A “No Regrets” Approach for Tech Leaders
We’re all familiar with the feeling of buyer’s remorse. Whether it’s the realization that a purchase may not have been the best choice, discovering a better alternative after it’s too late, or feeling overextended financially, these regrets can take many forms.
Amid the frenzy of the AI boom, some tech leaders have experienced similar sentiments about their AI deployments. Many projects have delivered lower-than-expected ROI, incurred soaring costs, or simply failed to achieve their intended outcomes.
How can tech leaders ensure their AI deployments lead to success rather than regret? A thoughtful, phased approach where costs, risks, and expectations are managed upfront is essential. Here are three key considerations.
Regrets from AI deployments usually stem from being underprepared. Many business leaders do not realize the incredible complexity and surprise costs associated with managing AI systems. Tech leaders should ensure their team is fully prepared for AI’s demands well before deploying these technologies.
Ideally, in-house teams should possess the necessary technical proficiency to autonomously manage the full lifecycle of AI services. This includes skills in areas such as machine learning and AI operations (e.g., containerization for model deployment, CI/CD pipeline and model registries, ETL pipeline optimization, model auditing frameworks) as well as proper security measures including authentication, data encryption and comprehensive logging of model actions.
An unqualified or ill-prepared team can lead to unexpected, ballooning costs — for example, leaning on externally managed services that charge by the hour — or the failure of the entire project, or both.
By doing their homework to understand the needs of their AI projects and laying the groundwork to ensure the proper support system for those needs is in place, tech leaders can set the stage for success and scalability in their AI deployments.
Tech leaders should start small on their AI deployments by focusing on automating low-risk, repetitive tasks that require minimal expertise but deliver measurable productivity gains. This type of deployment might include appointment scheduling or an email-based workflow that sends automated responses.
By taking a measured approach to AI in a relatively controlled environment, tech leaders can not only wrangle costs but also foster trust in AI systems within the organization. As the AI deployment proves successful and confidence grows, organizations can gradually scale these technologies to handle more sophisticated and critical tasks – for example, using AI to enhance collaboration with partners, vendors, and customers by providing shared, real-time insights.
I also encourage tech leaders to consider investing in or building small language models (SLM) wherever possible. SLMs offer significantly greater cost efficiency versus large language models (LLM), as they can be hosted on cloud platforms, sometimes at one-tenth the cost. While SLMs require more extensive training compared to LLMs, which often work “out of the box,” the long-term benefits are substantial. Once properly trained, SLMs can replace LLMs, providing a more cost-effective solution that frees up budget for additional AI initiatives and deployments.
Here again the skillset of your team will be important, as they need to fully grasp the multiple approaches to training SLMs like retrieval automated generation (RAG), retraining with labeled data, and reinforcement learning online networks.
While AI can efficiently handle routine tasks, its performance in complex, unpredictable situations is still imperfect. This, combined with rumblings about job replacement, has resulted in cautious — if not outright cynical — perceptions of AI.
If a tech leader has any hope of implementing a successful AI strategy across an organization, they must address and minimize these concerns.
In the face of distrust, tech leaders must ensure any potential AI malfunction will have as small an impact as possible. Specifically, strategies like monitoring and early detection, fallback to a rules based deterministic system, multi-agentic framework, and chaos engineering will be absolutely critical. Leaders should develop robust contingency plans early on and ensure human oversight for critical tasks.
And in the face of disruption, tech leaders should thoughtfully plan for changes in roles and workflows. Personnel plans should allow for continuous learning and adaptation — retraining and repurposing employees for higher-level tasks as AI adoption grows across the organization — as well as consistent and clear communication so employees know exactly what to expect.
Far beyond the typical technological advancement, AI has launched a profound shift in how businesses operate. In the mad AI rush, it’s been all too common for tech leaders to launch underdeveloped initiatives that end up falling flat.
But there’s no need for regret. By deploying AI thoughtfully — starting with small use cases, managing costs and risks, and preparing their teams for inevitable changes — tech leaders will better position their organizations to reap AI’s long-term benefits.
