Choosing the right AI solution: Five build or buy mistakes to avoid (and what to do instead)

As a data scientist and AI expert, I work with clients on both AI strategy and hands-on solution design. One of the biggest challenges they face is navigating the “build or buy” question: Should we build an AI solution or buy one off the shelf? 

This sounds like a straightforward question, but the reality is more complex. The market is becoming crowded with AI solutions, and board members, stakeholders, internal teams and consumers often have conflicting demands on how to adopt AI and its proper use. This creates pressure to not only make quick decisions, but the right ones.  

To help you navigate these complexities, here are five common mistakes organizations make when deciding on their AI investment, along with examples and recommendations on more effective approaches.

Mistake #1: Believing AI can solve every problem

When I first meet with clients, they are often surprised when I begin by asking: “Does your organization really need AI?” Surprisingly, the answer is sometimes no. Many clients assume AI can solve everything, but this can lead to overcomplicated and costly solutions when simpler approaches could work better.  

Client example: Government agency with mismatched datasets

• Challenge: A government agency wanted AI to compare two datasets from different systems to understand why they didn’t match.
• Solution: In this instance, AI wasn’t needed at all. Instead, their need could be addressed by running two daily SQL queries. 
• Recommendation: Start with your problem, not the technology. Depending on your goals, non-generative AI, or even non-AI solutions, can be cheaper and better suited for your needs.

Read more on this topic from my colleague Diane Gutiw: Let’s stop talking about AI.

Mistake #2: Focusing on GenAI and not exploring all available AI and non-AI options 

Despite being around for decades, many still equate AI with GenAI (e.g., LLMs and chatbots). While GenAI solutions are undeniably useful, there are other AI and non-AI tools that can be more accurate and cost effective. 

Client example: Extracting information from documents 

• Challenge: A client wanted to use a large vision model to extract information from multiple reports.
• Solution: A traditional OCR (Optical Character Recognition) tool like Tesseract, built specifically for this purpose, provided the client with greater precision and efficiency—without the use of GenAI. 
• Recommendation: Explore the full range of existing technologies and solutions. Compare GenAI with traditional approaches, and consider combining tools (e.g., OCR with GenAI) to yield better results. 

Read more on this topic from my colleague Dave Henderson: AI is more than just GenAI.

Mistake #3: Not anchoring decisions in your organization’s strategies and capabilities 

Your organization's digital strategies and capabilities should form the foundation of your build or buy decision. Keep in mind: AI is not one-size-fits all. The same solution can produce different outcomes between organizations, depending on their business processes. 

Client example: Choosing a solution to align with current needs and future strategies

• Challenge: A client identified a need requiring an external solution but was unsure whether to buy an existing tool or custom building one. 
• Solution: While not the only factor, a good starting point when making your decision is reviewing your organization’s IT footprint and strategy. From there, options include: 
  
  • Engaging partners to build AI solutions that they later manage in-house.
  • Partnering with vendors like CGI to not only build solutions but also maintain and continually update.
  • Using an off-the-shelf product can be faster and more affordable.
• Recommendation: Align your AI investment with your digital strategy and capabilities. Engage vendors to evaluate solutions based on your organization’s needs, not what worked for others. 

Mistake #4: Custom-building a solution without considering existing options 

Clients sometimes assume they need specific customizations, but often there are off-the-shelf solutions with embedded AI or have features that could meet their needs in the future. 

Client example: Build or buy a data query solution?

• Challenge: CGI builds custom Ask Data AI solutions that can work across multiple sources like Databricks and Snowflake. There are also off-the-shelf tools and platforms embedded with AI capabilities. In cases like this, should they build or buy? 
• Solution: For smaller organizations that store data within one source, there isn’t always a need to build a custom solution because the AI capabilities within off-the-shelf solutions are sufficient. However, when needs become more complex—such as multiple data platforms or deep domain knowledge requirements—customization can be the better path. 
• Recommendation: Work with service providers or a trusted third-party to see if your needs can be met with existing or tailored solutions. CGI has alliances and advanced insight into product roadmaps across providers and solutions to help you make informed decisions. 

Read more about the client story above: GenAI cuts query response time to just 45 seconds for telecom firm.

Mistake #5: Chasing the latest tools without clear objectives

The pace of AI advancements and new tools can be overwhelming to track and integrate into your workflows and solutions.

Client example: Existing AI solution overtaken by new tools

• Challenge: A client needed to choose between an existing AI solution and new tools with extra capabilities. 
• Solution: Despite the features newer options offered, the original solution still generated accurate results, strong ROI and time-savings—making a replacement unnecessary. 
• Recommendation: Focus decisions on your business objectives. Continually assess new tools with the help of AI experts or trusted partners but weigh them against your objectives and performance goals. With AI, it’s not just about its capabilities. Factors like security, trust, data, and regulatory can introduce complexity not previously found in non-AI solutions.  

When and how to best engage third-party expertise 

With AI evolving rapidly and competitors continually strengthening their capabilities, clients often engage CGI to support their AI strategy, solutions implementation and enterprise scaling. 

Contact CGI to find answers to these and other AI questions

• What are the pros and cons of fine-tuning versus pre-training LLMs? 
• How should an organization evaluate standard GenAI business versus deep domain use cases? 
• Our organization is new to AI. To maximize our ROI, should we consider GenAI or other options like traditional machine learning, computer vision, or standard automation? 
• What is the best way to implement an AI solution end-to-end, from data processing to production? 

Learn more about our latest AI success stories and listen to our newest podcast for practical AI advice.