Forecasting and Expectations

“Artificial Intelligence” has become one of the most overused phrases of our time. It’s driving some tech stock prices, and has become a hot topic of conversation, and simultaneously a hopt topic for misunderstanding. While a lot of this is fun, it can also be a time-waster for busy non-technical managers, who- in our opinion – could use some guidance in understanding capabilities, and setting priorities that realistically propel efficiency and profits.

So in a brief two-parter, at the risk of repeating things many of you know, we’ll offer a quick recap of definitions and a couple of best practice pointers that are helpful at Vertical Circle as we support clients.

The Layers of Intelligence

Artificial Intelligence (AI)

AI is the broadest term, coined at the Dartmouth Conference in 1956 (Go Green). It refers to machines designed to perform tasks that typically require human intelligence—things like reasoning, planning, problem-solving, or natural language understanding. Think of AI as the umbrella under which everything else falls.

Machine Learning (ML)

Machine Learning is a subset of AI. Instead of programming explicit rules, we train models with historical data so they can learn patterns and make predictions. For example, predicting customer churn, classifying invoices, or recommending products.

Deep Learning (DL)

Deep Learning is a specialized branch of ML. It uses multi-layered neural networks to automatically extract complex patterns from data. It’s the technology behind facial recognition, self-driving cars, and large-scale language models, and much more.

Put simply:

  • AI = the goal (simulate human intelligence).

  • ML = the approach (learn patterns from data).

  • DL = the advanced technique (neural networks with many layers).

AI complexity escalates fast as more money pours into the field, and marketing promises arrive with legitimately expanding functionality. For example ChatGPT and its friends are a big deal, but arguably not as important to organizations as techniques for modeling Continuous Improvement or predicting product growth or new engineering development. AI also helps

developers with shortcuts in complex or long code, doctors with surgery, farmers with irrigation, and the list of functionality gets longer by the day.

Why Forecasting Doesn’t Matter Without Good Data and Human Intelligence (HI)

At Vertical Circle we work with industrial clients seeking visibility into revenue, gross margins, product and sales team success, cash flow, and many other areas. We also as part of this do forecasting, and sometimes, sadly, have to tell our clients that we don’t have magic wands.

In other words, AI really isn’t anything new, it’s just getting more advanced, sometimes quickly.

Which is why the above definitions are helpful to keep in mind, and the fundamentals haven’t changed very much.

  • Clean, Harmonize: Poor, incomplete, or biased data will lead to unreliable predictions. Ensure it can be used openly, for example with neural network standards

  • Historical Data Matters (A LOT): Forecasting relies on past behavior. If your history is inconsistent or sparse, your model has nothing solid to learn from. The more the better, but three years is a good start.

  • Meaningful Trends: It’s not just about collecting data—it’s about capturing the right signals. For example, recording every website click may be noisy, but tracking purchase conversions gives real insights. Or: focus on your most or least profitable items and units, and attack priorities.

  • Models are Secondary: Many teams spend months debating which algorithm to use—Random Forest, SARIMA, or the latest deep learning architecture. But actually, a simple linear regression, or basic SARIMA with clean, consistent, trend-rich data will often outperform a cutting-edge neural network trained on messy, irrelevant data. Don’t perseverate on models.

In the end, no matter how sophisticated your algorithms are, they are only as good as the data you feed them applied to organizational (human) experience.

That’s right! We still need HI, and in fact, more than ever. A computer doesn’t know why one customer spiked weirdly last August; doesn’t know the key quirks of our products, customers, of our organization. A computer doesn’t really know what the CEO really needs and wants to know, and when.

In other words, nothing about AI shortcuts the need to invest in, to hire and retain, the best possible people we can.

Khanh Kieu