The era of AI was ushered in with dazzling displays of its seemingly endless capabilities and a promise to change the world. After OpenAI sparked the generative AI boom in 2022, the pace and scale of AI growth in just about every major industry have been staggering.
These days, although many of the promises have been materializing, it’s left the AI industry itself with a self-inflicted headline problem.
It’s the headlines about spending that have been most troubling. For example, $675 billion in infrastructure investment this year alone and major AI companies like Nvidia surpassing $5 trillion market caps are the kind of news that dominate business headlines. When it comes to adoption, according to KPMG, 93% of US companies will deploy AI in finance within the next 18 months. Look at headlines about layoffs and you’ll find CEOs flattening organizations in the name of efficiency.
Headlines like these have begun to slowly turn AI into public enemy number one. More so because there’s one headline that still barely exists after all the hype and the outrageous amounts of money being spent: “AI generated $X million in measurable new value for this company.”
Far from it, MIT actually found that 95% of enterprise AI pilots deliver zero measurable P&L impact. Cambridge’s 2026 Global AI in Financial Services Report found that while 81% of firms are adopting AI, only 14% consider it transformational. The money is certainly flowing in but, for now, it seems clear that the results aren’t flowing out.
Afrozy Ara, Founder and CEO of San Jose-based LuminaData, believes she knows why and her background explains why she sees the problem differently than most AI founders.
Before LuminaData, Ara spent over a decade in enterprise consulting. She first began at Mu Sigma, where she advised Fortune 500 companies on data and analytics strategy. She was later the VP of Consulting at Incedo, where she led teams helping large enterprises operationalize data across complex, multi-system environments.
The difference was that her background wasn’t in AI research. Instead, Ara’s experience had exposed her to the messy reality of making technology work inside organizations that run on tangled processes interlaced with tribal knowledge, and cross-functional coordination.
That experience helped shape her core thesis:
“AI adoption is not AI transformation,” Ara says. “That’s the distinction most of the market still hasn’t made, and the cost of missing it is enormous.”
The coordination problem #
Ara’s stance is deceptively simple but can carry enormous implications for the way it changes how enterprises should approach AI.
Think about it this way, every individual approaches and interacts with AI at the speed and limits of their own curiosity. Provide them unfettered access to tools like Claude or ChatGPT and their productivity will inevitably increase in measurable ways. Once they begin to grasp the possibilities, intelligence can flow between form factors.
From writing documents and emails to creating bespoke spreadsheets or even code, AI can have a remarkable effect at individual level. However, an organization’s growth can only be as fast as its slowest coordinating seam. The problem is, those seams were built for a slower world. That world is limited by handoffs and ownership boundaries that work on the assumption that change will only happen occasionally and gradually. “Organizations aren’t individuals,” Ara says. “They’re people coordinating toward a common goal. They need shared understanding. They need to see results together, not just as individuals working faster in their own silos. That’s where adoption hits a wall.”
According to Ara, these are the kind of patterns she comes across all too often in virtually every finance team she works with. These teams inevitably work with tools deployed at multiple points in a workflow, each performing well in isolation. That means the process between those points remains manual and undocumented, essentially leaking value as a result.
“I’ve seen order-to-cash processes where the cash application is automated, the reports look beautiful, the dashboards are AI-generated and DSO is still climbing every quarter. Even though the tools work, the process between them is broken and you end up with beautiful outputs from a broken input chain.”
The pipe dream and the 99.999% #
Ara is blunt about the gap between Silicon Valley’s vision and enterprise reality.
“The startup ecosystem loves the story of the one-person billion-dollar company. And sure, maybe it could happen. But it would be more luck than a playbook. For the 99.999% of companies out there, that’s a pipe dream.”
Real companies, she argues, are a complex, tangled combination of people, technologies, and systems. This means the real processes are convoluted in incalculable ways. Another truth is that tribal knowledge runs through the ecosystem. It begins by existing in people’s heads but then quickly makes its way into things like undocumented spreadsheets, or just in the way things have always been done.
“Even when we talk about ‘ human in the loop,’ the humans in that loop have to be able to keep up with the AI,” Ara says. “
They need to understand what the AI did, why it did it, and whether to trust it. They need to coordinate with each other around AI-generated outputs. That’s not a technology challenge. That’s an organizational design challenge.”
It is within the somewhat paradoxical nature of this challenge, the organizational design challenge if you will, that LuminaData is built around.
What LuminaData actually does #
LuminaData is backed by Techstars and has built an AI transformation platform for finance operations. It specifically targets order-to-cash and record-to-report workflows. The company uses FinEdge LLM, a patent-pending LLM trained on GAAP, SOX, ASC 842, IFRS 16.
The platform itself is structured around two products that mirror the company’s framework. The first is Lumina Discover, a tool that maps a finance team’s real workflows. That means understanding the actual handoffs, the tribal knowledge behind the workflows, and all the undocumented rules that accompany their execution.
This establishes measurable baselines before any AI agent is deployed. It’s the diagnostic layer that most AI implementations skip entirely.
The second tool is Lumina Activate, which deploys finance-specific AI agents on the rule-based components of the redesigned workflow. That means handling reconciliation, invoicing, cash application, and lease accounting, all with documented matching logic, exception routing, and audit-ready documentation built into every action. While this kind of automation is becoming more popular these days, the kind of deep expertise and personalization offered by LuminaData is quite novel.
The two are designed to work as a sequence, not a menu. Discovery informs activation, so activation without discovery is what produces the 95% failure rate the rest of the market is experiencing.
Ara insists the products are only half the story.
“The agent handles the rule-based work. But the reason most AI deployments fail isn’t that the agent doesn’t perform. It’s that nobody examined the process the agent was deployed into. You can’t automate a broken workflow and expect a transformed outcome.”
This means that the framework isn’t theoretical. Ara goes on to describe a typical engagement.
“We had a customer who thought their order-to-cash process had five handoffs,” she says. “When we mapped it, there were sixteen. Most were manual and undocumented. Several were contradictory between business lines. An AI tool was sitting at one of those sixteen handoffs, working perfectly. The other fifteen hadn’t changed.”
The discovery process is essential because it surfaces what is usually invisible. Imagine the billing rules that live in three people’s heads or the spreadsheet only one person understands. It’s these kinds of scenarios that create the handoffs like “that’s how we’ve always done it.” It also establishes baselines. Think things like DSO, billing timeliness, exception rates, dispute resolution time, and close duration because these can make the impact measurable from day one.
Activation follows the discovery, so it’s not merely part of a sales cycle. Agents are deployed on the rule-based components of the redesigned workflow, with human oversight at critical judgment points. In this way, it means the roles are redefined around how and what AI gives you back. There’s less matching and more investigating, something that flies in the face of how typical AIs work but also what makes them so predictable and monotonous in their output.
“We’ve seen companies uncover 7–15% of revenue as EBITDA savings through this process,” Ara says. “But the savings won’t come from the AI alone. They come from the transformation underneath. With the standardized rules, the eliminated handoffs, and things like the connected systems, the AI just makes the new design scalable.”
The market opportunity #
LuminaData is entering a market with significant momentum but that also carries significant confusion with it.
KPMG projects that within 18 months, half of US companies will be orchestrating multi-agent AI systems across finance workflows. Bain found that 83% of CFOs plan to increase AI budgets by 15% or more over the next two years, with the largest allocation going to financial planning, analysis, and reporting.
Despite all this, the gap between spending and results still remains wide. Only 8% of organizations have established ROI from their AI deployments, according to KPMG. S&P Global found that 42% of companies abandoned most of their AI projects in 2025. In a very real sense, many companies are finding the limits of AI hype out the hard way. However, that doesn’t mean AI is a failure, just that the real winners will be those AI innovations that learn from the early mistakes and correct them in meaningful ways.
“There are gazillions being spent on tokens and data centers,” Ara says. “The ROI headlines will come. But they’ll only come for the companies that solve the transformation problem, which is the thorny, messy, human problem of making AI work across teams, systems, and processes that were never designed to talk to each other.”
What’s next #
LuminaData is currently scaling its Discover + Activate platform beyond early adopters. Its early focus is on making the discovery process fast enough and accessible enough that any finance team can run it without a consulting engagement.
“The teams that succeed don’t start with better AI,” Ara says. “They start with better questions. Where are we spending the most time on work that follows rules? What are the handoffs? What’s the baseline? What does each person do with the time AI gives back?”
She s. “Those questions are the transformation. The AI is just the tool that makes the new design possible.“
Get the TNW newsletter #
Get the most important tech news in your inbox each week.
Opinions expressed by The Next Web authors are their own