The 3 Steps to Enterprise AI Readiness: A Practical Guide

The talk about AI has gone from science fiction to DIY, seemingly overnight. First, we saw the roll out of ChatGPT, then the announcement of Microsoft 365 Copilot, and finally…. Bam! Now, everything is AI. There is even an AI toothbrush and an AI-powered beehive. With so much hype, it’s not surprising pressure is mounting within organizations to ‘do something’ with AI. It’s not a stretch to say that business today is experiencing AI FOMO. That’s why a 2023 Gartner report finds a whopping 85% of organizations are in early stages of exploring AI, with 70% investigating and 15% piloting Generative AI programs.

But before you rush off to install AI tools, you need to be aware of the risks and dangers of rolling it out willy nilly. Your first inclination might be to expose ChatGPT to employees either via the native OpenAI interface or via enterprise apps using APIs. While this might acclimate employees to the experience of Generative AI experience, it will not unlock the business value from information contained in your existing content and data stores. Plus, inherent risks, challenges, and costs of working with a generic LLM may not be worth the trouble.

Start with Microsoft 365

If you are a Microsoft 365 shop, the natural place to start your enterprise AI journey is Microsoft 365. Your M365 tenant already contains a treasure trove of valuable, business-rich content in the form of documents, emails, and Teams interactions between colleagues… all of which can be mined to provide valuable business insights using AI. For example, a lawyer or consultant could ask an AI Assistant for a list of acceptable terms and conditions that can be included in a contract with a partner, based on previous agreements.

Understand the risks: what could go wrong?

Regardless of where you begin your AI journey, all AI solutions require you to address both security risks and data governance challenges. Security risks associated with exposing sensitive information to unauthorized personnel inside and outside your organization, and data governance challenges associated with providing the AI engine with only up-to-date and authoritative content so that employees attain accurate, reliable, and contextually relevant answers.

The reality is that few organizations feel ready today for AI. A McKinsey report found 56% of organizations fear inaccurate answers as the biggest risk. Furthermore, only 21% of organizations have policies governing employee use of generative AI like ChatGPT; policies that cover factor such as verifying AI model sources, aligning outputs to intended topics and audiences, validating facts and ethics, reviewing content before publishing, conforming to regulations and standards, and weighing costs.

With these concerns and challenges in mind, let’s examine the three steps you need to take before exposing employees to enterprise AI.

The 3 Steps to Enterprise AI Readiness

Step 1: Limit access to sensitive content

It has always been important to institute high-quality content management practices, even without AI. Yet, for most organizations, content oversharing and data governance is an ongoing challenge. Content oversharing happens when information is exposed to unintended and unauthorized personnel, either intentionally or accidentally. The best practice is to apply ‘just enough’ access to documents, so that people have access to only what they need to do their jobs… and no more. But most organizations have been lax in enforcing access control because it requires consistent vigilance to get it right.

While oversharing is not a new problem, in the past, it was less acute because employees would usually not find unintended information because they didn’t know to look for it. With AI, unintended information is now delivered by AI as a matter of course, in response to naïve questions. Because the impact of oversharing can be catastrophic, it has become imperative to overcome this challenge before introducing AI and exposing employees to AI tools.

The solution is to limit access to content based on job and role requirements using knowledge management best practices. Practically speaking, you can address this with a next generation, intelligent knowledge management platform, installed to your Microsoft 365 tenant - and connected to multiple data repositories - before giving employees keys to the AI kingdom.

Step 2: Set up auto-tagging

With or without AI, the richness of the Microsoft 365 experience depends on data sources being properly and comprehensively indexed. To index content properly, it must be properly classified using metadata. The problem is that this task typically falls on busy employees, who often won't spend time tagging content every time they create a new page or document. Plus, most employees don’t know how to apply tags to accurately classify the content. Yet, without this critical step, AI engines struggle to correctly interpret content and context of business-critical documentation. Without accurate classification, the possibility of AI hallucination increases, while the probability of generating contextually relevant answers is reduced.

The simple solution would be to have employees accurately tag documents. Since - being realistic - this is an impossible expectation to set, the practical solution is to employ an intelligent knowledge platform to automatically tag content when created or uploaded.

Step 3: Establish authoritative content with knowledge collections

In the normal course of business, SharePoint, Microsoft 365, and other document repositories often turn into "document graveyards".

That’s because employees typically work with multiple versions and copies of documents when they share them as email attachments with colleagues and partners. The result is multiple versions of the same document stored in SharePoint with names like ‘Contract1_1_1_1_final’.

When this happens, it is impossible to know which document version represents the latest and most authoritative copy. When these documents are contracts, policies, or other business-critical files, the consequences are dire. Then, when AI engines tap data sources to probe for answers, they process data from all copies of these documents, leading to unreliable and inconsistent answers.

Said differently, if it is not clear and easy for end-users to locate the correct versions of content, e.g. through a search, why should we expect AI to do much better. It might do "just as well" but faster if we are lucky. 

The solution is to build sets of authoritative knowledge collections for the AI engine to tap.

Practically speaking, the way to establish sets of authoritative documents is to create knowledge collections, where knowledge owners and subject matter expects are in control of "what is given to the AI".

Atlas AI Starter Pack gets you ready for enterprise AI

The Atlas AI Starter Pack is a limited time offer that gets your organization ready for AI by delivering the following three unique, enterprise-grade capabilities. With the Starter Pack, ClearPeople experts work with your team to:

1. Deploy the Atlas Intelligent Knowledge Platform (including configuring Azure OpenAI), set up Atlas auto-tagging, and configure the Atlas experience in SharePoint Online (SPO) and Microsoft Teams.
2. Configure an Atlas Knowledge Collection for your organizations, ensuring security and proper settings. Our experts also identify and load authoritative documentation, creating a single source of truth across multiple knowledge bases.
3. Enable Atlas to provide governance of permissions-based usage and cost controls, logging of user actions and prompts, and analytics of usage.

With the AI Starter Pack, your organization gets prepared for and experience with Generative AI inside Microsoft 365, in a matter of weeks, and with minimal effort. Register for the Starter Pack now.