Agentic commerce, six months in: the delivery promise is now part of the pitch
By Johan Hellman, Chief Product Officer, nShift
Back in May, I wrote here that AI shopping agents were already ranking stores, whether retailers knew it or not. Two months and one company report later, I can put numbers on it, some of which might take you by surprise.
AI and agents drove 20% of retail sales over the 2025 holiday season, roughly $262 billion, according to Salesforce. Adobe measured traffic from AI sources to US retail sites up 393% year on year in the first quarter of 2026. The figure I keep quoting to retailers, though, is the conversion flip. In March 2025, a visitor referred by an AI tool converted 38% worse than other traffic; twelve months later, it converted 42% better. Shoppers who arrive via an assistant have already compared, shortlisted and decided; they no longer browse like tourists, and the visit is the formality.
The takeover that did not happen
The most notable development of H1 was a retreat. In March, OpenAI closed its Instant Checkout programme after fewer than 30 merchants ever went live. The future many of us were told to prepare for, agents completing purchases wholesale on our behalf, has not arrived. Customers still buy on the merchant’s own checkout, and the merchant still owns the delivery promise.
The future of delivery: 2026 trends midyear check-in, the report nShift published this week, calls this the agentic inversion. AI took over discovery and comparison, the top of the funnel, and left the transaction alone. The catch: when an assistant compares five stores on behalf of a shopper, your delivery dates, fees and returns terms are being weighed before anyone reaches your site. Our research makes it clear: if your delivery options cannot be read by a machine, they are starting to be invisible to one.
The pitch, in other words, now happens without you in the room, where your delivery data gets invited in your place.
What an agent actually reads
When an AI assistant evaluates a retailer, it skips the hero banner and parses whatever structured information it can find: which delivery options exist, what they cost, when the parcel would arrive, how a return works and when the money comes back.
Retailers publish machine-readable delivery terms with wildly varying quality, and most have never audited what a machine sees. In my experience, any audit would result in humbling findings: delivery dates that only render in JavaScript, fees that appear three clicks deep, returns terms buried in a PDF.
But none of this requires exotic technology to fix, because the same data already powers your checkout. It just needs to be structured, accurate and consistent wherever a machine might look. And the commercial case was proven on humans first: Flying Tiger Copenhagen rebuilt its delivery options at checkout around better pickup points and recorded a 20% increase in conversions. The retailers winning agent-referred traffic in the second half will be the ones whose delivery data works as hard as their product data.
The promise got harder to keep
Being legible is half the job – the other half is keeping the promise, and the first six months gave retailers plenty of new ways to break one:
- Since 1 July, every parcel under €150 entering the EU carries a flat €3 customs duty, with a €2 handling fee planned for November.
- A FedEx-led consortium agreed to buy InPost for €7.8 billion, and DHL and Evri completed their UK merger, so the locker network your customers rely on may have a new owner by peak.
- From August, the EU AI Act requires that people be told when they are talking to a machine, which covers every delivery chatbot and notification flow.
- And since 19 June, EU consumer rules require online sellers to make cancelling a distance purchase as easy as making it.
Customer behaviour moved just as fast: out-of-home volumes and locker estates at Europe’s biggest parcel networks each grew by roughly 30% in a year, and returns led the shift, with shoppers now dropping most of their returns at lockers and parcel shops rather than waiting for a courier. A delivery menu tuned for 2024 habits misrepresents what your customers, and their assistants, actually want in 2026.
Our report groups these pressures under one term: the orchestration gap. A delivery promise at checkout is assembled in real time from carrier capacity, routing rules, pickup-point availability, warehouse cut-offs, customs data and returns flows. Those usually run in separate systems but the customer, and now the customer’s assistant, experiences them as one promise. The gap between the two is where sales go missing: the promise a machine read on Tuesday has to be the promise your network keeps on Thursday.
What to fix before peak
The report closes with several questions and recommendations for logistics leaders. If you only have capacity for three before October, mine would be these:
- Run the legibility test: give someone a day to check what a machine can extract from your storefront: options, dates, fees, returns terms, in structured form. You cannot fix what you have never seen through an agent’s eyes.
- Treat the delivery menu as merchandising: tune which options appear by market, postcode and basket, and suppress any promise your network cannot reliably keep. A slightly slower date you always hit beats a fast one you miss in December, with a human customer and doubly so with an agent that remembers.
- Rehearse the switch. The businesses that came through the spring tariff shock were the ones that could change carrier, customs data and customer promise together within days. A delivery management platform with the alternative carrier already integrated turns plan B from a quarter-long project into configuration; whatever your setup, measure your switching time now rather than discovering it in December.
On an encouraging note from the wider data we looked at, this work pays back on measures you already track. Gartner found in May that 83% of supply chain organizations apply AI incrementally to specific use cases rather than grand redesigns, and the programmes that graduate from pilots are tied to reported numbers such as delivery promise accuracy. Agent readiness belongs on the same list: it’s an audit and a data discipline you can fit in the weeks between now and peak.
Six months ago, agentic commerce was a prediction we defended at conferences. Mid 2026, it is a measured acquisition channel with a conversion premium. The retailers who benefit will be the ones who realised early that a delivery promise is no longer read only by people, and made sure the machine-readable version was worth recommending.
Johan Hellman is Chief Product Officer at nShift, whose delivery and experience management platform supports more than one billion shipments a year across 190 countries.
