7. Are You Agentic‑Ready?
Beyond AI features: Five questions European commerce leaders use to test their stack
In conversations with CDOs, SI founders, AI product teams and investors across Europe, the same pattern keeps coming back: everyone is “doing AI in e‑commerce”, but very few feel genuinely agentic‑ready.
Teams have pilots for GenAI copy, chatbots, new search tools and a re-platforming project on the side. Vendors tell them their platforms are “agentic‑ready”, but nobody offers a simple way to test that claim against the messy reality of European commerce stacks.
This article distils five questions that keep surfacing in those real roadmaps, steering committees and due‑diligence calls with merchants, system integrators, AI developers and investors who have to live with the consequences of their decisions.
If you can answer these questions with specifics, most likely you are further along than most organisations in your segment.
1. Strategy: what real problem would an agent solve for you?
Most “AI in commerce” roadmaps still look like lists of experiments: better product descriptions, a conversational search pilot, a few A/B tests, maybe a chatbot for customer service. These are fine as projects, but they do not add up to an operating model.
In work with mid‑market and upper‑mid merchants, the conversation only moves when we tie agents to very concrete P&L problems: margin erosion in specific categories, rising service costs, stock‑outs, slow assortment localisation, or high return rates in a few key segments.
An agentic‑ready strategy can answer questions like:
Which 2–3 business problems would change materially if an agent could run that loop end‑to‑end?
How would that show up in KPIs in your current planning horizon, not just on an abstract three‑year roadmap?
If you cannot name those problems yet, you are still in “AI feature” mode. To move beyond that, it helps to view AI as part of a broader commerce operating model built on three always‑on agents:Catalog & Product, Agentic Commerce and CRO & Experimentation.
2. Architecture: are you adding AI to a legacy core?
In many platform selection and re-platforming reviews, there is a quiet moment where someone admits, “We have three different sources of truth for stock and price.” That is survivable when humans reconcile the gaps in their heads. It is not survivable when agents need to query, compare and act against your stack.
Agentic‑ready architecture does not start with an AI tool; it starts with a clear, boring picture of how your core actually works:
Where product, price, stock and customer truth live today (PIM, ERP, CDP, data warehouse).
How many transformations happen before this information reaches the storefront, marketplaces and media systems.
Which APIs an internal or external agent would call to search, quote, reserve and place orders safely.
When we look at stacks for merchants considering platforms like Shopify, VTEX, SFCC, Adobe Commerce or commercetools, a simple litmus test is: can the team sketch this architecture on one page, and point to where an agent would plug in? If the answer requires three architects and a follow‑up meeting, the constraint is probably the foundation, not the AI.
3. Data: can an agent understand what you sell and what is allowed?
In early tests with marketplace‑heavy brands, most “AI errors” did not come from the model; they came from the catalogue. Missing attributes, inconsistent values, unclear compatibility rules and thin content made it impossible for agents, and often humans, to make good decisions.
Agentic commerce depends on your product data being legible to both people and software agents. That means:
A coherent product data model: attributes, relationships, bundles and variants that are consistent across categories.
Explicit constraints: what cannot be sold together, what cannot be shipped to specific regions, what requires certain certifications or approvals.
Identity and consent patterns: so that an agent always knows who it is acting for and which data and actions are permitted.
Many organisations recognise themselves in the catalogue problems: supplier feeds with inconsistent data, manual translations, cross‑sell logic stuck in spreadsheets, and teams constantly behind on enrichment.
A Catalog & Product Agent is one practical way to address this: turning catalogue quality from a one‑off clean‑up into continuous infrastructure.
4. Governance: what can an agent do without a human?
As soon as you move beyond simple recommendations or chatbots, governance questions appear. SI founders and product leaders often share the same story: a client wants an “autonomous pricing agent” or “self‑optimising journeys”, but nobody has written down who is allowed to change what, and within which boundaries.
Agentic‑ready organisations make these decisions explicit before they deploy agents:
Which actions can be fully autonomous (for example, replenishment within defined thresholds, low‑risk merchandising changes)?
Which actions always require human approval (large B2B quotes, changes in contractual terms, high‑impact promotions)?
What limits apply (price bands, discount ceilings, inventory buffers, compliance rules)?
They also insist on auditable logs: for any significant action, it should be possible to reconstruct what the agent saw, how it made its decision, and what it did. Without this, “let the agent handle it” easily turns into operational and regulatory risks.
5. Ecosystem: which partners can actually help you become agentic‑ready?
No mid‑market or enterprise merchant will build agentic commerce in isolation. In practice, your readiness will be determined as much by your systems integrators, AI builders, platform vendors and data partners as by your internal team.
In discussions with ecosystem partners, the same tension shows up repeatedly:
SI’s and vendors say, “Clients are asking about AI readiness and agentic commerce, but we do not have a clear narrative or framework.”
Merchants say, “Every vendor tells us they are agentic‑ready, but we cannot validate that claim during selection or implementation.”
Agentic‑ready ecosystems have partners who can:
Explain your current readiness in plain language to your board or investment committee.
Show how their product or service maps onto the three‑agent model: Catalog & Product, Agentic Commerce, CRO & Experimentation.
Commit to an at least 6 month roadmap that improves architecture, data and governance step by step, rather than selling isolated features.
A note for investors and M&A teams
Investors and corporate M&A teams evaluating commerce, MarTech and SI targets now see “AI‑ready” and “agentic‑ready” on almost every deck. In due‑diligence support, three questions consistently separate reality from positioning:
Does the target’s architecture show clear places where agents could plug in, with APIs and events that a technical team would trust – or does “AI” sit in a separate box?
Are there already processes or tools that behave like the three agents (catalogue, journeys, experimentation), with measurable outputs – or only pilots and prototypes?
Do logs, KPIs and governance frameworks exist that would survive increased autonomy and traffic, or is everything reliant on a few key people?
If the answers are vague, “agentic” is likely a story that will not survive deeper diligence.
How to get started
For CDOs, Heads of Ecommerce, founders and partners who recognise elements of their own landscape in this article, the first steps are usually straightforward.
Clarify your data foundations.
Map where product, price and customer truth actually live today.
Trace how many transformations happen before this information reaches customers and external systems.
Ask what an external agent would conclude if it queried your stack as it is now.
Define a minimum agent‑ready architecture.
List the 5–7 capabilities you consider non‑negotiable (APIs, events, catalogue quality, identity and consent, governance).
Align this with ongoing or planned work on platforms, data and integrations instead of treating it as a separate “AI project”.
Choose one agentic use case and commit.
Anchor it in a clear KPI: churn, AOV, service cost, stock‑outs, time‑to‑catalogue or similar.
Let this use case drive the architectural, data and process improvements you wanted to make anyway.
From there, the three‑agent operating model, Catalog & Product, Agentic Commerce, CRO & Experimentation, becomes a way to keep improvements compounding across your entire commerce stack, for both human buyers and AI shopping agents.