A Practical Guide to AI Adoption: Build, Buy or Partner?

Key Takeaways

• Balancing Build vs. Buy Decisions Requires Strategic Assessment
  Companies must evaluate their unique needs, considering factors such as intellectual property value, data access, and organizational priorities. For core competencies, building in-house may provide strategic advantages, while off-the-shelf solutions can be sufficient for non-core functions.
• Hybrid Approaches Offer a Compelling Middle Ground
  A hybrid strategy, such as partnering with AI-as-a-Service providers, can combine the strengths of in-house development and pre-built solutions. This allows organizations to customize AI to their needs while benefiting from rapid deployment, scalability, and reduced risk.
• Hidden Costs and Scalability Challenges Should Not Be Overlooked
  Building in-house solutions requires substantial investment in talent, infrastructure, and time, with hidden costs such as compliance and ongoing maintenance. Scaling these solutions effectively poses additional challenges, making it essential to weigh the long-term implications of both options carefully.

When organizations adopt new technologies they often find themselves at a crossroads: should we build our capabilities in-house or purchase ready-made solutions? 

This is no different for AI. According to a survey of 600 U.S. enterprise leaders, companies reveal a near-even split. 47% are developing AI solutions in-house, while 53% are sourcing them from vendors. 

However, the “build versus buy” debate in the context of AI does not necessarily need to be confined to a binary choice. Organisations may wish to follow a hybrid approach and purchase a solution that builds on their proprietary data. This is where finding an AI partner that can offer their expertise and craft solutions tailored to your organizational needs is crucial. 

To help leaders determine the best approach for their organisation, we’ve developed a key set of factors that should be considered when devising a strategy for AI adoption. 

Value of Intellectual Property 

A key consideration is how closely the technology supports your company’s core strategic objectives. If you have proprietary data and it is your unique selling point (USP), building your own custom solution may be preferable. This ensures the intellectual property remains within your company which can confer long-term strategic benefits. An often overlooked benefit is that it will also enhance the technical skills of your organization which may be useful for future projects. 

Consider the example of Hackajob, a tech career marketplace where employers can find and hire tech talent. CEO and Founder Mark Chaffey believed they could build the best recommendation engine in the world for jobs because they had around fifty million first party data points on individuals which they could use to fine tune their own models. 

Conversely, for non-core business processes, buying an off-the-shelf solution is often sufficient. Mark Chaffey highlighted how developing an in-house customer service would not be worth the return on investment for Hackajob as it would divert resources from their primary mission and existing market solutions sufficiently met their needs. 

Access to High-Quality Structured Data

Having access to high-quality data is crucial if you want to build an effective AI solution as organisations are unlikely to gain competitive advantage from publicly available datasets. While Marcus East was technical director at Google, he built KI, the world’s first AI-driven insurance syndicate. Google worked with Brit Insurance to create a unique proprietary model which gave them a sizable strategic edge.

Data structure is as critical as data quality. If you do not partition your data well, you might end up querying your entire database, unnecessarily driving up costs. Likewise, If your data is scattered across hundreds of databases rather than being held in a single global customer database, it becomes more difficult to guarantee the accuracy of data which is essential for building reliable AI solutions. 

Data Privacy and Regulation

For organisations handling sensitive data, particularly in healthcare and finance, building in-house may provide better control over data security and privacy. Third-party AI solutions may rely on public cloud infrastructure with potentially opaque security policies, introducing risks of intellectual property leakage. However, the right partner should ensure complete regulatory compliance and will protect every aspect of your client and company data. You can check for key security measures such as ISO 27001 certification and AES-256 encryption. 

Even the build approach is not completely without risk. Projects may be subject to data risk if they are built on top of open-source models. Additionally, AI owners most factor in the high cost of ensuring compliance with regulation like the recent EU AI Act. The financial stakes are substantial, with non-compliance penalties ranging from €7.5m (or 1.5% global annual turnover) to €35m (or 7% global annual turnover). While purchasing pre-built solutions might remove some regulatory risks, organizations must carefully evaluate potential legal liabilities.

Flexibility & Scalability

Vendor Lock-In Pitfalls

Consider your future growth and how easily the solution can scale. While third-party providers offer the benefit of lower switching costs and immediate implementation, organizations must carefully evaluate potential vendor lock-in scenarios. External providers may modify service offerings, or face business challenges that could disrupt your AI strategy.  Usage-based pricing can escalate costs as scale increases, while hidden fees for customization, upgrades, and support may inflate total ownership costs. 

Potential Risks of Building Custom Solutions

Custom solutions offer the advantage of being designed for scalability from the outset, allowing organizations to precisely align technological infrastructure with their unique operational needs. However, this needs to be weighed up against the risk that a custom-built solution is not guaranteed to deliver a competitive advantage. A prime example is BloombergGPT, a specially trained finance large language model (LLM) which was trained on all of Bloomberg’s data in 2023. However, that was quickly superseded by OpenAI's GPT-4 which demonstrated superior performance across finance-related tasks without specialized finance training.

How Difficult is it to Scale In-House?

Whilst it is relatively easy to build an in-house minimum viable product (MVP), it is easy to underestimate the challenge of scaling an AI system. To scale effectively, companies need to address a range of technical requirements, including integrations with diverse knowledge sources, low latency retrieval systems and high-volume data ingestion pipelines. As the system scales across an organization, it also has to be optimised to handle a diverse set of data sources and user flows. Companies also need to invest in product design to ensure that the AI system is usable and meets the needs of end users. 

Hidden Costs of Building AI solutions

Building AI solutions in-house offers the advantage of customization but comes with a range of hidden costs that are often overlooked. Below are some of the most significant costs companies should consider when deciding whether to build in-house AI solutions.

Time to Market

Building custom AI is time-intensive. Development can take weeks or months, whereas buying and implementing solutions is typically faster. This extended timeline is often compounded by the need to recruit specialized talent. According to LinkedIn data, engineers take the longest amount of time to hire (49 days) compared to other professions. 

AI Expertise

AI talent is in high demand and commands premium salaries. To develop a successful AI solution, companies need a team of highly skilled professionals, including data scientists, machine learning engineers, and software developers. The median annual salary for a data scientist in the U.S. in 2023 was approximately $108,020 according to the U.S. Bureau of Labor Statistics. This can be particularly burdensome for smaller companies, forcing them to either stretch limited budgets or compete fiercely with larger tech giants to attract top talent.

Cloud Infrastructure 

AI demands far more computational power than traditional software, leading to higher operational costs, particularly in cloud environments like AWS or Microsoft Azure. The cost of training AI models can easily run into the hundreds of thousands of dollars, especially as models that process images and video demand substantial storage and compute resources. These costs are compounded by the need for continuous retraining to keep models relevant and accurate as the data that enters AI models evolves over time (a concept known as "data drift"). In addition, running AI at scale can result in hefty costs for data ingress and egress, particularly when models are transferred between cloud regions to improve performance or comply with regulatory requirements.

Keeping “Humans in the loop”

Training AI models typically requires extensive manual effort to clean and label large datasets, a process that is both costly and time-consuming. Humans also play a vital role in verifying AI outputs, especially in tasks demanding cognitive reasoning like content moderation for social media sites. Although the need for human involvement may decrease as AI performance improves, it is unlikely that humans will be completely removed from the equation. 

Should I Build AND Buy?

Given the limitations of solely buying or building, many organizations are finding success with a hybrid strategy by forming a partnership with an external provider. 

Partnership or AI-as-a-Service (AIaaS) can combine the benefits of both approaches in the following ways:

1. Gain Immediate Value from Proven Solutions
   AIaaS offers tailored solutions with the expertise of solving problems for numerous similar businesses, enabling organizations to bypass long development times and start delivering value from day one. 
   
   
2. Flexible and Scalable AI Solutions
   Adapt quickly to changing needs or scale across your organization with AIaaS, blending the speed of pre-built solutions with the customization needed for evolving challenges.
   
   
3. Focus on Your Core Strengths
   Offload routine or complex AI tasks to AIaaS providers, allowing your team to concentrate on strategic priorities and maintain competitive advantages.
   
   
4. Access to Expert AI Talent
   Partner with a diverse team of data scientists, engineers, and designers, ensuring high-quality AI solutions that would be difficult and costly to assemble in-house.
   
   
5. Cost Efficiency
   Avoid the high upfront costs of fully building in-house while reducing long-term vendor dependency costs.
   
   
6. Mitigate Risk & Validate ROI
   Before committing to a custom solution, you can test a product and validate ROI before investing further for in-house development.

How to Evaluate the Right Path?

To determine whether to build, buy, or partner for AI, consider these key steps:

1. Define your AI objectives
   Clearly articulate the goals of your AI initiative to align with business needs.
   
   
2. Assess Internal Capabilities and Resources
   Evaluate whether your team has the necessary expertise, resources, and time to build in-house, or if outsourcing is required. Consider the timeline for deployment and your budget, as well as the ability to attract and retain top talent.
   
   
3. Research Available Pre-built Solutions
   Look for products that meet current needs and have the potential to scale with future growth. Assess the reliability, reputation, and integration capabilities of potential vendors.
   
   
4. Conduct a Thorough Cost-Benefit Analysis of Build vs. Buy Options
   Compare the financial, time, and resource costs of building vs. buying. Consider long-term expenses and hidden costs for both options.
   
   
5. Consider a Hybrid Approach
   If building in-house isn't feasible, partnering with an AI provider could combine the strengths of both building and buying, offering tailored solutions with faster implementation.
   

Overall, the decision to build or buy depends on your individual business requirements and resources. Buying off-the-shelf solutions, while faster to deploy and lower in initial cost, may lead to vendor lock-in and limit long term scalability. Building-in house offers control and flexibility but is often far more difficult than anticipated with significant hidden expenses. 

There is a third and potentially overlooked option. Organizations can benefit from a hybrid approach by partnering with an AI provider who can deliver tailored solutions that combine the best of both worlds. This hybrid model reduces risk, accelerates time-to-market and optimises costs, while ensuring the AI solution evolves alongside the business. 

If you are looking for a trusted AI partner to supercharge your growth and productivity, you can book a call with our team.