When MIT dropped its 2025 report claiming that 95% of AI projects are failing, the headlines practically wrote themselves.
Cue the “AI was overhyped.” “The bubble is bursting.” “We told you so.”
Pause. The question isn’t whether AI has potential. It’s why so many implementations stall, disappoint, or outright fail.
The answer is this: most AI isn’t integrated, personalized, or optimized for real business outcomes.
Where most AI goes wrong
MIT’s research confirmed what support leaders have felt for a very, very long time:
- Sexy, shiny AI tools often get deployed without deep integrations into existing systems. The result is bots that sound helpful but can’t actually do anything.
- Generic “AI in a box” solutions ignore context and personalization, treating every customer the same.
- Success is measured in “deflection” instead of real outcomes like conversions, saves, or retention.
- Teams treat AI as “set it and forget it,” rather than a system that needs to be tested, trained, and iterated on continuously.
It’s no surprise that when you roll out disconnected bots with no feedback loop, customers hate them, leaders lose trust, and everyone can start writing reports about how AI in general has totally flopped.
Why integration is everything
AI can’t succeed if it lives in a silo.
At KODIF, we’ve learned that AI has to be deeply integrated into the ecommerce stack.
Carts, subscriptions, order management, loyalty, returns, and beyond.
That’s what makes the difference between a bot that says “I can’t help with that, let me transfer you.” vs. an AI agent that actually pauses a subscription, edits an order, issues a refund, or recommends a bundle.
Without these integrations, “AI for CX” is just a glorified FAQ. With them, it becomes a revenue driver.
Personalization isn’t an option, it’s survival
Personalization gets watered down in most AI discussions. Slapping a “Hi, [FirstName] 👋” on a canned reply is not personalization.
Real personalization means:
- Knowing whether this customer is a VIP subscriber or a first-time buyer.
- Adjusting responses based on order history, sentiment, and channel.
- Offering the right retention save (skip, discount, substitution) instead of a one-size-fits-all cancellation flow.
The data flywheel is key: pre-purchase + post-purchase conversations generate richer insights → which improve personalization → which boost conversion and retention → which feed back into the system.
That’s how AI compounds over time, instead of stagnating.
Experimentation > “set and forget”
Another reason MIT says AI fails: companies launch it once and walk away.
We’ve said it a million times, and here it is again: AI isn’t magic. It’s a system that needs to be tuned, tested, and optimized.
For example, KODIF lets CX leaders A/B test workflows:
- Does offering a skip retain more subscribers than a discount?
- Which flows convert more cart abandoners?
- How does containment vary across refund vs. exchange flows?
By measuring resolution, retention, and AOV, not just “deflection,” AI actually gets better with use.
The 5% that succeed
If 95% of AI fails, the 5% that succeed share a few traits:
- Integrated into the business stack. Not a bolt-on.
- Personalized across the full customer journey.
- Outcome-driven, tied to metrics that matter (conversion, retention, saves).
- Continuously tested, with humans in the loop.
- Vertical depth, solving real workflows in ecommerce instead of pretending every industry is the same.
This isn’t theoretical. It’s what we see with customers like Dollar Shave Club, Liquid I.V., JustFoodForDogs, and many, many others: AI isn’t replacing humans, but empowering them while scaling outcomes that move the business forward.
The bottom line
The MIT report is right: a lot of AI is failing. But the lesson isn’t that AI is broken, it’s that bad AI is broken.
Disconnected, generic, one-and-done bots were always doomed to fail. We could have told you this a few years ago.
The future belongs to AI that’s integrated, personalized, and continuously evolving. AI that helps brands deliver experiences once reserved for companies with Uber- or Amazon-level resources.
That’s what we’re building at KODIF. And it’s why we believe the next wave of AI won’t just survive the 95% failure rate, it will define the standard for customer experience.
