In our enthusiasm to embrace generative AI for content creation, we often overlook all the other ways we can apply it. Case in point: predictive analytics and forecasting.
Forecasting in a top-of-funnel demand generation context means predicting “when something is likely to happen,” says Christopher S. Penn, chief data scientist at Trust Insights. You can apply that approach “to any data which is a time series data set,” like search traffic, topics discussed on social media, or even seasonal sales trends.
Such predictions aren’t guarantees but “what could happen with a certain amount of probability.” And that may be enough to help you decide what to do next and where to put your money.
When you add generative AI to the equation, forecasting becomes more accessible to marketers; it’s no longer the mysterious realm of analysts and data scientists.
The key is having access to data, choosing the right tools, and understanding how to write effective prompts.
Data Access
First up: data. Generative AI or not, you’ll need time series data to make the predictive analytics magic happen.
According to Chris, a time series data set is “numerical in nature, like the number of people who search for ‘demand generation’ on Google Trends,” measured in weeks, months, quarters… any unit of time.
And “for every period forward that you want to forecast, you need three periods of back data. So if you want to forecast next month, you need three months of back data.” At least. More back data is better for more accurate forecasting.
For the top-of-funnel awareness stage, for example, choose sources such as those that can be found in Google Search Console or your social listening tools. Again, use only time series data.
Chris goes into further detail in this clip from his recent MarketingProfs presentation, “How Predictive Analytics and Generative AI Help You Forecast and Plan Your Future Marketing Success”:
How much forward should you forecast? Chris recommends two times your decision frequency: “If you need to forecast a year in advance, you should forecast at the very minimum in six-month increments. So, at the half-year point, you can look at the forecast” to determine how well things are going. Were your predictions on point? Great! Not so much? You still have time to make changes to your plans.
The Right Tools
Next, select your tools. If you have ready access to business intelligence software, such as Tableau (Salesforce) or Power BI (Microsoft), AI-enhanced predictive analytics and forecasting capabilities are built in. But if you don’t, you’re not out of luck; Excel and access to a common large language model (LLM) interface will do.
“The platform I recommend for most people who do not have technical skills is ChatGPT,” says Chris. “The paid version is best, but the free version, now that it has access to the ChatGPT-4 Omni model, is pretty good.”
It’s true: generative AI tools like ChatGPT can’t do math—at least not directly. But they can write code. And code can do calculations.
The All-Important Prompt
Now that you’ve rounded up your data and selected your generative AI tool, it’s time to flex those prompting muscles.
Chris recommends using the RACE method for prompt writing—Role, Action, Context, Execute.
“You specify who you want the AI model to be, the action you want it to take, the data that you’re providing (that’s your context), and then execute the format you want it to have.”
(If you’re new to this method, check out his team’s prompt and power questions sheets for help getting started.)
And, he continues, “the more relevant, specific words you use, the better your prompts will perform. So, if you’re doing predictive analytics and have a good vocabulary about how to ask a machine to do predictive analytics, you will get better results than just saying, Hey, I want to know what’s going to happen next week.”
If knowledge of major predictive methods is not your forte, take a few minutes to research before you get going. (Hint: start by asking AI to explain it to you.)
Using the language of predictive analytics and asking ChatGPT or other AI to explain what it knows about these methods at the beginning of your session also helps prime the AI. To that end, you should tell ChatGPT, “Your first task is to explain what you know about best-practices for forecasting time series data,” Chris says.
That “is going to do what’s called priming the model. The model will then load all of its knowledge about time series forecasting and fill in the hundreds of words you should provide to get it to forecast,” Chris explains. Ultimately, priming helps deliver better results.
Then, load your CSV file and ask it to step through the different types of analysis. Don’t have a technical or statistics background and don’t understand the analysis you see? Ask it to educate you on reading statistical analysis. “At any point, you can say, ‘Hey, explain this to me. I don’t know statistics. Is this good? Is this bad? Is this forecast reliable?'”
Asking those questions is a good thing,” says Chris: “It allows more relevant keyword text to flow into the session,” and it “can encourage the model to check its work. Even though it’s writing code, the model still can hallucinate.”
And just like that, generative AI has done the predictive analytics work for you.
But that’s not the end of how generative AI can help with your forecasting. It can also recommend what to do next based on your predictive analytics. Check out Chris’s full AI for Demand Gen Marketers presentation for an in-depth discussion about doing that.
More Resources From the AI for Demand Gen Marketers Series
Can AI Save You From Marketing Inferno?
AI Use Across the Customer Journey Means Aligning Across Teams
Using AI to Build Your Personas: Don’t Lose Sight of Your Real-World Buyers
AI Can’t Write Thought Leadership (But It Can Do Something Else)
