I find it useful on occasion to interview people who are marketing-adjacent, and thus possess a different and perhaps more highly clarified perspective on the tools we use for data-driven online advertising. These are the supply-chain managers, data scientists, merchandisers, finance analysts and others who must either appraise the CMO’s performance or are an internal stakeholder to the CMO.
One such recent briefing was with Joe Torreano, VP of product and UX for outdoorsy gear and clothing online retailer Backcountry. He joined Backcountry more than 10 years ago in call center customer support, moving up through sales and product management until taking over product and UX responsibilities a year ago.
We discussed the brand’s extensive roster of software vendors – Torreano juggles software contract management for the company “and it takes way too much of my time, unfortunately,” he said – and we zeroed in on the recent beta test of a product called Commerce Graph from commerce ad tech and analytics company FERMÀT.
Story of a SaaS startup
Backcountry’s onboarding of FERMÀT is a useful case study not just of the beta-tested product in question, but of the delicate balance between online sellers and their SaaS vendors.
Initially, Torreano said, FERMÀT pitched its services as a way to dynamically personalize product display pages when people clicked to the Backcountry site via a social media ad. As a low-stakes way to test the vendor, Backcountry handed the startup a few underperforming segments to see if there would be a turnaround.
It worked. And so Backcountry gave FERMÀT a chance to outperform its in-house dynamic page rendering product for Google product listing ads.
FERMÀT then asked Backcountry to participate in the beta test for its new product, the aforementioned graph, which uses a pixel on customer sites and/or apps to connect a customer’s behavior with their broader activities on social platforms and elsewhere.
“I was skeptical,” Torreano said, and the first iteration of the graph was, admittedly, “kind of rough” in the way of a beta product.
But the product improved drastically over the course of the beta test, he said. Now, it has useful analytics like heat maps for the homepage and product display pages that demonstrate how different elements on a page affect user sessions or if there are points of frustration for shoppers.
For example. does traffic via email campaigns degrade faster compared to other traffic sources? Or does featuring certain product listings on the homepage improve the overall average order volume?
These are the types questions that the product can answer, Torreano said, including via agentic-style natural language processing.
Backcountry had access to similar features from legacy vendors that it had been paying hundreds of thousands of dollars per year for, he said. The FERMÀT Commerce Graph did the job just as well in about six months, he said, and also provides more granular tracking since it has pixels and the graph spans outside platforms.
Also, Torreano said, other SaaS platforms and even the big enterprise LLMs – Backcountry uses Anthropic’s Claude among other tech – aren’t necessarily adept at ecommerce marketing or product management. He might do large data dumps to help a Claude agent get a full understanding of how the site works, but Claude can’t reliably execute on the nuances of ecommerce.
For example, he said, traffic that comes from social media has lower intent and brand familiarity, in general, compared to organic site visitors. A human understands this. Someone who came via a social post may have mistakenly fat-thumbed an ad or clicked an influencer post that isn’t clear about what kind of site is being linked to. They may not have known the company at all a minute prior.
A human gets the context for why those social visitors won’t convert at the same rate as organic traffic, not to mention having different kinds of baseline thresholds for when certain kinds of traffic really are underperforming.
Torreano said Backcountry has gotten great value out of the Commerce Graph test, in no small part thanks to being able to replace far more costly SaaS solutions. To be fair, though, because Backcountry is a beta partner, it’s currently getting the tech for free.
“I don’t know the details on how they’re going to end up charging for it,” he acknowledged.
He does, however, have a sense of where the broader market is headed.
Eventually, Torreano said, the entire crowd of martech and ad tech vendors may be ready for a real agentic overhaul.
Right now, there needs to be a human to vet the AI output, he said. Agentic software doesn’t yet understand the particulars of ecommerce traffic and product management. It also can’t replicate human intuition. A person can see, for instance, when something happening on TikTok is affecting on-site conversion rates, or recognize that traffic from email, search and different social channels naturally behaves differently.
But once the big LLMs and vendors get better at teaching these human instincts and known factors to agentic tech, he said, we’ll probably be able to start letting agents make decisions without a human approving each one.
It’s not hard to see from what the LLMs are building that they’re moving toward agentic tech optimizing on the fly, what Torreano called “human out of the loop.”
“This isn’t how I expected this conversation to go,” he quickly added. “I hope I’m not getting in trouble.”
