Only about 8% of enterprise AI transformations succeeded last year, with the vast majority failing due to poor data readiness and lack of organizational preparedness rather than technology limitations.
Arunima Thakur, Founder of Innoengg Technologies and former Consulting Managing Director at Deloitte, argues that AI cannot be deployed effectively until enterprises clean up their processes through small, incremental changes across the entire value stream.
Employee empowerment is the prerequisite to useful AI: workers closest to the process are the only ones who can surface the data and make the adjustments that AI models need to deliver value.
Seventy-eight percent of AI transformations failed last year, and more than half are expected to falter again in 2026. The cause? Organizations are trying to layer AI onto messy, disconnected value streams. Without disciplined process improvement, aligned priorities, and empowered frontline employees generating clean data, AI has nothing to amplify and nowhere meaningful to embed.
Arunima Thakur is the Founder of consultancy Innoengg Technologies and a former Consulting Managing Director at Deloitte, where she built the firm's Manufacturing-as-a-Service solution practice. With 33 years leading billion-dollar transformations across GM, Ford, and Booz Allen Hamilton, Thakur is a certified change management professional and Six Sigma Master Black Belt whose career rests on one conviction: transformation starts with process discipline, not technology deployment.
"If you don't look at the entire value stream in an enterprise, from inception all the way to selling the product, and make tiny incremental changes, you cannot really implement AI in a meaningful way," says Thakur. That conviction is rooted in the concept of Kaizen, the Japanese continuous improvement methodology Thakur learned firsthand at Toyota's plants in Nagoya early in her career. The principle is simple. A Kaizen practitioner walks the production line and makes one small adjustment at a single node, then moves to the next. Over the course of a shift, dozens of minor fixes snowball into a measurable transformation across the entire value stream.
Snowballing progress: Thakur illustrates the compounding effect with an example from a Toyota plant. A line worker is given a 90-degree attachment for a torque gun so he holds the tool differently. Less fatigue means fewer worn-out screw threads at the next station, fewer quality defects downstream, and fewer vehicles in the repair bay. One ergonomic change cascades through the entire operation. "Now let's talk about it in the sense of AI," Thakur says. "You can go very granular based on all the data you get from every single node. AI can tell us before we even put that change in place what the changes need to be, how much we can expect in terms of repeatability, and what it will snowball into in quantified numbers."
But the data AI needs only exists if the underlying processes are clean enough to produce it. Thakur points to a simple example: a TV bracket manufacturer and a retailer like Home Depot operate on disconnected product data. A small change to the part-numbering system on either side could link them, and AI could detect that gap and connect the datasets. Without that incremental fix, two silos persist and no AI deployment can bridge them.
The disconnect test: Before any transformation, Thakur runs a diagnostic she has refined over decades. She asks one question at every level of an organization: what are your top two priorities for the next six months? The CXO may say cost and quality while the plant manager says absenteeism. A logistics manager says supplier backlogs, a line supervisor says morale. "Those are all symptoms of the bigger root cause," Thakur explains. "Usually just haphazard, disconnected processes that have not been able to keep up with growth." The exercise builds the case for change before a single AI model is deployed.
Boots on the ground: Thakur is equally direct about why employee empowerment is the prerequisite, not the byproduct, of AI transformation. She tells a story about the Ritz Carlton in Tysons Corner, where a housekeeper noticed a guest's snow-covered leather boots and had them cleaned while the guest was out. Ritz Carlton empowers employees to spend up to $250 on guest experience without approval. "AI cannot figure out if a guest had snow-covered boots," Thakur says. "But AI can figure out that during winter months there was a 20% spike in snow-covered boots being cleaned at Tysons Corner. From that you can predict, and from the prediction you can measure loyalty and sales." The insight only exists because an empowered employee made a judgment call and created the data point.
The wider pattern holds. Over the past several years, Thakur has watched the disconnect between leadership and frontline workers widen, accelerated by remote work. Leaders are not seeing employees daily, trust is harder to build, and ground-level intelligence no longer flows to the top. Without that feedback loop, the data AI needs never gets logged. "You're adding more work to employees who think they're losing their jobs to this technology," she says. "There's a whole vicious cycle. You just can't go guns blazing with technology and expect things to fall into place."
The fix is not smaller AI budgets or better models. It is the willingness to start with the process, clean the data at the source, empower the people closest to the work, and let the improvements compound. "You do need to make these tiny changes every day for AI to really get embedded in the process."
