A recent MIT report delivered a striking conclusion: 95 percent of enterprise AI pilots fail to deliver measurable value. So influential was the publication of this paper, it resulted in a significant sell-off of publicly traded stocks that have AI exposure.
After reviewing more than 300 implementations, interviewing executives from 52 organizations, and collecting 153 executive surveys, researchers at MIT uncovered a stark “GenAI Divide.” In one group are the 5 percent of companies seeing millions in returns. In the other, making up the vast majority, are companies stuck with expensive experiments that never scale.
The difference comes down to whether the AI can learn.
Why most AI projects fail
The MIT researchers found that most systems don’t retain memory, can’t adapt to specific workflows, and never improve from user feedback. Users may enjoy tools like ChatGPT for quick drafting, but they abandon it for high-stakes work because the systems forget context and repeat mistakes. This learning gap explains almost every failure.
As the report put it, “tentatively right beats confidently wrong.” Static AI systems confidently produce the same errors over and over. For finance teams, that difference matters. When your AI doesn’t remember that certain vendors always invoice in a particular way, or that specific categories need special handling, it becomes a novelty rather than a partner you can trust.
“Organizations that cross the GenAI Divide discover that ROI is often highest in ignored functions like operations and finance. Real gains come from replacing BPOs and external agencies, not cutting internal staff. Front-office tools get attention, but back office tools deliver savings.”
MIT, Project NANDA
The GenAI Divide: State of AI In Business 2025
Finance AI delivers ROI
The MIT report revealed that over half of AI budgets flow into sales and marketing, yet “back-office automation often yields better ROI.” High-return use cases included accounts payable and receivable automation, outsourced risk management reduced by more than a million dollars annually, and BPO contract elimination.
In other words, finance AI delivers the fastest payback periods. AppZen customers see it every day, through shorter cycle times, faster audits, and millions saved through compliance and fraud detection.
Partnerships beat built-it-yourself
Critical to this ROI are external partnerships, which researchers revealed succeed twice as often as internal builds. Building AI that truly learns requires years of domain expertise, continuous customer feedback, and the scale to recognize the rare exceptions.
According to the paper's authors, “Organizations on the right side of the GenAI Divide share a common approach: they build adaptive, embedded systems that learn from feedback.” Enterprises that attempt in-house builds often hit the roadblocks they described, including brittle systems, stalled pilots, and tools that never adapt.
AppZen closes the learning gap
Ninety percent of employees already use personal AI tools at the individual workflow level. Yet they often report the learning and memory capabilities are unreliable within enterprise systems.
AppZen’s AI was built to avoid these problems. It adapts and improves from every interaction. The system remembers when your team corrects a categorization or flags a policy exception. Instead of repeating the same errors, it applies what it learned across future transactions. This adaptive intelligence turns finance automation into a system that grows stronger over time.
Unlike generic tools, our platform was also designed specifically for finance. Expense policies, compliance requirements, and vendor management practices are embedded into the AI from the start, giving it the domain fluency enterprises demand.
The philosophy of “tentatively right” lives in our confidence framework. When our AI encounters a transaction where confidence is low, it doesn’t force a decision. It flags the case for human review. This keeps your team in control while the system learns from the outcome. This balance of speed, accuracy, and oversight is what builds long-term trust in AI-driven finance.
AppZen brings the ease and reliability of personal AI tools into official, enterprise-grade systems, adding the learning capabilities, compliance guardrails, and domain depth that enterprise-level AI users require.
What enterprise executives really want
Arguably the best part of this article (from our perspective) are the direct quotes from the executives the researchers interviewed. Their thoughts reveal how AppZen sits solidly in this report’s sweet spot:
“We’re more likely to wait for our existing partner to add AI than gamble on a startup.”
Our decade-long track record of successful enterprise deployments means you’re dealing with a vendor you can trust. And building a dedicated AI platform specifically for finance beats bolting one onto an existing solution.
“Most vendors don't get how our approvals or data flows work.”
We get you. Not only are our solutions built specifically for finance operations, they also learn from and adapt to your unique workflows.
“If it doesn’t plug into Salesforce or our internal systems, no one’s going to use it.”
Not only do we work with your existing tools, our integrations are easily installed with minimal disruption or IT effort.
“I can’t risk client data mixing with someone else’s model, even if the vendor says it’s fine.”
Our platform performs expense report auditing and invoice processing using only aggregated, anonymized data. It maintains clear data boundaries with the highest data privacy standards.
“It’s useful the first week, but then it just repeats the same mistakes. Why would I use that?”
Because our products have the ability to learn, they actually continue to improve over time. As does your ROI.
“Our process evolves every quarter. If the AI can’t adapt, we’re back to spreadsheets.”
Flexibility and adaptability are the hallmarks of AppZen’s AI models.
Finance teams that have adopted our AI-driven Mastermind platform are benefitting from a partner that crossed that GenAI learning divide long ago. Our continued success across some of the world’s largest global enterprises is proof of that.
Time is running out
The researchers warned that the window for adopting the right AI tools is closing fast, noting that enterprises are likely to lock in AI vendor relationships over the next 18 months. Once these systems are trained on specific workflows, they will be difficult to unwind.
The organizations that choose learning-capable partners today will compound their advantage, while those that hesitate risk falling permanently into the 95 percent of failed implementations.
Locking yourself into tools that forget everything and repeat the same mistakes doesn’t make sense. Success comes from AI that learns and improves over time, partners who understand your workflows, and solutions that focus on clear ROI areas like finance operations. AppZen delivers all three.
Ready to cross the AI divide?
Our Mastermind AI Platform was helping the world’s largest global enterprises, well before ChatGPT became part of the public consciousness. They trust our solutions to automate their accounts payable, audit expenses, and deliver measurable ROI every day.
Schedule a demo today to discover how AppZen’s learning-capable, finance-specific AI solutions put you on the winning side of the GenAI Divide. Join the 5 percent seeing real returns with AI that actually learns.
