The euphoria around AI and its promise of productivity in organizations has passed its peak of hype. With a solid year and a half of AI experience under their belts, CIOs now understand that AI is not the panacea it first appeared to be. Initially, it seemed like an instant miracle, but simply bolting ChatGPT onto enterprise systems didn’t cut it.
Monumental AI flops have been all over the press. Take the now-infamous case of the law firm that was fined by a judge after its lawyers blindly relied on ChatGPT and submitted a motion with six completely fictitious case citations (ouch!). If only that were an isolated incident, but it’s not. Using ChatGPT in a capricious, unchecked manner continues to happen. Two more similar examples have surfaced recently, one in the US and another in Australia; both involving major firms. And it’s not just happening in the legal space; those stories just went public due to the nature of the infraction. The reality? These are just the tip of the iceberg. It’s fair to assume that the damage caused by indiscriminate use of ChatGPT and other publicly available AI tools is widespread and detrimental to businesses worldwide.
But are organizations faring any better with enterprise AI tools? Apparently not. Case in point: there is widespread disappointment with tools like Microsoft Copilot, which hasn’t performed as promised. So, what’s the problem? And more importantly, what do we need to do to fix it?
Let’s break these down.
Get your data AI-ready. AI success depends on structured, well-governed data. If information isn’t cataloged properly by providing context, removing ambiguity, and ensuring only the most recent, approved, gold-standard content is consumed by the AI engine, you’re likely to end up with inaccurate and misleading answers. To prepare for AI readiness, organizations must tag content with metadata and define which content qualifies as ‘gold standard’ so the AI can generate reliable, high-quality responses.
As we’ve seen from those published horror stories, one of AI’s biggest problems is generating false, inaccurate, or outdated answers. Sifting through AI-generated content to separate fact from fiction negates AI’s intended efficiency. To ensure AI-generated answers are useful, organizations need a way to tap into only the most authoritative, up-to-date content within their enterprise.
‘Garbage in, garbage out.’ Knowing how to ask the right question, along with providing the right parameters, constraints, and context makes all the difference between getting great answers and getting… well, garbage. The key to success with AI? Generating the most effective prompts. But expecting typical business users to master prompt engineering isn’t realistic. Instead, AI tools should have built-in mechanisms that automatically apply the appropriate ‘wrappers’ – context, constraints, and compliance restrictions, so users get the best possible response without needing to be AI experts.
One of the biggest disappointments with AI so far? The cost. The hype-sters didn’t exactly highlight that part, but AI’s high cost is limiting its use and potential for growth. And with cost comes another unintended consequence: AI’s impact on an organization’s carbon footprint. For AI to be viable in the long term, solutions must be both cost-efficient and energy-efficient, ensuring scalability without breaking budgets… or the planet.
Atlas is purpose-built on Microsoft 365 to address the GAPS in enterprise AI, making it easy to get AI-ready and ensuring businesses get authoritative, accurate, and reliable answers without needing to learn new tools. By covering critical Microsoft gaps and delivering contextual AI solutions for key verticals like legal and professional services, Atlas enables businesses to extract real value from AI today and sustain their competitive edge as they scale to meet tomorrow’s challenges.