To Build or to Buy AI?

In recent years, Artificial Intelligence (AI), which combines computer science and robust datasets to enable problem-solving machine learning, has been making great strides in terms of analytical capabilities, and different use cases are rapidly emerging. As awareness of these use cases spreads, public and private sector organizations are exploring how to harness this technology and asking — should we build or buy AI?

If you’re in the position of deciding whether to build or buy AI to solve a problem, you can follow the steps below to guide your decision-making process and figure out the best way forward for your organization. This decision deserves careful consideration as AI is a complex discipline.

Step 1: Identify the need

Use these guiding questions to help determine if AI is needed:

• What problem are you solving?

• What are the inputs/outputs in terms of data? Does the data exist to input?

• How would an algorithm solve this?

• Does a solution already exist?

Step 2: Define your decision criteria

Defining your decision criteria can serve as a navigational compass for your decision-making process and help you prioritize your desired outcomes for an AI solution. Use the factors below to evaluate whether building or buying AI is the right choice for your initiative.

• Competitive Advantage: Does the technology provide an asset of value? Is the proprietary asset high quality? How important is it to have the intellectual property (IP) know-how? Does this IP offer increased value to stakeholders?
• Requirements Alignment: How well does the commercial product meet your needs? Are your objectives met with the product? Are customizations required? Are data security requirements met?
• Financial: Is there a specific net present value (NPV) or return on investment (ROI) that must be achieved? What is the total cost of ownership, including human capital, software and hardware? What are the costs to maintain it?
• Time to Implement: What is the speed to develop and deploy?
• Maintenance: What is the effort and cost of support for human capital, hardware and software?

Step 3: Consider the pros and cons

When choosing whether to build or buy your AI solution, there are advantages and disadvantages to consider. You should consider these in each situation. 

Pros of building AI in-house

+ Customization

Leveraging in-house AI solutions enables companies to potentially reduce overheads by building only what they need. This means training algorithms on the specific data they'll be working with against a tightly defined use case. Packaged solutions may have unnecessary features or require heavy customizations.  

+ Flexibility

With customization, organizations can also expect increased flexibility when developing their own AI solution. Going in-house offers the opportunity to develop plug-and-play functionality that aligns with future needs or ad-hoc requests. If freedom to modify is a top priority for your business, consider building an ideal platform to ensure maximum flexibility and control over product design.

+ Intellectual property (IP) ownership

It’s no secret that investing in a maturing technology area can create value through IP and patents. The possibility of owning IP can be a competitive advantage, providing an asset to your company. If you’re building a potentially competitive AI solution, consider IP ownership when deciding whether building or buying is best for your business.

Cons of building AI in-house

- Time commitment

Building a production AI solution is typically a multi-year effort — model development is complicated and requires careful thought around automation, reporting, and user experience. Plus, the solution needs to meet scale and be integrated with source systems and control frameworks.

Even with the support of open-source software, the duration to build AI solutions can extend far beyond desired outcomes depending on what models you are training. For example, Natural Language Processing (NLP) requires deep learning models to solve extremely complex problems. Deep learning uses a model inspired by the brain with artificial neural networks that continually improve its ability to make predictions based on data. Building NLP solutions requires rigorous testing and training to ensure the models perform well and you reduce false positives. The rules that dictate the passing of information using natural languages are not easy for computers to understand, making building NLP difficult. The challenges in training, reducing false positives and quality production may take years to overcome.

- Large financial investment 

Going in-house requires investment in the millions. Whether it's ten or a hundred people involved, all the different tools and platforms required will make this a significant financial investment. And without a solid strategy and organizational buy-in, your AI and machine learning efforts could waste a lot of money and resources and never become production-ready (suitable for the end user).

For solutions in the cybersecurity and safety tech space, these problems need to continually improve and evolve to keep up with the rapid pace of adversaries. Even if you were to build the most advanced solution, that might become irrelevant a few months or years as the data changes – this is known as performance drift. Building AI platforms requires massive continual investment to stay ahead of the threat curve. You need to be engaged with adversaries' techniques, tactics, and procedures and come up with a long-term roadmap. You also need humans in the loop to review outputs, suggest improvements and then retrain and redeploy models. 

Sharing the cost with other agencies under a SaaS license subscription model allows the costs to be spread across many years and not front-loaded (where poor performance may lead to early termination).

- Lack of expertise

While employing existing talent to build software may seem like a good idea, it may not be the best route. With an increasing demand for AI talent, maintaining AI solutions can become a concern for organizations, putting knowledge retention and product intimacy at risk. You need a substantial team to deliver and maintain AI solutions, including data scientists, engineers, machine learning experts, and product experts.

- Lack of suitable training data

A machine learning program requires access to clean, integrated data and data processes. You’ll need a strategy that includes setting definitions, eliminating data silos, establishing governance and aligning business processes. Ownership and restrictions on data are big roadblocks to gathering training data and AI solutions require access to tremendous amounts of data that must be trained and revised. 

Pros of buying AI

+ Cost savings 

Software vendors pay for research and development (R&D) costs by spreading expenses over many clients instead of incurring them with in-house solutions. Buying an AI solution off the shelf will save substantial investment in R&D and human resources (HR). 

+ Quality assurance

Gartner predicts through 2022, 85% of AI projects will deliver erroneous outcomes due to bias in data, algorithms or the teams responsible for managing them. This can lead to harmful outcomes and reputational damage for organizations. Commercial packages are often more rigorous in quality assurance, which leads to higher quality and more stable code for businesses.

+ Compatibility

The vendor handles the cumbersome task of ensuring compatibility with other software systems, allowing the customer to focus on other demanding priorities.

+ Security

Information security is usually built-in with vendor options and continuously upgraded to safeguard data. A bought solution reduces risk and comes with pre-built integration with enterprise systems and security frameworks.

Tech solutions also need to prove compliance around the globe, whether it’s FedRAMP in the US or ISO in the UK. You need to consider additional compliance requirements when you're engaging with certain public sector or financial bodies, so you need governance there too.

+ Knowledge share

Using tailor-made AI solutions allows you to leverage the learnings of other clients, including upgrades and enhancements to improve performance. Even if you have a skilled data science team, they will need an end-to-end understanding of how a specific problem manifests itself and to be experts in that subject matter to build a solution. 

If we are talking about complex issues that are continually changing, that requires extensive research. For example, although the industry around combatting misinformation is new, there's an evolving playbook of solutions that have been tested in real-world scenarios by niche solution providers and adapted to be more and more effective at tackling this complex problem. 

+ Reduced time-to-value

Businesses want to deliver AI benefits quickly and sustainably to focus on reducing time-to-value while ensuring the solution can be scaled and maintained across the organization to deliver analytics and automation capabilities.

Cons of buying AI

- Ethical concerns

The AI software industry is growing rapidly, evolving at a faster rate than legislators can keep up with. Many frameworks and guidelines exist, but they are implemented unevenly, and none are truly global. Ensure vendors align with existing frameworks and use AI responsibly and ethically.

- Lack of compatibility

Evaluating compatibility is critical to ensuring AI solutions will work successfully within your current ecosystem. Compatibility issues can lead to cost overruns and decreased business continuity.

- Learning curves

Vendors often use clients as part of their research and development. Consequently, the lack of product optimization can result in many unnecessary installations, costing time and money. 

- Generalized solutions

While packaged offerings are ubiquitous, many businesses need unique solutions. The need for custom solutions requires additional cost and time for implementation.

Step 4: Make the right decision for your organization

With strategic decision-making, it’s essential to evaluate how a decision will affect your company objectives carefully. In the case of building or buying AI, ensure that the AI competency fits into your organization’s strategy.

Additionally, don’t forget to convene with key stakeholders to review decision criteria and determine the pros and cons. 

Lastly, talk to vendors as you weigh up your options. Be transparent with them and get their honest opinion about your choices and situation. For example, some vendors might be open to training their AI on your data to meet your requirements and will therefore be open to discussions about bespoke solutions.