower costs and risks involved in pursuing these.

Three AI operating models

The question of strategic versus tactical impact of an AI solution may be addressed differently depending on the operating model each company adopts to execute AI projects. Companies tend to employ one of three models:

  • Distributed: Highly distributed ways of working are the sign of mature technology, tools and usage patterns. Here, data scientists have considerable resources at their disposal and can be productive working in loosely connected groups with established ways of reusing and providing knowledge and expertise to the wider business.
  • Centralized: At the other extreme are highly centralized models, in which all data science and AI projects are executed centrally by a specialized team. Local groups that want to use the technology consult the center for all aspects of development. While efficient, this model can become a bottleneck of value and growth for companies, especially large ones in which command and control at this level might be difficult to achieve.
  • Hub-and-spoke: This is a great model of working when you are setting yourself up for a journey to maturity. A central hub helps drive standards, methods and tools, and it guides the prioritization of work as the organization matures to support the technology. It consists of people who belong to the hub (data scientists, engineers of different types and data specialists) and specialists who belong to the business units.
An AI project deemed to be of strategic significance may very well merit a decision to build internally, depending on the existence of internal capabilities.

Question #5: Capabilities

Do we have the required technical capabilities in-house?

If an AI project is deemed to be strategic, the last major question to ask is whether the organization has the internal capabilities to deliver it. If the answer is yes, developing internally will usually make sense. There could, however, be a case for using a vendor solution or partnering if speed is of the essence and mobilizing existing internal resources will take too long. In such situations acquiring a start-up with the right solution and ready-made capabilities could also be a good option.

If the capabilities are not on hand, the question then becomes whether to develop them internally, buy an off-the-shelf solution or use a specialist partner to help develop the project. Given that top AI talent is highly sought after—many organizations have trouble attracting and affording the right people—buying or partnering becomes an attractive option for many firms.

A hybrid approach is also possible, in which the company can draw on partners and/or buy off-the-shelf components (to ensure better results in shorter time with lower risk) while simultaneously developing a capability that can take on more of the heavy lifting on future projects.

In all cases, though, companies should always make sure that they are not reinventing something that’s commoditized and available at reasonable cost in the marketplace.

A hybrid approach is also possible, in which the company can draw on partners and/or buy off-the-shelf components while simultaneously developing a capability that can take on more of the heavy lifting on future projects.

The right decision

Few organizations will be able to create value by building AI capabilities in-house alone. Leveraging existing solutions and capabilities developed by others will often achieve the desired business result faster and more cost-efficiently. To get similar outcomes from building, organizations will have to invest more time, money and effort, which only makes sense if the project is a strategic differentiator for them.

This is why the optimal approach would be to invest in two ways:

  1. Get to market quickly by integrating what already exists. Take advantage of the wave of innovation and the investment of others.
  2. Invest in in-house capabilities to create long-term value and differentiation by building what will support the future business.

This common-sense approach will maximize the value that organizations get from AI and avoid wasted time, effort and resources.

Fernando Lucini

Managing Director – Applied Intelligence and Artificial Intelligence Lead, Accenture UKI

MORE ON THIS TOPIC


Subscription Center
Stay in the Know with Our Newsletter Stay in the Know with Our Newsletter