Ask an e-commerce team if they’re ready for AI and there’s a good chance the answer will be an immediate “absolutely.” And they probably mean it. Competitors are already using AI, plenty of everyday work could use automation, and there are usually a few ideas the team would happily start tomorrow.
The answer gets less automatic when the question becomes: is our e-commerce stack ready for AI?
McKinsey found a similar gap: 70% of respondents admitted they felt personally ready to use AI, while only 27% of leaders believed their organizations were ready for the changes it would require.
An AI readiness assessment helps answer that question. Once you know what AI implementation challenges you may face in e-commerce and where AI can create measurable value in retail, the focus moves to what has to be in place technically before development starts.
The good news is that AI readiness is something you can check in advance. That is exactly what we’ll help you do here.
What an AI readiness assessment looks into
An AI readiness assessment reveals whether your e-commerce systems can support the AI use case you want to build, from the data it needs to the way it will run in production.
For e-commerce teams, five areas usually deserve the closest look:
- Data: product, customer, and order data quality, consistency, and availability.
- Architecture: access to the systems and functions the use case depends on, plus the ability to change individual components without affecting the whole platform.
- Integrations: reliable connections between ERP, PIM, OMS, CRM, and other systems involved in the use case.
- Agent-specific requirements: structured data, transactional APIs, permissions, and controls for AI agents that need to take action.
- Operations: ownership, monitoring, and fallback procedures once the system is live.
We’ll start with the part every AI use case depends on: the data available to it.
Is your data ready?
A storefront may look perfectly fine to a shopper even when the records behind it are inconsistent. AI makes those inconsistencies much harder to ignore.
Say one SKU is described as “oak” in the PIM, “natural wood” in a marketing spreadsheet, and “light brown” on the storefront. A person can connect the dots. Search, filtering, or agent logic now has three different values to reconcile.
That is where data readiness for AI starts. The information feeding the chosen feature needs to be structured well enough for machine use. For product data, check that:
- Active SKUs have the attributes the feature depends on.
- Attributes follow a consistent format across the catalog.
- Taxonomy stays consistent across systems.
- Price, stock, and other time-sensitive fields.
Customer information brings another test. An order in the OMS should connect to the customer record in the CRM and their history in the service platform. Otherwise, the picture arrives in fragments.
McKinsey found that only 7% of companies have fully scaled AI, with data issues among the factors holding companies back. Here’s how you can get a quick read on your own setup:
- Pull a random sample of active SKUs and look for missing attributes, conflicting values, or inconsistent categories.
- Pick a customer with a completed order and try to answer: “What have they bought, and have they contacted customer service?” See how many systems you need to open before you have the full answer.
Once the data itself holds up, look at the architecture around it.
Is your architecture ready?
Imagine replacing search or adding a recommendation engine. If the same release also has to touch checkout, catalog services, and several integrations, the change is already bigger than the feature itself. Timing can create another issue: a stock update that reaches the storefront after an overnight job is too late for an AI shopping experience built around current availability.
A composable commerce or API-first architecture gives teams more freedom to keep that work contained. In practice, look for three things:
- Search, recommendations, checkout, and other parts of the platform can be replaced without rebuilding unrelated areas.
- The functions your AI use case needs are available through documented APIs.
- Price, stock, order, and other time-sensitive updates arrive within the required time window.
Then test what this looks like in your own stack:
- Write down every system and release involved if you replaced search, recommendations, or checkout today.
- Update a price or stock value at its source and record when it appears everywhere that uses it.
Together, those two checks show which parts of the architecture can stay as they are and which need work as part of the project.
If your use case includes AI agents
AI agents in retail may be expected to act inside the e-commerce stack, for example, by adding an item to cart, starting checkout, or updating an order. For that kind of use case, add three checks:
- Machine-readable catalog data the agent can interpret consistently.
- Transactional APIs for the actions the agent is expected to perform.
- Permissions and controls that set access limits, approval rules, and activity logging.
The transactional part deserves particular attention. In its 2026 review of 345+ e-commerce sites, Aidō Lighthouse found that 29% scored zero on transactability, meaning there was no programmatic path for an agent to complete a transaction.
For AI shopping agents expected to handle checkout, order changes, or similar actions, verify the transaction path early. Map the actions the stack already exposes and note where new API access or controls are required.
Are your integrations ready?
Most e-commerce teams already have plenty of integrations running. Orders reach the OMS, product updates come from the PIM, customer data lands in the CRM. Day to day, there may be little reason to question any of it.
Then an AI use case needs information from several of those systems at once.
That is often when old assumptions surface: an endpoint nobody documented, a point-to-point connection one engineer still remembers setting up, or an API limit that has never mattered at the current traffic level.
If those integrations are going to serve another consumer, three details should be easy to find:
- Which systems exchange data, through which APIs or middleware.
- What the actual rate limits are for the ERP, PIM, OMS, CRM, and e-commerce platform.
- Who maintains each integration and where its current documentation lives.
The e-commerce platform may sit in the middle as well. On Salesforce Commerce Cloud, BigCommerce, or Adobe Commerce, APIs and webhooks may connect an AI feature to the systems it needs to reach.
The integration side of AI is already causing problems at scale. Deloitte found that 67% of surveyed leaders considered integration cost and complexity as a barrier to scaling AI agents.
A quick check can tell you how exposed your own setup is:
- Find the documented rate limits for your ERP, PIM, and OMS rather than relying on an estimate.
- Pick two systems and ask someone outside the integration team to explain how information gets from one to the other.
If basic integration questions still depend on one specialist, AI will make that dependency harder to manage. Before adding another consumer, make ownership explicit: who is responsible for each connection, and what happens when it fails.
Is your organization ready?
Picture a simple support case: an AI assistant starts giving customers the wrong return information. The first complaint arrives, and the ticket gets passed from support to product to engineering like a hot potato, in the hope that someone will pick it up and fix it. Someone probably will. The harder questions are how quickly they notice the scale of the problem and who takes charge from there.
Those decisions should already be made before the first incident. The team needs to know who owns the AI-assisted process, how unusual behavior is detected, and how a customer-facing or business-critical flow can be paused or handled manually while the issue is fixed.
Ownership is still surprisingly uncommon. Adobe found that only 16% of consumer goods respondents had a formal AI owner with a defined title and budget authority.
Run the same support case through your own team:
- If someone reports “AI did something wrong” today, who exactly gets the ticket?
- If a customer-facing or business-critical AI-assisted process misbehaves, can you switch it off or route the work elsewhere without taking the rest of the service down?
If the first answer starts with “it depends” or the second requires an emergency release, ownership and fallback handling still need work.
AI readiness checklist for e-commerce teams
We’ve covered a lot of moving parts by now. This checklist brings the basic practical checks from each section together, so you can run through them against the AI use case you’re considering.
Data:
- Can you pull complete records for the active products the use case depends on, including the required attributes and images?
- Do the same attributes mean the same thing across categories and systems?
- Can you connect an order to the customer’s purchase and support history without matching records by hand?
Architecture:
- Could you replace search, recommendations, or checkout without rebuilding unrelated parts of the platform?
- Do price, stock, and other time-sensitive updates reach every dependent system within the time window the use case requires?
- Are the backend functions the use case needs available through documented APIs?
Integrations:
- Do you know the actual API rate limits for your ERP, PIM, OMS, and other systems the use case will access?
- If an integration failed today, would monitoring catch it before a customer reported the problem?
- Can your integration layer support another consumer without exceeding rate limits or relying on an undocumented connection?
AI agents:
- Can an agent read the product information it needs directly from a machine-readable catalog?
- Are transactional APIs available for every action you expect the agent to perform?
- Are permissions, approval rules, and logging in place for those actions?
Operations:
- Can you name the person to put in charge of an AI-related issue when one is reported?
- Can monitoring flag unusual AI-driven activity before it turns into a customer-facing problem?
- Can a customer-facing or business-critical AI-assisted process be paused or handled another way while an issue is fixed?
Each unchecked box points to something that still needs to be resolved before development starts.
So, are you ready for AI?
If the answer is still “absolutely,” you can now say why: your data is usable, your architecture can support the use case, your integrations are understood, and the team knows who owns it once it goes live.
If a few items on the checklist are still open, you know where to start. For gaps that need a deeper technical assessment, Vilmate can help work out what needs to change and what belongs in the development scope.
