One of the biggest issues with major LLMs is that they’re pretty much black boxes.
No one really knows the full extent of what’s going on inside, said Aniket Deosthali, CEO of agentic commerce company Envive. At least, that was true in the past, until companies like Envive made it their mission to help brands optimize how they’re showing up in LLMs.
Envive is a commerce-focused AI company that helps brands improve how they show up within AI search and develops brand-specific agents that marketers can deploy on their websites.
By studying a combination of past campaign results, CRM data and third-party data, Envive develops a clear understanding of a brand’s customer profiles. This understanding allows marketers to shift their messaging and content strategies accordingly and improves their odds of ranking better on answer engines, Deosthali explained.
Just “guessing what consumers are saying” isn’t effective, he added – which is why Envive also builds out on-site agents for brands like footwear company Clove. Marketers have full access to what their site viewers are asking the agent and can adjust their content and the agent accordingly.
Feeling chatty
Clove has been working with Envive since July 2025, when it first introduced a widget on the site that connects users with a brand-trained chatbot.
In the past, if people had questions about Clove’s products while perusing the site, they had to submit a ticket to Clove’s team and wait for someone to respond manually. It took time and resources for Clove to answer those questions, said Nick Sanetra, Clove’s director of marketing, and those interactions didn’t always lead to purchases.
Now, the chatbot answers those questions for consumers in real time.
Over the past nine months, Clove has seen a “demonstrable increase” in the average revenue per user among those who have been exposed to the widget, compared to those who haven’t, said Sanetra.
The widget appears on product description pages as well as collection pages and functions as a shopping assistant that can answer questions ranging from sizing guidance to what makes Clove’s products distinct from those of its competitors.
When someone clicks on the widget, several suggested questions appear. Depending on what page the customer is currently viewing, the questions are either broadly about the brand or specifically about a style or collection.
The suggested queries draw from previous searches, Deosthali said, and use “reinforcement learning” to understand what details customers care about. For clients like Clove, who are only using the chatbot feature (or “sales agent,” as Envive calls it), the result is an improved chatbot with more precise query suggestions.
Clients who are also using Envive for GEO can take advantage of the platform’s asset creation and copywriting agents, which use customer search data to help develop content like FAQs and more targeted landing pages.Clove is also able to directly access its chatbot’s question history, Sanetra added, so it can see exactly what its customers are asking.
Break it down
To create the first version of the chatbot, Envive’s AI crawled Clove’s website, including product descriptions, sizing guidance and collection pages, to understand the “raw content” of the site, said Sanetra.
Initially, only a small fraction of site viewers were exposed to the widget as Envive ran tests to ensure that it wasn’t somehow creating friction or harming conversion rates, said Sanetra. Over time, Envive increased the number of viewers seeing the widget. Now, about 95% of site visitors are exposed, with a 5% holdout for testing.
However, Envive didn’t initially break down its reporting by juxtaposing those who were exposed to the widget against those who weren’t. Instead, it separated people simply by whether or not they had engaged with the widget. The (extremely positive) results were likely skewed by response bias, said Sanetra, since people who choose to engage with the widget are likely more interested customers in the first place, with more specific purchase questions.
When Sanetra brought up the concern and asked instead for a breakdown of exposed vs. unexposed, though, Envive was quickly able to provide the requested results.
The original reporting, comparing engaged vs. not, showed a 5x increase in add-to-cart and checkout completion. In reality, Sanetra said, the increase was closer to 5%.
Both methods are “absolutely fair” ways of measuring performance, Deosthali said. Envive opts to measure by engagement rather than exposure to see the direct impact of a “guided selling experience,” he added. As to why both reports aren’t included on a user’s dashboard from the start, Deosthali said Envive wants to “ensure consistency in measurement,” and thus only offers engagement-based reporting unless asked otherwise.
Overall, the partnership has been a great asset for Clove, Sanetra said. Envive has been “proactive” in suggesting new tests, like exactly where on the page the widget appears, and automatically optimizing for the best performance.
“They’re very responsive,” he added, and will provide “pretty much anything you ask for from a reporting standpoint.”
