Recently, the CEO of Consumer and Community Banking at JPMorgan Chase, Marianne Lake, stated that agentic AI was “not going to be the thing that scales rapidly.” Her reasoning boiled down to consumer trust, highlighting the disparity between the growing number of users relying on AI platforms for search and discovery and the few that are willing to take the next step towards purchasing.
While the disparity is currently very real, this prediction ignores what we’ve seen to be true regarding consumer trust: it’s something the market manufactures. Just look at Airbnb or Uber – once upon a time it sounded impossible to think people would choose to sleep in strangers’ houses over hotels or hop into their cars instead of calling taxis. But once the value proposition got strong enough, skepticism was won over. This didn’t happen by accident.
Though agentic commerce may not be rapidly scaling at this moment, it would be a mistake to wait around until consumers are “ready.” Instead, retailers should be actively building consumer trust in agentic commerce by laying the groundwork for the inevitable scale to come.
Don’t Call it a Tech Rollout: The Reality of AI Transformation
A couple of months ago I attended the Commerce Next Growth Show, where retailers generally fell into one of three categories when it came to agentic commerce. The first was the small number of brands going all-in on AI investment. Sitch Fix, in particular, comes to mind – in one presentation they talked about running 40 consecutive AI pilots as they work toward building truly personalized experiences.
The other two categories are those that are sitting back and taking a “wait and see” approach and those that are panic-investing in bolt-on solutions as a means to “keep up.” Though the latter group isn’t wrong to invest in AI now, their approach in treating it as a tech purchase is leading to two key issues.
The first issue is that companies are using AI to automate their pre-existing workflows without ever stopping to ask if the workflow should even exist. As a result, they’re creating faster versions of the same inefficiencies, with added cost, complexity, and risk. This generates further skepticism about the value of AI and whether it’s worth the effort.
The second issue is they miss out on crucial customer intelligence. It’s widely understood that the value of AI is proportional to the data it’s trained on. So when AI is treated like any other point solution, all that valuable data and learning compounds in someone else’s product.
If the promise of agentic commerce is autonomous decision-making on behalf of customers, the best thing a retailer can do is build the systems and structures that will organize, capture, and ultimately learn from customer data. This is also why the “wait and see” approach is equally risky. Every month delayed is a month of missing out on valuable insights.
In an Agentic World, Trust is Functional
The thing about building consumer trust in agentic commerce is that it’s largely functional. Going back to one of the examples referenced above, if cars rarely arrived via Uber’s app, the company never would have gotten off the ground. In retail, if an agent takes too long to respond or promises a Thursday delivery, but inventory data is stale, resulting in a delayed package, trust is immediately broken.
Retailers who are genuinely concerned about building consumer trust in agentic AI should therefore be working now to aggressively clean up their product data so that it’s structured, real-time, and trustworthy enough for an LLM to act on – and not just within the confines of a demo or pilot.
Those who are investing heavily and looking to move from pilot to production encounter data challenges such as: finding the sweet spot in structuring data ontologies so they remain dynamic; dealing with context graphs breaking when applied to real enterprise data; agents struggling to reason across structured and unstructured data sources; and overspending on tokens due to poor memory usage or model inference.
Luckily, solving for these issues doesn’t require ripping out every legacy system and starting from scratch. Instead, new innovative tools like Oliver DB or Saphenia are entering the market to help retailers build an “agentic context layer” on top of what they already have. This organizes data into four key areas an agent can reliably act on: product details, customer history, session signals, and business rules like inventory and pricing.
By prepping for agents now, retailers will get ahead of competitors who will inevitably suffer the growing pains of messy data rupturing trust when agentic commerce becomes status quo.
The Governance Gap That Will Define Agentic Winners and Losers
Of course, trust isn’t limited to functionality, and the security concerns highlighted by JPMorgan are legitimate. Earlier this year, my company, Solvd, surveyed CIOs and CTOs across different industries and found that though retail AI projects were more likely than other industries to report positive ROI, they had less formal governance in place and were less proactive in their approach to governance overall. Trust may be built through functionality, but it’s maintained via strong governance.
And though there’s much talk about keeping a “human in the loop” with agentic commerce, it’s quite literally contrary to what agentic commerce is meant to achieve, but, more importantly, it’s not scalable. The math immediately falls apart if you have to hire more humans to approve every agent’s decisions.
A more sustainable solution is for retailers to define the objectives, constraints, and escalation rules that agents operate within – in other words, establishing Role-Based Access Control (RBAC). I recommend looking at the tasks agents would accomplish through a three-tiered model:
- Act-and-log: actions agents can take and keep a record of (no human required).
- Act-and-notify: actions agents can take whilst notifying a human in near-real time.
- Recommend-and-wait: actions agents can recommend taking, but require human approval.
By calibrating autonomy according to consequence, agents are empowered to independently make lower-tier decisions (like product recommendations) but are kept on short leashes for higher-risk tasks – like approving credit or making employment decisions. In creating these systems of governance, retailers can successfully scale agentic commerce without risking brand-destroying consequences from agents gone rogue.
None of this is going to be theoretical for much longer. Earlier this year Macy’s announced the launch of “Ask Macy’s” – a conversational AI shopping chatbot that assists with personal recommendations, product discovery, and virtual try-ons. When major players start getting involved, it’s usually not long before that momentum spills over into consumer expectation.
The retailers who set their foundations now with new processes and structures, clean data, and clear governance will be the ones who build consumer trust in the agentic future.
