Dennis Vorobyov is CEO of EltexSoft, a 40-engineer studio building software for Fortune 500s, and author of 42: The AI Builder's Stack.
getty
Every e-commerce leader I talk to is investing in AI, but where are those investments actually creating returns? The data of the last two years gives a clear answer: Companies that deploy generative AI well get an average of $3.70 back per $1 invested, and the top performers are reaching as much as $10 per dollar.
I look at this from the builder's side, since my team engineers AI-first products. The technology has moved from promise to production, and access to AI no longer separates the leaders from the rest. The difference is in knowing where to point it.
The Biggest Returns Hide In The Warehouse
Everyone wants to talk about chatbots and recommendations, while the highest returns actually happen in supply chains and logistics. A supply chain is the circulatory system of a business; customers see only the face (storefront, checkout, the delivery guy), but the money flows or clots in the veins nobody looks at. This is where AI delivers first.
One case from McKinsey's 2025 research describes an ordinary last-mile delivery operator with over 10,000 vehicles that put $2 million into AI virtual dispatcher agents and pulled $30 million to $35 million in savings out of it. Amazon's AI reduced out-of-stock rates by 19%, and DHL predicts shipment volumes with 90-95% certainty per facility, per day.
The next wave is agentic AI, autonomous agents making operational decisions without a human in the loop. Gartner predicts 15% of day-to-day supply chain decisions will run this way by 2028. The technology is ready before most of the market pays attention, and this is exactly the moment when moving early matters.
Product Discovery Is Being Reinvented
For twenty years, shopping online meant the same ritual: type keywords, scan the grid, filter, repeat. That ritual is ending.
Google Shopping queries in AI mode are 23 times longer than keyword searches, because consumers describe a situation, a need or a constraint the way they would talk to a knowledgeable friend.
The stores that can hold that conversation are being rewarded. Adobe's data shows AI-referred shoppers convert at higher rates and generate more revenue per visit. Amazon's Rufus assistant reached 250 million customers, and shoppers who engage with it are 60% more likely to buy.
This game is not reserved for the giants. Karaca, a Turkish kitchen and homeware company, built a GenAI shopping assistant and doubled conversion versus search, with 5x conversion versus unaided sessions.
The frontier moves even further. Wayfair's Muse generates shoppable room visualizations from a text prompt, so the customer describes a mood and receives an interior filled with products which can be bought on the spot. Discovery here stops being search at all and becomes co-creation between the shopper and the AI.
Practically, it means the following: AI systems now decide which products get recommended, and they can only recommend what they can read. Rich, structured, current product data is the new storefront.
The Winning Formula Is Human Plus AI
Klarna gave the industry the best free lesson. Their AI assistant handled the workload of 700 full-time agents in its first month and dropped resolution times from 11 minutes to under two. Then quality suffered once cost became the only metric, and their CEO said so publicly. They refined the model instead of retreating. Now, AI takes the routine, humans take the complex and emotionally loaded cases. Costs stayed down 40%, quality recovered and the whole market got to watch and learn.
The hybrid pattern shows up wherever AI performs best. Stitch Fix runs AI on 75% of product selections with human stylists making the judgment calls, and that combination delivered their first revenue growth in 12 quarters.
AI works as an amplifier of human judgment. This is the formula.
How To Get Into The 5% That Captures The Value
BCG found only 5% of companies are "future-built" and capture significant AI value today. The encouraging part is that what they do differently is neither a secret nor a matter of better models.
First, they fix the data before the AI. Clean catalogs, current inventory, consistent records. AI on top of great data multiplies every initiative at once.
Second, they redesign workflows around what AI makes possible instead of sprinkling AI over the old processes. Think of it as putting an engine on a horse carriage versus building a car.
Third, they buy before they build. MIT's research shows purchased AI solutions succeed roughly twice as often as comparable in-house builds, so spend the engineering effort where you are genuinely differentiated. Budget wise, none of this requires a moonshot, just clarity and sequencing.
The Compounding Effect
The thread running through all of it: AI advantage compounds. BCG's numbers show leaders see 5x the revenue gains and 3x the cost reductions of laggards, then reinvest those gains into more capability, and every cycle widens the gap a little further.
Bain projects agentic commerce (AI agents completing purchases autonomously) could reach $300 billion to $500 billion in the U.S. by 2030, up to a quarter of all e-commerce, through a channel that barely existed two years ago.
The companies winning at AI in 2026 started the flywheel earliest and kept it spinning. The window is open, and I would walk through it now.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?