AI for growing businesses. Hype or help? Let's find out.

Fools rush in

Fortunately, there's a smart middle ground. Ready your organization or department. Position it at the leading edge as AI technologies rapidly unfold. And importantly, use AI to gain and add your own experience to the evolution equation. Small wins in this direction will bolster your team and build confidence and momentum in a way that rushing to keep up never will.

Focusing on the middle ground seems simple, yet can be deceptively daunting. Fortunately, I looked at the evidence and found four key areas that organizations succeeding with AI tools have in place. Let's cheat and get a head start: 1) management quality and structured operations, 2) digital maturity and data infrastructure, 3) training investment alongside technology introduction and 4) matching task to AI abilities. Because doing is a great way to grow, I'll suggest ways you might use AI for each of the four. It's like your high school chemistry experiments: you have to combine chemicals, watch them explode or not, and reformulate before you are a prize-winning chemist.

AI is different than software. When new software is introduced, it's adopted. Data is migrated or hand entered, staff are trained and work continues in this new way. AI is something owners and staff alike adapt to from the start. Its behavior and usefulness are determined in real time. Readiness is key to getting them both to work for, and not against your growth.

Setting the AI table for managers

Managers must be comfortable with uncertainty and prepared to adapt. The scripted routine and repetition of working with software will make this understandably an outsized shift for some. When introducing AI, operations that are structured and procedural will win out over those that are anecdotal, spontaneous and rely on institutional knowledge being in the heads of key players.

Good news. Whether you are reluctant, hesitant or enthusiastic, AI tools will meet you where you are. As a software development company owner, in my first attempts using LLM chatbots (I will use these terms interchangeably for platforms like chatGPT, Claude and Gemini) my prompts were vague. Responses I got back reflected my lack of clarity. It was evident from the response what I needed to do, be more specific. Not every attempt was equal. Several sessions on topics I knew well turned into a shouting match. All caps included. I reminded myself, it's a machine.

Then I discovered that I could ask the chatbot to create a prompt for me by giving it a topic, context and goal. Gold. Far less back and forth. Once you've played around and gained your own experience, you and your team are ready to go pro. How would you use these tools to reshape or tighten your operations and procedures? Pop in any set of instructions or SOPs (standard operating procedures) you have, and iterate with the chatbot to see how they can be optimized. How can you reformulate your prompts and operational inputs to reduce costs or add more value? Keeping the same prompt and changing the operational input, changing the prompt with the same operations or mixing them both affords a level of experimentation that's useful. Reminder that these are LLMs. It can take a few go-arounds to get the hang of things. Worth it.

Digital transformation … the AI way

If you sense you're lagging in the digital transformation wave, this is the time to catch up. Quickly. My research shows, you can profit from those who have gone before you. Digital transformation is a process. Wherever you are in that process, you will pull even further ahead by auditing your operations.

Look for signs of analog activity, everywhere. Do you have paper checklists? Whiteboards broadcasting the day's work? Emails requesting status updates? Meetings designed to "get everyone on the same page"? These examples are ripe for digitization.

Integrating systems (getting them to talk to each other and share data) is a key milestone in modern operations. One clever way to find where these systems aren't talking to each other is to mine processes for duplicate data entry. It tells you precisely where you need to build automations between systems. Be zealous with cleaning your data, garbage in / garbage out is magnified in the AI universe.

The path to digitization is made easier using LLMs. Prompt the chatbot with a list of department specific analog activities in your organization and then ask it for ways to digitize them. Include context like the department's size, resources, budget and company's industry.

Describe your office systems, include key context like the system's purpose and its role in operations, and prompt it for how these systems could be integrated to talk to each other. Integrations like this save time by automating and reducing errors. Combining data from multiple systems generates more meaningful reports. These in turn make for tighter decisions. Chatbots are great for identifying novel ways to improve your business operations.

LLMs are also a resource for data cleaning. If the organization's chatbot is publicly available, describe the data set and its source: spreadsheet, database or software application. Then prompt it for ways to identify duplicate data. What fields or columns in this data set typically gather duplicates? How can duplicate data be found? If you have an enterprise chatbot and it is safe to share company data, upload the data and ask it to flag duplicates for you. Clean data means fewer errors, whether by humans or AI tools. Fewer errors means saved time and more impactful decisions.

If you believe you are behind, you are not. These very situations are ones I see daily … yes, still! It is a normal place to begin. You don't start a business automating everything. That bit gets done later on once the business is validated. It's also tricky timing. Automating in the middle of growing? No time. Auditing analog practices as and when you can, as you go, sets you firmly on the way to digitization … a prerequisite to AI introduction success.

AI training and why it's your single best move

One thing I was shocked to learn from my research is that the biggest multiplier of AI productivity is training. Organizations that invest in training see nearly six times the AI productivity gains as those that don't. That's a big number! It's shocking and makes sense.

Using software, you follow a structured and sequenced set of steps. Using AI requires the ability to create context (specific and nuanced instructions), verify results (LLMs notoriously supply incorrect information) and adapt (instructions working last month may fall flat the next given changing algorithms (instructions embedded in chatbots), priorities, regulation changes, etc). If software training was a good idea, with AI it's a must.

Getting training got a whole lot easier. You can use AI to train yourself to use AI! An AI instructor was one of my first use cases for a chatbot. Ethan Mollick, author of Co-Intelligence: Living and Working with AI, has a great instructor prompt on his More Useful Things site (https://www.moreusefulthings.com/student-exercises). I've used it to learn how to craft better prompts and gain insights on the origins and inner workings of LLMs and machine learning. An AI instructor is ideal for learning how best to create context and verify results. Using the Mollick prompt, you supply "best practices for creating context" or "verifying LLM results" as the topic and the instructor will guide you through the rest. These teachings strengthen AI skills in each of the four key areas. Applied to adaptation, an AI instructor is useful for keeping up … and getting ahead!

With AI, how you use it matters … a lot

We all love a good short cut, me likely more than most. This approach with AI can get you into trouble. Fast. An easy mistake with AI is believing the hype, "AI is taking our jobs." This is hyperbolic and implies far more ability than AI has on its own.

Research shows that AI performs well with discrete, well-defined tasks. Choosing which tasks is where experimentation and critical thinking come in. It might feel natural to take a priority task as the first candidate for AI. A lower priority task allows low-risk learning about what works and what doesn't. Not a short cut, but a J-curve. As we learn this ever-changing technology, productivity dips. As we master how AI works, productivity soars.

Keeping with the discrete and well-defined tasks guideline, here's an experiment: choose three to five tasks from an existing workflow. Write up the context and reasoning behind the workflow and prompt the LLM for which of the tasks is best suited to AI automation and why. Your own experience will vet and verify if the why resonates. You will start forming a picture of these tasks operating in real life using AI. You'll know if the LLM's suggestions make sense for AI automation experiments. If yes, go ahead and start on the task that delivers highest impact or matters the most. If not, refine the next set of tasks you submit based on learnings from the first set. Combining and reformulating in this process is exactly how you will achieve award-winning results. Eventually you will be more precise in where you start and consequently get the most from the LLM as an automation tool.

In two words, get ready. More is on the way … in a hurry.

In adopting a new technology, knowing where to start is a challenging obstacle. Fortunately, research identifies four key areas that pave the way for successful AI adoption. By learning how these tools work through experimentation and adding your own experience, you gain a strong understanding of how to deploy AI tools now and in our rapidly changing future. The work can feel grinding, frustrating and confusing at times. Your own and your team's experiences and experiments will prove that out. That's the J-curve at work. Pushing through has tremendous payoff. And, there is a lot of fun, insight and wow to be enjoyed along the way. Focusing on the four key areas and their practices will prepare your organization well for the next phase … from LLM chatbots to agents.