The problem Just Eat says it is actually solving
Just Eat Takeaway.com announced on August 11 that its AI Voice Assistant is expanding across Austria, Belgium, Bulgaria, Canada, Germany, Ireland, Israel, Italy, Luxembourg, Poland, Slovakia, Spain, Switzerland, the Netherlands and the UK, following earlier launches in the UK and Germany. The framing from Chief Corporate Development Officer Jorg Gerbig is the most useful part of the announcement, and worth quoting directly: Gerbig identified deciding, rather than finding a restaurant, as 'the biggest challenge in food delivery today.' The company built a voice interface specifically to compress the decision paralysis that comes after search already works, treating comparison fatigue rather than discovery as the real conversion bottleneck.
That distinction is easy to miss but strategically important. Most retail and marketplace AI investment over the past two years has gone into search relevance and recommendation ranking, on the assumption that better-ranked results solve conversion problems. Just Eat's own behavioral data undercuts that assumption for a high-choice category like food delivery: the real friction sits in evaluating too many good options, a stage that better search alone cannot resolve. That reframing has direct implications for where any high-SKU retailer should be pointing its next AI budget cycle.
The numbers behind the decision-fatigue thesis
Just Eat's usage data shows customers typically browse approximately three restaurants before ordering, with 43% proceeding directly to a favorite establishment, showing strong existing loyalty that any new interface has to work around rather than override. Nearly 25% of ordering journeys involve four or more restaurant reviews, the segment most exposed to decision fatigue and most likely to abandon a session without ordering at all. Decision time itself varies meaningfully by market: Swiss diners average seven minutes to decide, German diners nine, with burgers, pizza, chicken, kebabs and Indian food as the most common cuisine requests across the footprint. That market-level variance is itself useful operational data, since it tells Just Eat where a faster decision aid will move the needle most.
Against that baseline, the UK results are the ones worth scrutinizing closely: voice assistant users found options around 40% faster than text-only AI users, and were 30% more likely to choose a recommended restaurant or dish over a self-selected one. Both figures speak to the same mechanism, a conversational interface reduces the cognitive load of evaluating options in a way that a ranked list, however well-tuned, structurally cannot. Applied to a business Just Eat's size, shaving even a couple of minutes off the median nine-minute decision window in a market like Germany translates into meaningfully lower abandonment across millions of daily sessions.
Why voice specifically, not just better AI recommendations
It would be easy to read this as evidence that AI recommendations work, full stop, and skip past the voice-specific part of the result. But the comparison Just Eat draws is voice against text-only AI, not AI against no AI. Both groups had access to AI-driven recommendations, yet the voice cohort moved 40% faster and converted 30% more often on recommended options. That gap is about interface modality, not model quality, and it argues that the conversational format itself is doing real work in reducing the perceived effort of choosing.
For any retail or marketplace CTO evaluating conversational commerce, this is the specific claim to test rather than accept on faith: does a voice or chat-style interaction outperform an equally AI-powered but non-conversational recommendation surface in your own category. The answer will vary by product complexity and basket size, but food delivery's high-frequency, high-choice, low-consideration purchase pattern makes it close to an ideal test bed, and the result here is a genuine data point, not just a vendor's marketing claim, since it comes from Just Eat's own operational rollout rather than a controlled lab study.
The loyalty segment this interface has to protect, not convert
The 43% of customers who go straight to a favorite restaurant deserve separate attention, because a poorly designed voice assistant risks inserting friction into a transaction that previously required zero decision-making at all. Just Eat's rollout strategy has to account for this segment differently than the 25% deep in multi-restaurant comparison, and the company's phased UK-then-Germany-then-fifteen-market approach suggests it is using early markets to tune exactly that distinction before scaling.
The broader lesson for retail technology leaders building any AI-driven interface: aggregate conversion lift numbers can mask a segment where the new interface actually underperforms the old one. Before scaling a conversational or agentic interface across your full customer base, segment your rollout data by existing purchase pattern, habitual repeat buyers versus active comparison shoppers, because the interface that helps one segment can easily degrade the experience for the other if deployed uniformly.
What this means for your own conversational commerce roadmap
Just Eat's fifteen-market expansion is effectively a live, staged natural experiment other marketplace and retail operators can watch and learn from without spending their own pilot budget first. The company has already done the hard part, isolating decision friction as the specific problem worth solving with voice, and generating real before-and-after conversion data across two markets before committing to a wider rollout. That sequencing, narrow validated pilot before broad rollout, is the discipline worth copying regardless of whether voice specifically is the right interface for your category.
The concrete question for any high-SKU-count retailer or marketplace reading this is whether your own conversion funnel has a decision-fatigue problem or a discovery problem, since the two require different fixes and most organizations default to solving discovery because it is easier to instrument. Pull your own funnel data on browse depth before purchase, comparable to Just Eat's 'three restaurants' baseline, before deciding where conversational AI investment actually belongs in your roadmap, and before deciding whether a voice interface is the right format to build at all.
