Ahsan Shah, SVP AI & Analytics, Billtrust.
Forrester predicts that, in 2026, one-quarter of CIOs will be asked to bail out business-led AI failures in their organizations.
With the recent wave of generative AI and LLMs changing how AI is leveraged in an organization, there are some common failure patterns to avoid. Here are three reasons why AI projects fail—and some insights into how to get them right:
1. Building Solutions For The Wrong Fundamental Problem
When finance leaders ask me why their AI initiatives stalled, they usually expect to hear about data quality issues or model accuracy problems. Those matter, but they’re rarely the root cause.
The most common failure point happens much earlier: when teams build solutions for the wrong problem. A collections team wants to “automate dispute resolution” when the actual challenge is maintaining customer relationships while accelerating cash flow. A credit team wants to “predict payment risk,” when what they really need is a system that adapts credit decisions based on changing customer behavior.
This disconnect happens because organizations approach AI as a technology deployment rather than a business transformation. They ask, “What can AI do?” instead of “What outcomes do we need to achieve?”
How To Get It Right
To see real returns from AI, finance teams should start by first identifying the business need and understanding where AI can be applied, as opposed to applying AI arbitrarily.
Instead of saying, “We need an AI chatbot for customer inquiries,” successful teams ask, “How do we resolve customer issues faster while strengthening relationships?” In collections, the outcome might be “reduce days sales outstanding while maintaining customer satisfaction scores.”
2. Replacing Human Judgment Instead Of Amplifying It
In B2B payments, every customer relationship has nuance. Payment delays might signal financial distress or operational challenges. AI can surface patterns and anomalies across thousands of accounts, but finance professionals provide the context that turns data into strategy.
The adoption numbers tell a clear story. Recent Billtrust research found that 90% of financial decision-makers now rely on AI for financial decisions, with 83% reporting that AI has positively influenced their approach to managing financial risk. But adoption alone doesn’t guarantee success.
When organizations design AI systems that bypass human expertise rather than elevate it, they lose the judgment that creates real business value. The result is either systems that produce technically correct but strategically useless outputs, or teams that don’t trust the AI enough to use it.
How To Get It Right
Build decision frameworks before deploying models. Get specific about when AI should act autonomously, when it should recommend actions for human approval and when it should escalate to human experts. Allow AI to act without human approval at a certain threshold.
For credit decisions, my company established clear thresholds: AI can approve standard transactions within defined parameters, flag unusual patterns for review and escalate cases with conflicting signals. These guardrails let the system operate at scale while maintaining control over risk.
Just as important: Invest in building feedback loops. AI systems need to learn from human decisions, and humans need to understand how AI reaches its conclusions. Build systems that capture when specialists override AI recommendations and why. This data improves model performance over time and helps identify when the AI is right and human instinct is wrong, which happens more often than people expect.
3. Underestimating The Organizational Change Required
Leaders focus on algorithm selection and model training while ignoring the harder questions: How will decision-making authority shift? What happens to existing workflows? How do we measure success differently?
I’ve watched promising AI projects collapse not because the technology failed, but because the organization wasn’t ready to work differently.
How To Get It Right
The companies getting real value from AI in finance have redesigned how work happens, not just added AI tools to existing processes.
This requires a fundamental shift in leadership thinking and a reassessment of skills across the team. Traditional management assumes human oversight of every decision point. AI-augmented organizations define the strategic boundaries and let intelligent systems navigate within them. This approach scales decision-making beyond human limitations while maintaining direction.
It also requires new skills across finance teams. The valuable capability in an AI-augmented workflow is the ability to provide context, articulate business logic and interpret results strategically. Technical comfort with AI tools matters, but critical thinking about when to trust or question AI outputs matters more. Team members who are versatile and understand either the technology or the business process, and can also articulate those business processes for context for AI, need to be cultivated as assets.
Train teams on how the AI works, but spend equal time on scenario planning: What do you do when AI flags something unusual? How do you validate recommendations? When do you escalate?
Building AI Systems That Actually Work
Success starts with picking the right first project. Identify one area where human judgment is valuable but constrained by volume or complexity. Build a pilot with clear success metrics and strong feedback mechanisms. Focus on learning what works in your specific context rather than chasing perfect accuracy from day one.
Most critically, involve your finance teams in designing how AI will augment their work rather than surprising them with finished systems. The teams that will use these tools daily understand the nuances of customer relationships, payment behaviors and risk factors better than any algorithm. Their input during design prevents the exact failures Forrester predicts CIOs will need to fix.
The goal is to create environments where AI handles the routine so humans can focus on strategy. When finance professionals spend less time chasing data and more time interpreting what it means, that’s when AI delivers real returns.
Organizations that get this right will have incredible leverage in the years ahead and better position their human capital for scalability and growth. Those that treat AI as just another efficiency tool will struggle to keep pace with competitors who’ve fundamentally rethought how financial decisions get made.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
