As generative AI continues to reshape industries, organisations are eager to integrate this technology into their operations. While the temptation to have in-house IT teams build custom AI solutions is strong, this approach can lead to significant pitfalls that are easy to underestimate. Hidden complexities in data management, security, reporting, cost control, and governance can turn well-intentioned projects into costly misadventures.
The allure of in-house development
Empowering internal IT teams to develop AI solutions seems advantageous for several reasons:
- Cost savings: Avoiding vendor fees and licensing costs
- Customisation: Tailoring solutions to specific business needs
- Skill development: Enhancing the team's expertise and capabilities
However, these perceived benefits often overshadow the risks and challenges inherent in building AI systems without adequate experience.
Hidden complexities that can derail projects
Data management
AI systems require vast amounts of data. Managing large datasets — especially unstructured data — demands robust infrastructure and expertise. Poor data quality leads to inaccurate models. Gartner estimates that poor data quality costs organisations an average of $12.9 million annually. Combining data from disparate sources is complex, and inconsistent formats and siloed databases can hinder AI development significantly.
Security vulnerabilities
Handling sensitive data increases the risk of breaches. According to IBM, the average cost of a data breach in 2023 was $4.45 million. Regulations like GDPR and CCPA impose strict data handling requirements, and non-compliance can result in significant fines. AI models themselves can be targets for adversarial attacks.
Reporting and explainability
AI systems, especially deep learning models, can be opaque. This lack of transparency complicates decision-making and accountability. Industries like finance and healthcare require explainable AI for compliance. Without clear explanations, stakeholders may distrust AI outputs, reducing adoption and effectiveness.
Uncontrolled costs
Projects often exceed budgets. A study by McKinsey found that 17% of IT projects go so badly they threaten the existence of the company. Costs for cloud computing, data storage, and talent acquisition can escalate quickly. Ongoing support and maintenance adds to the total cost of ownership, often underestimated in initial planning.
Governance gaps
Without established AI governance frameworks, projects can drift without clear objectives or oversight. Issues like bias and fairness require careful governance. Projects may not align with business goals, leading to wasted resources and strategic misalignment.
Talent shortages
AI development requires specialised skills in machine learning, data science, and software engineering. Allowing teams to build expertise on live projects delays timelines and introduces risk. Investment in training can be lost if employees leave, taking their expertise with them.
Real-world consequences
A mid-sized financial services company decided to develop an AI-powered risk assessment tool in-house. The IT team, eager to build their skills, embarked on the project with minimal prior AI experience. The project took 18 months longer than planned and costs exceeded the initial budget by 200%. The tool did not meet regulatory requirements for explainability, resulting in significant fines. A lack of proper security measures led to a data breach affecting thousands of clients.
A survey by Gartner revealed that 85% of AI projects fail to deliver on their objectives due to issues like data quality, lack of expertise, and poorly defined use cases.
Advantages of established platforms
Established AI platforms provide expertise, robust infrastructure, security, compliance, and scalability that in-house builds struggle to match:
- Pre-built foundations: Tested and optimised components, with technical support and continuous updates
- Security and compliance: Advanced encryption, regulatory adherence, and comprehensive audit trails
- Scalability: Flexible resource management and easy integration with existing systems
- Cost predictability: Clear pricing models, reduced overheads, and tools to monitor returns
Mitigating risks if building in-house
If an organisation still opts for in-house development, mitigation strategies include starting with small, manageable pilot programmes to build experience, combining in-house efforts with platform services, investing in comprehensive training, and engaging external experts to guide the project.
In the rapidly evolving landscape of AI, organisations that recognise the pitfalls of learning on the fly and opt for more secure, efficient foundations will be better positioned to succeed and scale with confidence.