“You’ve got to use the “new thing” to do old things better. Then, you use the new thing to … do new things.”
– N.N., Microsoft Employee

It’s been almost a year since I penned a WR titled „There’s an AI for That.“

The message appears to have resonated. According to a survey by Microsoft and LinkedIn, around 70% of employees currently use AI tools, although their company does not provide them.

And it shows. Last week, a client told me that posts on her company’s social intranet had recently improved considerably in readability and form. Even colleagues known for their limited mastery of English now published well-formulated posts. However, she added, there was a certain uniformity to them.

This contrasts starkly with recent Deloitte findings, indicating that only 14% of managers globally anticipate immediate impacts from GenAI usage.

However, 68% of employees feel overloaded (Microsoft/LinkedIn research) and, for that reason, take things into their own hands, even if policies are missing or limiting AI use. Programmers and communication functions stand to gain in the order of 50% productivity, so they are taking to it: permission, mandate, strategy, or not. And the rest is not far behind.

My clients‘ BYOAI (Bring Your Own AI) outstrips the typical cautious trials provided by the IT departments.

So, even if you are one of the 73% of German managers who feel unprepared for AI (Deloitte), it’s time to take a serious look.

We see five general use cases where AI and Machine Learning excel

Making sense of complex data: discovering intricate relationships between factors and making predictions, even when there are gaps in the data. Various types of forecasting are commonly cited applications.

Classification and tagging: assigning labels or categories with high accuracy and consistency. One of our clients is currently exploring the creation of inventories of the components of undocumented machinery installations based on their images.

Information retrieval: scanning through unstructured, often textual, data and summarizing relevant information. Condensing actionable product requirements from customer feedbacks is being trialed at a company we are in discussions with.

Coding: GitHub Copilot, ChatGPT, Tabnine, and others vastly enhance developer productivity by providing feedback, suggesting snippets, and even writing new code based on natural language text prompts.

Creating new content based on existing content: drafting job descriptions, summarizing meeting transcripts, writing intranet posts, and even brainstorming mission statements. This is where we currently see the most significant productivity enhancements for office workers. 

Applying AI tools daily has taught me a few things

Start with an outcome. Don’t get invested in tools and technology too early. Consider the benefits you want to achieve for yourselves or your customers. Where can AI help unlock new levels of productivity, improved service, or other customer benefits?

Experiment and learn. Embark on an agile discovery journey of limited experiments and bets to pursue your desired outcome, and be ready to pivot quickly. Brace yourself to be amazed or utterly disappointed. I recently created a brilliant animated explanatory video clip in less than ten minutes and saw very disappointing image recognition results only a few days ago.

Don’t try to “fully understand“ or possibly even rebuild the underlying technology.“ Understand how it behaves, not how it works.“ as Bloomberg’s Zane van Dusen recently put it at a conference. Experiment with different models, consider your security requirements and use what makes sense in your context.

Input data is everything, so it’s worth investing in it. When adding your own data, ensure its accurate, consistent, and as unbiased as possible. A few months ago we created a Product Management bot using RAG (retrieval augmented generation) and an approx. 400-page corpus of materials we wrote or edited. However, its responses were heavily biased towards a specific method. Upon reviewing the input materials, I realized that I hadn’t balanced them carefully enough and had provided increasingly detailed materials on the particular approach. With machine learning, data integrity becomes an even more fundamental requirement for building effective and reliable systems. Without it, the results produced by these systems may be inaccurate, unreliable, and potentially harmful.

However, unlike other advanced technologies in history, AI is very accessible, even for non-specialists (like me!). Many of the available tools offer free trials. In addition, there is a large body of Open Source materials. Considering the potential gains, it’s worthwhile to identify your key use cases and move them from informal and unauthorized use to a more strategic – and compliant – approach.

Here are two additional perspectives:

Build a Winning AI Strategy for Your Business

Microsoft EVP Christopher Young shares his thoughts on how to do it.

Click to view!

How to Implement AI in Your Business: A Step-by-Step Guide

Single Grain’s Eric Siu prefers a more hands-on approach. Despite the marketing focus, the article offers numerous ideas for use cases across an organization.

Click to view!