Most retailers believe they have an AI tool problem. They switch vendors like pants, searching for the perfect fit, and still see no significant results.
The uncomfortable truth? All major AI models are functionally identical. The problem lies elsewhere.
“One in a thousand retailers create opportunities to go beyond basic implementation,” says Julian Krenge, Co-Founder and CPO at parcelLab. “Yes, basically only a few retailers that really doubled down on this and wanted to drive real customer experience improvement.”
That’s not a typo. One in a thousand. Which raises an obvious question: What are those 0.1% doing differently?
Julian Krenge works with hundreds of retailers from SMBs to enterprise giants every single day. He sees which AI implementations deliver results and which ones fall flat. More importantly, he’s identified a pattern: Retailers often don’t know where they stand on the AI maturity scale. And because they don’t know where they are, they choose the wrong next steps.
From these observations, Julian developed a stage model that allows retailers to assess their current position and plan their next move with clarity. Picking a tool before understanding your current state is like navigating without a map. First assess where you are, then define where you want to be.
Why the traditional approach doesn’t work
“AI for Customer Service” is too broad to be actionable. Without clear progression logic, retailers don’t know what’s realistic. They expect a 30% ticket reduction and run into the reality that a basic FAQ bot is about as effective as a digital parrot.
The jump from “informing” to “acting” gets massively underestimated. That’s where implementations stall. AI success isn’t a question of the tool. It’s a question of maturity, and that maturity builds in clearly defined stages.
The AI CX maturity ladder
Before retailers can progress through the maturity stages, they need to see the full customer experience (CX) landscape. Most think “AI in CX” equals “AI in Customer Support” and miss three other domains entirely.
The complete picture includes four areas:
Customer support – covering WISMO, returns, and warranties.
Pre-purchase support – handling sizing, product guidance, and compatibility.
Automated communication – managing order confirmations, dispatch, and cancellations.
Legal obligations – processing returns, recalls, and chargebacks.
This domain map becomes your use case backlog. It shows where AI can create value and helps you prioritize where to start. With the full landscape in view, retailers move through a predictable development.
Each stage has clear characteristics, measurable outcomes, and specific requirements for the next step:
Stage 0: AI makes an informed guess (FAQ-Level)
Stage 1: AI handles one thing well – Informative
Stage 2: Branching path – “Taking action” OR “One AI for everything”
Stage 3: Combined multi-use case AI with cross-pollination effects
Understanding where you stand on this ladder changes everything. It sets realistic expectations, clarifies the next actionable step, and prevents the most common mistake Julian sees: switching tools instead of enabling them.
The detailed blueprint: Stage by stage
Stage 0: AI makes an informed guess
The FAQ-bot is better than nothing, but far below potential.
At this stage, your AI uses FAQ data combined with standard LLM knowledge. No live data or customer specificity. When someone asks about their order, the typical response is: “Here’s the link to our order status page.”
The AI knows general eCommerce knowledge from its training, your FAQ content, and standard processes like “FedEx typically delivers within 3-5 days.” What it doesn’t know: individual order status, customer history, or product-specific details.
The expected impact remains modest. A small reduction in customer service inquiries and less traffic on your FAQ page. But there’s no significant monetary impact.
Stage 0 works when you’re selling simple products like smartphone cases, using Fulfillment by Amazon, or running highly standardized processes. It falls short with complex logistics, high product variance, or B2B-adjacent markets with specific requirements. On the opposite end of complexity sits a retailer like IKEA, who operates their own fleet to deliver furniture. For them, such a setup would essentially solve zero problems.
The most common mistake at this level: Tool-switching instead of tool-enabling. Retailers change providers instead of integrating more data into their existing system. As Julian puts it:
“Fundamentally, even if you look at all of the benchmarks and the underlying AI models, ChatGPT and Gemini and Claude are identical. They are as good as the other. The limiting factor is how do you put in more knowledge, more context so it can give the right answer.”
Stage 1: AI handles one thing well (informative)
The AI accesses real data and delivers individual answers.
This is where things get interesting. Your AI now has live data for at least one use case, retrieves customer-specific information, and provides personalized responses.
Take WISMO inquiries as an example. Instead of a generic link, the AI now says: “There’s this one order for you. It’s two packages. One is coming from our East Coast Fulfillment Center, one is coming out of the West Coast Fulfillment Center, and both are with FedEx. And for one, we already have a delivery date because it has been dispatched, and the other one will probably be dispatched tomorrow.”
That’s a fundamentally different experience.
“Realistically, you can have an expectation for something between 20 and 30% of the WISMO inquiries that are currently hitting your customer support team, you will be able to resolve with AI before it actually ever hits a human. Those are non-value adding contacts anyways that are deflected ahead of time.”
Retailers like YETI, with their Ada-powered order lookup, and True Classic, with their codefuse-powered chatbot, operate at this level.
Moving from Stage 0 to Stage 1 requires four steps.
Identify which use case has the highest impact.
Map what data the AI needs.
Integrate real-time data sources technically.
Test and evaluate response quality.
When prioritizing, multiply Impact by Confidence. Low confidence means a bigger discount on expected results.
Stage 2: The branching path
The strategic decision: Go deeper or go broader?
After Stage 1, retailers face a fork in the road. You can either go vertical, enabling your AI to take action, or go horizontal, expanding to multiple use cases.
Stage 2A: Taking action
The AI executes. Not only is it informing, but it’s doing.
Your AI learns to trigger processes autonomously. It becomes transactional, requiring significantly more technical integration. Consider the returns experience. At the informative level, the AI can look up your order and tell you: “I see you ordered three items. One is a hygiene item. You cannot return this. Everything else is returnable until December 14th.”
At the action level, rather than only giving you information about the return policy, the AI actually asks you what items you want to return and why, and you just chat with it. At the very end, you get a return label and an email saying you can drop off your items.
If you’re at this stage, you belong to the 0.1% of retailers most invested in AI. Pandora works toward this vision with their “Gemma” AI – a one-stop shop that can take actions on behalf of the customer. They’re not there yet, but that’s what they’re building toward.
They expect the impact to be a 50-70% reduction in manually processed cases, and approximately a 20-25% repurchase rate for loyal customers.
But here’s the caveat, Julian warns: “The step between this informative AI is so large that it’s kind of its own stage. Moving to having it take action is larger from an effort level than the stage from zero to one.”
Stage 2B: One AI for everything
One chatbot, many use cases with compound effects.
Instead of deeper action-taking, you expand broader. The same chatbot handles order status, sizing help, and shopping assistance simultaneously. This creates cross-pollination effects:
“Maybe before you were able to resolve 20% of your WISMO queries. But now people go for shopping assistance and sizing help pre-checkout, and already interact with your chatbot. So they’re much more likely to use it post-checkout. Maybe your resolution rate goes up from 20 to 25 or even 30% because you have more adoption.”
Choose this path when multiple use cases have similar effort levels, when action-integration seems too complex, or when pre-purchase conversion represents a significant pain point. Julian sees huge potential here:
“I think the potential of having the AI help with this product discovery over a search, for example, is huge. So you start maybe with reducing costs. But ultimately, it helps you generate upside potential. And the upside is much more interesting.”
Stage 3: Combined multi-use-case AI with cross-pollination
The vision: One central agent for everything.
At this pinnacle, you achieve multi-domain coverage with action-taking capability. The AI understands customers holistically across pre-purchase, post-purchase, returns, and service, using context across all domains.
But how do we get to this Iron Man 3-esque stage?
By combining multiple use cases with action-taking ability, your probability of adoption increases because the more your chatbot can do, the more likely people are to come back to it. You get all of the individual case improvements combined, plus higher adoption overall.
However, please don’t crash into your IT department yet, waving this blog post around: Even leading brands like Pandora are still building toward this. It requires massive technical and organizational investment, and may only be realistic for retailers with dedicated AI teams.
From invisible to unbeatable: What each stage delivers
The transformation becomes tangible through concrete benchmarks:
At Stage 0, your AI answers: “Here’s the link to our order status page.” Customers research everything themselves with no personalized experience.
At Stage 1 and beyond, your AI responds with real, individual data. Customers feel understood. 20-30% fewer tickets reach human agents, freeing them for complex cases requiring genuine expertise.
At Stage 2 and beyond, your AI acts proactively. Customers experience frictionless processes. Retention and revenue rise measurably.
Conclusion: Your next move on the maturity ladder
In order to achieve AI success in CX, you need to build maturity systematically: stage by stage. The required mindset is threefold. Manifest this:
Patience: The jump from Stage 1 to Stage 2 takes time.
Focus: Better to execute one use case properly than three halfway.
Re-evaluation: Assess your position every six months as technology evolves rapidly. As Julian notes: “Whatever you thought six months ago doesn’t work anymore.”
With all the caveats about complexity and effort, one thing remains clear. Julian emphasizes: “The cost of not doing it is way higher.”
Wonder where to start?
Assess: Where do you stand on the maturity ladder?
Prioritize: What’s the next realistic step?
Execute: One use case, done right
Evaluate: Measure, learn, iterate
Those who lay the groundwork now will be best positioned when AI capabilities continue to advance.
Your questions, answered
AI maturity describes how advanced your AI implementation is, from basic FAQ bots that guess answers to intelligent agents that take action autonomously. Most retailers are stuck at Stage 0, where AI only uses general knowledge without access to real customer data.
Because they lack context. All major AI models are functionally identical. The difference lies in data integration. Without access to live order data, customer history, and product details, AI can only give generic answers that don’t resolve individual inquiries.
At Stage 1 maturity, where AI accesses real-time order data, retailers can expect a 20-30% reduction in “Where Is My Order” inquiries. These are non-value-adding contacts that get deflected before reaching human agents.
Every six months. AI technology evolves rapidly, and what seemed impossible six months ago may now be achievable. Regular reassessment ensures you’re always taking the next realistic step on the maturity ladder.
AI
