The initial excitement around artificial intelligence (AI) and generative AI in particular has faded as fear of missing out (FOMO) turned to fear of messing up (FOMU). Building generative and agentic artificial intelligence (AI) that works has proven to be an uphill battle for many organizations. At Valoir, we've seen companies spend months, sometimes years, fine-tuning language models, struggling with data integration, and crafting precise prompts to get AI to function accurately.
Even after all that effort, security remains a concern, as implementing robust guardrails to prevent data leaks and AI hallucinations requires advanced expertise. Add to that the challenge of designing a seamless user interface and fine-tuning AI responses, and managing those projects to maintain performance and accuracy over time, and it's no wonder that many homegrown AI projects stall before delivering real value.
However, now is not the time to give up on generative AI. Properly implemented, it can accelerate onboarding, reduce training and coaching costs, increase productivity, and even improve customer experience while reducing human agent-managed interaction volume. In our research on generative and agentic AI successes over the past few months, we've identified a few common themes that help drive AI success. Here are some of them:
Take a platform approach.
A lot of the initial enthusiasm and failure from generative AI came from IT teams that thought they could build their own large language models (LLMs), guardrails, and prompts. Organizations that embrace a platform-driven approach see radically different outcomes. With pre-built models, integrated data connectors, and structured prompt engineering tools, these businesses significantly reduced development time while improving accuracy. Security and compliance became far more manageable with built-in trust layers and guardrails, and user interface development was streamlined with low-code tools. Instead of months or even years of trial and error, companies using a platform approach can launch production-ready AI agents in weeks or months and achieve accuracy rates that far exceed their do-it-yourself counterparts.
Prioritize application of AI based on benefit, not coolness.
One of the common themes we hear from organizations is that they don't know where to start with generative AI and are afraid to invest in development in areas that might not pay off or be locked into an investment in one channel or generative AI project that doesn't deliver what they expect. The key to maximizing value from AI is not in picking the coolest application but the one that delivers the most value. If you take your list of potential AI projects and rank them based on four key factors: people (how many people do they touch), process (how many steps are in the process), potential (potential upside value), and price (the cost of the alternative, or what it costs today); you can quickly determine which ones are likely to deliver the biggest bang for your buck.
The benefit of the platform approach is that you should be able to quickly build, test, and pilot AI solutions in a relatively short period of time and assess their value so you're not sitting on a big investment that doesn't pay off. You should also be working with a vendor that lets you flexibly spend AI credits on their platform across different pilots and projects and helps you maximize value from your AI investment over time.
Don't let data hold you back.
Another common theme we often hear from organizations waiting to jump into AI is that their data isn't ready. I would have agreed with many of them a few months ago, but the platform approach enables organizations to rapidly ingest data and content to inform and ground AI, use low-code tools so business process owners can build their own instructions and processes for AI to follow, and spin up pilots to rapidly test and improve accuracy. They can rapidly identify data gaps or inconsistencies based on the answers their test agents are giving, effectively reverse-engineering the data cleansing and prep process. We've found that organizations reach production-ready agentic AI 16 times faster with a platform approach, largely because they can dramatically cut time to accuracy.
Focus not just on accuracy but on quality of experience.
The highly publicized failures of DIY AI set the bar for generative AI pretty low: a factually correct answer. While accuracy is important, it's really only the first step in ensuring AI can deliver quality interactions. As you think about your mid-term AI strategy, you'll also want to think about which thresholds you want to have for both accuracy and deviation from the correct response. Do some agents today have discretion to offer discounts to certain customers or make other exceptions? Moving forward, if you expect to maintain the quality of customer experiences while making some portion of them completely AI-agent based, you'll need to replicate some of these exception workflows within your AI environment to provide the same level of quality and personalization. The upside is that if you do you'll likely find more consistent application of those exceptions than you currently get with humans making the call.
Work with HR to establish HR policies, practices, and training.
AI is going to continue to transform customer and employee experiences, and working with HR to craft policies, training, and reskilling efforts will help drive both effective use of AI and the most effective use (and retention) of human resources. Only 10 percent of organizations today have development or reskilling programs for workers who might be replaced by AI, and only 30 percent have employee training on the effective use of AI. Training in areas like critical thinking is going to be key, particularly for employees who have traditionally been incentivized to minimize call duration and now have to make the choice of a shorter call or taking the time to ensure AI is giving them the correct answer. Working with HR to get the ball rolling on policies, practices, and training will both reduce risk and ensure you're making the most of your agents while leveraging AI's value.
Despite some initial disappointments, real organizations are getting real value from generative AI today, and the next generation of platforms for agentic AI are already delivering results. A platform approach reduces cost and risk, flattens the AI learning curve, and dramatically increases accuracy and quality. You can leverage that platform to manage your portfolio of AI projects and prioritize them based on benefit, quickly identify data consistency and hygiene gaps, and move beyond just accuracy to quality of experience for both customers and employees.
Rebecca Wettemann is founder and CEO of Valoir.