Want AI With Impact? These 4 Essentials Are Your Starting Point

The average person punches with a force of around 300 pounds per square inch (psi). Do you remember Rocky IV?

The one where Sylvester Stallone faces off against Dolph Lundgren’s terrifying Soviet persona, Ivan Drago? Ivan’s punch supposedly measures in at a whopping 2,150 psi. So, say goodbye to your face.

Now, that’s unrealistic — a number made for cinema. However, top-tier boxers and MMA fighters do often surpass 1,000 psi both in the lab and in the real world.

This isn’t especially surprising. Professional fighters train tirelessly to hone their bodies for precisely this task. The food they eat. The exercise they do. The routines they practice. All of these are in service to their punching power, and the work leading up to the punch is what makes all the difference.

The same could be said for almost any organization’s implementation of artificial intelligence (AI), small language models (SLMs), or large language models (LLMs). Any punch will produce force — just like any AI will produce some level of gains for your company. After all, the implications for AI in business are incredibly vast, and its integration is already augmenting a wide array of functions across every industry. That said, only the right kind of AI will yield serious impact to your bottom line. The kind backed by the right training and the right data.

How do you tell the difference? I’ve been on the cutting edge of supporting organizations with AI implementation since the very beginning. These are the four essential points of evaluation I tell every executive to use when sizing up their AI and its eventual impact on the company.

1. The size of the engine. 

Size matters. Well, sort of.

The volume of data held by an AI engine determines the number of permutations or relationships between data points. In turn, those levels impact the number and quality of insights that can be generated by the model. So, every organization should aspire to implement an AI engine that is large enough to make an impact. For example, ChatGPT pulls from 570 GB of text information, or about 300 billion words.

Conversely, bigger isn’t always better. Tests confirm that the quality of the data and the efficiency of the engine matters a lot more than the size, especially when you consider the cost advantage of employing SLMs over LLMs. Smaller models can be built from scratch, which makes them more customizable and secure. At the same time, they often lack the extensive reasoning and broader thinking of a large engine. Everything’s a tradeoff.

2. The quality & type of data.  

Did you hear about the man who went to Antarctica to start selling ice? Or the woman who went to the Sahara to start selling sand? You didn’t, and there’s a good reason for that. You can have a whole lot of something (data), but if that something doesn’t meet your needs, it doesn’t really amount to anything.

Every AI engine runs off of data, and high-quality data is absolutely essential for success. Without it, your AI models can get things wrong, hallucinate, drift, and even collapse altogether. In fact, warnings about the potential risks of bad data implementation have been coming from parts of Silicon Valley for years now, and the consequences are just now starting to become a reality for almost every business.

Does your organization need images, web-references, or financial data as part of its engine? Does this data sit in one place, or is it separated by silos? Is this data full of duplicates and missing information, or is it comprehensive and complete? These are the questions that will make or break the efficiency of your AI models, so they are questions you need to be asking right now.

3. The maturity of the engine. 

I love kids, but I wouldn’t trust a 5-year-old to make complex or high-stakes decisions. For that matter, I wouldn’t trust many teenagers to do either — even the really smart ones. Why? Because maturity matters when making decisions, and it matters in much the same way when you’re selecting or developing an AI engine.

I’m not talking about “AI maturity,” which is mostly about how prepared your organization is to implement or integrate AI solutions. I’m talking about the maturity of the engine itself. Over time, AI improves data accuracy, increases in volume, and reinforces the relationships built between data points, resulting in a more reliable and more capable engine. In fact, AI maturity is already receiving attention attention from the health care community as it looks to depend on models in lifesaving and life-altering situations.

How long has the engine in question been training on high-quality data? If the model is being built from scratch, how much training will it undergo before being trusted to sustain itself? If you aren’t asking these questions, you aren’t getting the full picture about the AI models you’re using.

4. The expertise & experience of your team or partner. 

Did you know that 80% of companies that embark on AI implementation will hit barriers relating to unstructured data? Many organizations simply can’t surmount all the challenges along the way. AI models are highly complex, data-dependent tools, so it should come as no surprise that not every problem can be easily solved by someone inside your organization. Are you partnered with AI experts who have experience with your specific industry? This can mean the difference between more delays or more progress.

Even once your models are up and running, it’s important to have someone to turn to in times of crisis. 80% of companies don’t have a plan in place for AI-related risks, and the fallout from AI collapse is potentially devastating. If your organization is prepared to take advantage of AI models, it also needs to be prepared to respond when something goes wrong.

Not just AI — AI with impact. 

When it comes time to take a punch, you want to make a real impact — you want to hit with the force of Ivan Drago. That said, without the right training and the right routines, you’ll never be able to muster up the strength.

You want your AI models to make waves at your organization — waves that roll out to positively impact every part of your bottom line. To do that, you need an engine with the right size, the right data, and the right maturity, all backed by the right team of experts. Getting there isn’t impossible, but it does require a little guidance.

Read more here.

This article originally appeared in CFOtech Magazine. Read the source article here.