Himanshu Jain, Co-founder and Head of Products at CommerceIQ – Interview Series

Himanshu Jain, Co-founder and Head of Products at CommerceIQ, leads product strategy and development with a focus on building AI agents that automate complex retail and ecommerce workflows. Since co-founding CommerceIQ in 2017, he has helped shape the company’s approach to applying data, automation, and AI to digital commerce. Before CommerceIQ, Jain held product strategy, business development, and customer success leadership roles at Boomerang Commerce, where he worked with major retailers on pricing, assortment, and ecommerce growth strategies. Earlier in his career, he was a management consultant at A.T. Kearney and spent nearly five years at Capital One UK across product management, credit risk, marketing strategy, analytics, and statistical modeling. He has also advised startups on go-to-market strategy and contributed articles on Amazon and ecommerce to Marketing Land.

CommerceIQ is an AI-powered retail commerce platform designed to help consumer brands manage and optimize their performance across ecommerce channels. Its platform brings together sales, retail media, digital shelf, content, inventory, and competitive data, giving teams a unified view of performance across more than 1,500 retailers and 85 countries. CommerceIQ has increasingly focused on agentic AI through its AllyAI technology, with specialized agents that can analyze retail signals, recommend actions, automate reporting, optimize advertising bids and budgets, and execute selected ecommerce workflows within defined controls. The company positions this approach as a shift from dashboard-driven retail management toward continuous, AI-assisted execution across sales, media, content, and the digital shelf

You’ve spent more than a decade working across product management, customer success, business development, and more before co-founding CommerceIQ. What made you get into agentic retail, and which lessons from those earlier technology transitions have shaped how you view this one?

Most software before agentic AI was designed to make a human’s job faster. Excel helps you build a model faster. A dashboard helps you make a decision faster. What changed with agentic retail is that software can now do the work itself, not just speed up how a human does it. An agent can learn your environment the way a new hire would. Your processes, your internal guidelines, how you actually make decisions, and then act toward a goal end-to-end, with a human providing context and auditing the output rather than doing every step. That is a completely different category of software than anything I worked with earlier in product management or customer success, and it is why I think this is the most fundamental technology shift I have seen, not an incremental one. The pace is also unlike anything before it. Smartphones took years to become ubiquitous. Agentic AI has changed how entire industries work and it is not slowing down. The lesson from earlier transitions that I take most seriously is that people who project from the past make the wrong calls here. The approach I have found works best is to keep an open mind, get your hands dirty, and learn from what is actually happening right now rather than what happened in the last cycle.

You’ve drawn a distinction between “agentic commerce,” where consumers use AI agents to discover and purchase products, and “agentic retail,” where brands use agents to manage how they sell. As both sides become increasingly automated, how do you see the relationship between consumers, brands, and retailers changing?

Retail discovery has gone through a few real shifts. In the 80s and 90s, it was circulars in the mail, a store deciding what to spotlight, and a human choosing from that. Amazon created the endless aisle. Every keyword search became its own unlimited shelf, but a human was still reading titles, skimming reviews, and making the choice alone. What is changing now is that humans and agents are buying together. You express an intent; an agent can read every product on the page in full, every claim, every review, every bullet, run an intent analysis against what you actually need, and hand you four or five recommendations. You still pick one, but the agent has already done the comparison work a human used to do by skimming. On the brand side, agents are optimizing PDP content based on what consumers and their agents are looking for, deciding which products to bid on, and helping manage merchandising. The retailer sits in the middle of that loop and controls how much signal flows between the two sides and what gets shown to whom and under what promotion. That data and decision-making layer is becoming increasingly proprietary to the retailer.

Product discovery is moving beyond traditional keyword searches toward AI assistants that compare products, summarize options, and make recommendations. How should brands rethink product content and digital shelf strategy when they are increasingly trying to earn a recommendation from an AI agent rather than simply rank highly in a search result?

A human skims a page, reads a review summary, and looks at a couple of images. An agent can read every pixel. Even inside a shopping agent, your question gets broken into keywords, those keywords get searched, the results get extracted, every product gets read in full, and then an intent algorithm ranks the shortlist it shows you. That means content has to work for traditional search and for that pixel-level read at the same time, not one or the other. It also matters which kind of agent you are optimizing for. Shopping agents like Rufus, Sparky, and Alexa Shopping are where actual purchases close, and what moves the needle there is the product page itself and every attribute and claim on it. General-purpose agents like ChatGPT or Perplexity behave more like top-of-funnel research, and what earns you a mention there is closer to earned media, citations, blog posts, and third-party write-ups. Brands that treat those as the same problem end up optimizing for the wrong surface.

The term “agentic AI” is now being applied to a wide range of products. In retail specifically, what separates a true AI agent from traditional rules-based automation or an analytics platform with a generative AI interface layered on top?

Rules-based automation is if X happens, do Y. It has no intelligence, and it cannot handle anything you didn’t explicitly program for. A generative interface on top of a dashboard just makes the same static insight easier to read. A true agent can think, plan, execute, and then continuously assess and adapt, like a junior analyst you have trained. You give it context, your processes, your guidelines, and the decisions you have made before, and it starts executing against a goal rather than waiting for the next rule to fire. That is the test I actually use: can it operate the way a person you trained would, or is it just a faster version of something that was always static?

Commerce teams have traditionally had people monitoring inventory, pricing, product pages, retail media, search performance, and competitive activity across thousands of products. Which of these operational gaps are best suited to autonomous AI agents today, and where does human judgment remain difficult to replace?

I think about this as a 2×2 of how much judgment a decision requires and how much impact it has. Low-judgment work gets handed to the agent outright. High-judgment or high-impact work keeps a human providing strategic guidance and auditing the outcome. In practice, that means starting with your long-tail SKUs and long-tail retailers, the ones nobody has touched in a year anyway, and putting those on autopilot first since the blast radius of a mistake is low. Your top SKUs and top retailers keep a human layer, and that layer is not just for audit; it is how you train the agent. When you catch something, you tell it what it should have done and why, and that feedback is what makes it better over time rather than just consistently mediocre.

Specialized agents can potentially create their own problems if they optimize for isolated objectives. A media agent might increase advertising for a product that is running out of stock, while a pricing agent could improve conversion at the expense of margin. How do you ensure multiple agents work toward the broader commercial goals of the brand rather than simply optimizing their individual metrics?

This is exactly why we built an orchestrator rather than a set of independent agents. Ours is called Ally, and its goal is sales growth, not any single channel’s metric. It figures out where the opportunities and risks are, works through the trade-offs, and then assigns specific tasks to the worker agents, media, content, and pricing, sharing the same information across all of them so nobody is optimizing in isolation. Without that layer, you get exactly the failure mode in the question: a media agent spending against a product that is about to go out of stock because it has no visibility into inventory.

Retail data is often fragmented across retailer portals, advertising platforms, inventory systems, product information systems, and internal analytics tools. How important is data quality and data unification to successful agentic retail, and does giving AI agents greater autonomy make poor or incomplete data more dangerous?

Getting data into the right shape, with real semantic structure and a proper ontology, not just raw feeds sitting next to each other, is most of the actual work here. A lot of teams think you can upload something into ChatGPT or wire up an agent on top of whatever systems already exist, and what you get is inconsistent; sometimes it works, and sometimes it hallucinates, with no way to depend on it. Autonomy makes that worse, not better, because a human with bad data usually catches the error before acting on it. An agent with bad data acts on it immediately, at scale, before anyone notices.

As brands become more comfortable with AI agents, how should they decide which actions can be executed autonomously and which should continue to require human approval? What does the progression from “AI recommends” to “AI acts” look like in practice?

It should move gradually, not in one jump. Most teams start around 10 percent automated and 90 percent requiring approval. As the feedback loop runs, as humans correct the agent and the agent incorporates that correction, trust builds and that ratio shifts to 20 and 80, then further, on both the volume of actions and the value of what is being automated. The starting point for that progression is the low blast radius work, long-tail SKUs and retailers you were not actively managing anyway, and it only extends into higher-impact, higher-judgment decisions once the track record is actually there.

Retail media has traditionally been heavily focused on metrics such as return on ad spend, but there is growing interest in determining whether advertising is actually creating incremental sales. How can AI help brands make better decisions about incrementality, profitability, and where advertising dollars should be allocated?

Brands can get too enamored with measurement. Running a deep incrementality study for a quarter sounds rigorous, but the market has usually moved on by the time the study is done. I would rather see teams build a set of practical heuristics and act on them faster: if you are already ranking high organically, do not bid more; you are paying for a sale you would have gotten anyway. If you are ranking low organically, that is when the ad spend actually earns its keep. If there is an in-market audience that does not know your product exists, spend there to educate them. Arguing over whether a dollar returned 1.48 or 1.68 dollars incrementally is time spent not acting. Getting to the action faster and accepting that the number will be directionally right rather than perfectly precise is what actually moves sales.

If shopper-side AI agents increasingly handle discovery, comparison, and purchasing while brand-side agents manage content, pricing, advertising, and inventory, we could eventually have machines influencing both sides of a transaction. What do you think ecommerce looks like in that environment, and what will ultimately separate the brands that win from those that become invisible to these AI-driven purchasing systems?

I do not think we are heading toward agents transacting with each other autonomously any time soon. What I see instead is humans and agents buying together on both sides, agents doing the comparison and recommendation work, humans still making the final call, and the same pattern on the brand side, agents executing while humans set strategy and approve the decisions that matter. In that world, the physical product still has to be good; no amount of optimization saves a bad product. Past that, the separator is context: how much context an agent has about your business, how much it has learned, and how much feedback it has been given. That is what separates an agent that makes dumb mistakes from one that actually understands your brand voice and strategy well enough to be trusted with real decisions. The brands that win are the ones that treat giving agents good context and a working feedback loop as core work, not a side project, and the ones that skip that step are the ones that become invisible.

Thank you for the great interview, readers who wish to learn more should visit CommerceIQ.