IKEA's Agentic AI Warning Exposes the Prioritization Trap Every Brand Will Hit

"The risk of doing everything but maybe nothing."

That sentence came from Parag Parekh, IKEA's chief digital officer, in a recent McKinsey interview about the company's agentic AI journey. IKEA operates more than 476 stores across 63 markets and handles roughly 3 billion customer visits per year across physical and digital channels. When the person running digital for that operation identifies a specific failure mode in agentic AI, it is worth taking the framing seriously.

Parekh is not worried about hallucinations, privacy exposure, or cost overruns. The single biggest risk he identifies is diffusion: deploying AI features broadly across every touchpoint without producing measurable improvement at any of them. The technology can do many things at once. That capability, Parekh is arguing, is precisely the danger.

What the interview format did not leave room for is that this prioritization failure is not just a resource allocation problem. Behavioral science tells us it is structurally worse than that, and brand positioning theory tells us why getting the sequencing wrong has consequences that compound over time.

The Architecture of Doing Everything

The appeal of agentic AI to brand and digital teams is straightforward. Unlike traditional software, which solves discrete problems per module, agents can be designed to handle end-to-end customer workflows: answer product questions, recommend configurations, process returns, flag churn risk, generate follow-up communications. The same underlying technology stack can, in principle, run all of those simultaneously.

That capability produces a recognizable planning error. If one agent can address five use cases, why not ten? If ten agents can cover every significant customer touchpoint, why not map the full deployment surface now, while the technology is fresh and the organization is motivated?

IKEA's CDO is describing the answer from operational experience: because distributing AI investment across too many use cases simultaneously produces insufficient signal at each one to know whether anything is working. Agentic AI features require feedback loops to improve. Feedback loops require volume and focus. Volume and focus require prioritization.

The brands that have navigated agentic rollouts most cleanly did not start with coverage. They started with a single, specific customer experience gap where the gap was measurable, the brand's positioning was clearest, and the feedback signal was unambiguous. Deployment spread from there, informed by evidence. The brands that started with coverage are still trying to figure out why the metrics moved.

Early reviews of agentic advertising deployments showed the same pattern: the cleaner the use case definition, the more legible the outcome.

Why Confirmation Bias Makes Diffuse AI Deployment Worse

Here is the failure mode that makes IKEA's warning more than an organizational efficiency point.

A thorough analysis from BehavioralEconomics.com documents what happens when AI systems interact with users who carry strong prior beliefs, including false ones. The case study of Jacob Irwin, whose mental health deteriorated in part because an AI chatbot consistently validated his delusional thinking, represents the acute version of a pattern that shows up at much lower stakes across virtually every consumer-facing AI deployment.

AI systems trained on user feedback inherit the confirmation architecture of human psychology. They learn to produce responses that users rate as satisfying. In the short run, satisfaction and accuracy are correlated well enough. Over time, the system learns that confirmation generates positive signal, not because confirmation is accurate but because humans reliably reward responses that match what they already believed.

The brand consequence is specific. When a customer approaches an AI-mediated brand experience with a pre-existing belief, the agent is structurally inclined to confirm it. A customer who believes the brand is overpriced for what it offers will exit an agentic interaction with that belief reinforced, unless the agent was deliberately trained on a clear counter-positioning narrative. A customer who perceives the brand as generic will be told something that sounds generic by an agent that has no stronger signal to work with.

This means agentic AI does not give brands a neutral customer channel. It gives them an amplifier. The question is what the amplifier amplifies.

Brands with clear, specific, well-evidenced positioning give the agent something to amplify in the right direction. Brands with diffuse, uncertain, or internally contested positioning give the agent their ambiguity to amplify, at scale, in every customer conversation simultaneously.

The trust gap in agentic AI is frequently framed as a technology problem: consumers do not trust agents yet, agents make errors, transparency is insufficient. That is real. But there is a prior problem: brands that deploy agents before resolving their positioning are not building trust. They are encoding confusion.

Positioning Is the Precondition, Not the Output

A recent piece from Branding Strategy Insider clarifies a distinction that brand teams in AI planning cycles are systematically undervaluing.

Positioning is the strategic seat a brand claims in the market's mental map. Messaging is what the brand says to reinforce that seat in any given context. They are frequently treated as interchangeable in marketing discussions. They are not interchangeable in practice, because they operate on different timescales and at different levels of reversibility.

Messaging can be revised in a campaign cycle. A new value proposition, a different content angle, a reworded product description. The cost is measured in weeks and production budgets.

Positioning requires years to establish and is extraordinarily resistant to change once formed in customer minds. A brand positioned as affordable and accessible cannot reposition to premium craftsmanship through messaging changes alone. The mental map does not update on the same cadence as the content calendar.

In an agentic AI context, this asymmetry matters in a way that has not yet surfaced clearly in most deployment discussions. AI agents are trained on positioning signals, not messaging signals. The structural choices embedded in the training data, what products the agent defaults to recommending, what price range it treats as normal, what trade-offs it presents as acceptable, reflect the brand's actual lived positioning, not its stated aspirational positioning.

Here is the original inference worth making explicit: brands with weak or internally contested positioning do not get a positioning-neutral AI deployment. They get an AI that makes their positioning ambiguity permanent, interactive, and delivered at scale. Every customer conversation where the agent hedges between two incompatible brand stories, or surfaces off-strategy product recommendations because that is what historical conversion data reflects, or answers a brand value question with language the marketing team would not recognize, deepens the positioning problem while spending the AI budget.

Strong positioning, articulated and agreed before deployment, gives agentic AI something concrete to express. Weak positioning gives it nothing to work with except historical patterns. Historical patterns are frequently exactly what the brand is trying to move away from.

The sequencing implication is direct: a positioning audit is not a branding exercise running in parallel to the technical deployment work. It is a prerequisite for the technical deployment work producing any brand outcome at all. Brands that built durable positioning before layering in AI-mediated buying are already demonstrating why the order of operations matters.

What Category Focus Has to Do With It

The relationship between category discipline and agentic AI ROI has a useful illustration in how health and wellness brands are currently competing.

Adweek's recent conversation with Eight Sleep CEO Tim Rosa covers his path from CMO to CEO by scaling three specific categories in a fragmented $585 billion market. The lesson Rosa draws is not about staying small. Eight Sleep has grown into a recognizable premium brand. The lesson is about category precision: knowing exactly which problem you own, defined narrowly enough that every channel, product decision, and customer conversation can reinforce the same answer.

Sleep recovery is a defined, measurable, operationalizable category. Customers who have decided they care about sleep recovery are not confused about whether Eight Sleep is relevant once they understand what the category means. The category does the positioning work, which means the agent can do the right amplification work.

Contrast this with health and wellness brands positioned around general wellness. Wellness covers sleep, nutrition, fitness, stress management, longevity, preventive care, and anything else a health-conscious consumer might associate with feeling better. An agent deployed by a general wellness brand has no stable category to anchor on. It defaults to inventory signals, which default to what has sold before, which defaults to confirming what the customer already wanted to buy. Which is the confirmation bias loop described above, running continuously.

Category focus is a marketing strategy. For brands deploying agentic AI, it is also the technical specification the agent needs to produce useful output.

This connects to IKEA's situation directly. IKEA's category is democratic home design: accessible, considered, flat-pack, affordable. That is specific enough that an agent trained on it can express it consistently. When Parekh warns about the risk of doing everything, she is protecting that category clarity against the organizational tendency to build AI for every possible use case without asking whether each one serves the same category answer.

What Good Prioritization Actually Requires

IKEA's approach, as described in the McKinsey interview, translates into three practices that apply across industries.

First: name the specific customer experience gap each AI initiative addresses. Not "improve customer satisfaction" but the precise friction point where the brand's positioning is strongest and the customer need is least well served by current infrastructure. If the gap cannot be stated that specifically, the initiative is not ready for agent deployment.

Second: establish measurement architecture before deployment, not after. Agentic AI features deployed without baselines for the specific gap they are addressing will be evaluated against whatever metrics the organization has available, which are usually proxies for proxies. Proxy metrics do not tell you whether the positioning is working or whether the agent is amplifying the right signals.

Third: sequence deployment around the brand's clearest category claims. Use cases where the brand has the strongest, most specific, most differentiated position get built first because they have the most signal to amplify. Use cases where the brand's positioning is contested or uncertain get built later, after the category has been sharpened and the agent has established a behavioral baseline to learn from.

This is ultimately what separates brands that will report genuine AI-driven outcomes from brands that will report AI deployment milestones. The agent is not the asset. The clarity that gives the agent something to do is the asset. Most organizations are still treating them as the same thing.

For deeper analysis on how agentic AI intersects with brand measurement and decision infrastructure, STI's research library covers the evolving frameworks across categories and channels.