Using AI to outsmart competitors
About 10 years ago, as the chief operating officer of a large ad agency, Erin Kelly could see the writing on the wall: Companies would need to find a better way than social media clicks to optimize marketing campaigns—and artificial intelligence (AI) would help.
“I could see that clicks and likes were not good sales predictors because people were clicking and liking instead of buying,” she says. “For example, let’s say you’re raising money for a charity and people ‘like’ your cause—they may feel they’ve done their part by supporting you, but they haven’t actually made a donation.”
Kelly had the idea of developing a tool that could derive insights from billions of online posts. Unlike traditional market research, which involves asking people questions they might not normally consider, the AI would scrape social media data for behavioural insights.
In 2015, Kelly co-founded Ottawa-based start-up Advanced Symbolics Inc. (ASI) with physicist Kenton White, and they began developing Polly, an AI that learns and predicts the behaviours of target markets to help create more impactful advertising and messaging.
“I knew what I wanted to do, but I didn’t have the technology,” says Kelly. “Then I met Kenton, and he was doing all sorts of cool things. He said, ‘I measure people the way some scientists measure atoms.’” She had found the right partner for the job.
Harnessing the power of AI
Many popular AI tools, like ChatGPT or Gemini, require training before they can be useful. But Kelly and White had a vision: they wanted Polly to be a “zero-shot” AI, which means it would be able to go straight to work with no training.
But building it would turn out to be a “huge undertaking,” says Kelly—one that required the right people with the right skills. According to a 2024 BDC study, only 20% of small businesses build their own custom AI tool, with most opting for out-of-the-box solutions.
The company’s staff is almost entirely technical. In addition to Kelly and White, there are nine employees, all but one with advanced degrees in fields like physics or machine learning. Led by White, they made up the research and development team tasked with developing Polly.
While the team worked, the company operated as a regular market research agency, running as a service business and gradually introducing Polly’s abilities.
Before long, Polly could sift through social media to make predictions about human behaviour for clients.
In fact, Polly’s name comes from its first project, a study of the “poli”tics of Brexit: Polly made headlines in 2016 after predicting the outcomes of both the Brexit referendum and the U.S. federal election that brought Donald Trump to power.
Businesses don’t just want to know whether people like their product. They want to know if people are going to buy the product. Getting Polly to [be able to predict that], and then making it zero-shot—that was years in the making.
Erin Kelly
Co-founder and CEO, Advanced Symbolics Inc.
Achieving a major milestone
In October 2023, Kelly and White finally achieved what they had originally set out to do: Polly became the envisaged “zero-shot” AI, and their clients gained direct access to its capabilities through “askpolly,” a self-serve AI product and platform.
“That was the big ‘A-ha!’ moment for us,” says Kelly, who recently closed an early-stage investment round with U.S. investors.
Before the breakthrough, Polly needed more supervision and training, and relied more on identifying patterns in data to make predictions. It could take the company two weeks to train Polly on something a customer wanted to know about.
“In market research, businesses don’t just want to know about sentiment—that is, whether people like their product. They want to know about stance—that is, whether people are going to buy the product,” says Kelly. “Getting Polly to achieve stance, and then making it zero-shot—that was years in the making.”
Polly is protected by 6 patents, she adds.
Kelly says the next 18 months will be exciting for her company now that Polly is available to customers on a self-serve basis.
A lot of people think all AI is generative [like ChatGPT]. But for certain tasks, there are better AI tools.
Erin Kelly
Co-founder and CEO, Advanced Symbolics Inc.
Kelly’s advice on using AI in your business
Kelly and White spent years building Polly so their company would stand out from competitors that relied on traditional market research methods. This might make you wonder if AI could give your business a competitive edge.
1. Determine if AI can help realize your vision
The answer lies in knowing what problems AI tools were built to solve, then figuring out whether a tool exists—or could be customized or invented—to reflect your vision.
“A lot of people think all AI is generative [like ChatGPT],” says Kelly. “But for certain tasks, there are better tools.” For example:
- Applied AI makes calculations based on data you provide to solve problems with real-world applications, like diagnosing a disease or managing inventory.
- Optimizer AIs improve performance by finding the most efficient solutions to problems.
- Predictive AIs use data and machine learning to forecast events, trends or behaviours. They are useful for forecasting demand or assessing risk.
2. Prioritize out-of-the-box solutions
Setting out to build a custom AI to solve a particular challenge may not be ideal for every business, says Kelly. AI is complex and the technology is constantly changing. In addition, many AI tools must access proprietary data to yield insights. Cybersecurity expertise can be essential to keep it safe.
For these reasons, Kelly recommends choosing an existing product, if possible. But if no out-of-the-box solution will do the job, an external partner might be able to customize or build a tool for you.
Kelly suggests assigning the project to a staff member who is familiar with both the AI landscape and your company’s vision, strategy and goals.
3. Watch out for these red flags when buying AI
Kelly recommends being wary of an AI tool whose provider can’t tell you how it works (a “black box” solution). A truly innovative product is likely patented—and therefore, protected—so the company should be able to tell you
what’s under the covers, she says.
Another red flag is a company trying to charge exorbitant up-front fees. Many AI tools now have quite affordable monthly subscriptions.