The argument in 30 seconds:
AI is making people more productive, but productivity alone isn’t changing the economics for most companies. The bigger opportunity is to put AI inside the work that drives growth, margin, working capital and cash, then measure AI ROI against outcomes the business already values.
I’ve spent nearly four decades in enterprise software. I’ve run a few companies, sold one of them to Microsoft, and sat through more technology cycles than I care to count. They all end in the same reckoning, when somebody from finance asks: How does this improve company value?
Why AI isn’t showing up in EBIT
AI is getting that question now, and frankly it’s overdue.
I’m a believer. I use AI every day, and I think it’s the most important shift of my career since the internet, probably bigger. But companies don’t make more money because a document got summarized faster, and the data is starting to say so. McKinsey’s latest State of AI survey found 80 percent of respondents say AI has improved their individual productivity, but only 37 percent report that AI has contributed positively to their organization’s EBIT. Everybody is getting more done. Far fewer companies can show it changed their economics.
AI has been pointed at the wrong work. The technology isn’t the problem.
Consider how a manufacturer produces cash. They decide what to build, configure it for a customer, price it, quote it, sell it, make it, ship it, invoice it and collect. That’s the order-to-cash cycle, and it lives or dies on how fast inventory turns and whether what was promised can be delivered.
That is the business, and most of the first wave of enterprise AI has been parked next to it instead of inside it. Search, summarize, forecast, write the deck. Some of it is genuinely useful. Some of it is cute. None of it has ever stopped a credit memo.
The gap between AI productivity and AI ROI
Companies have spent a lot of money teaching AI to understand their business. Letting it help run the business is a much harder job with a much bigger payoff, and the first wave’s returns show the gap. Deloitte’s Finance Trends 2026 research found 63 percent of finance leaders have fully deployed and actively use AI, but only 21 percent report clear, measurable ROI. Gartner found 45 percent of AI investments in finance lean toward productivity and just 20 percent toward better decisions and warns there’s a ceiling on what productivity gains return. Making one task faster only goes so far when the process around it still breaks.
Earlier this year I wrote about what I call the CFO problem. Analytics can tell you about leakage after it happens, but they can’t stop the wrong transaction from being created. So, I keep asking CFOs: Are you real-time in reporting, or real-time in execution based on truth? AI raises the stakes, because now we’re choosing between explaining the problem faster and finally fixing it.
Picture one sales order for a made-to-order manufacturer. The configuration is slightly off, which changes the price and throws an exception that stalls the order. The rep promises a ship date against inventory already allocated to another customer. The shipment splits, the invoice doesn’t match the purchase order, the customer short pays, and accounts receivable spends the next month reconciling a problem created six steps and three systems earlier. Finance owns the credit, the rebill and everything around it.
What one broken sales order costs in DSO and working capital
Most companies have lived with this so long it sits in the budget as a cost of doing business. Why?
Because in the end, it’s about cash. For a company doing $1 billion in annual credit sales, meaning sales invoiced to customers for later payment, every additional day of days sales outstanding (DSO) ties up roughly $2.7 million in receivables instead of in the business. Same with hard assets. A plant running at 70 percent while orders wait on a configuration no one can validate is capacity you paid for and aren’t getting a return on. Inventory in the wrong warehouse because nobody trusted the availability number is working capital doing nothing. Finance can report on all of it beautifully. Reporting doesn’t give you the cash back.
Finance leaders know where the value is. When Deloitte asked where agentic AI could help most, the top answers included sales and profitability management at 48 percent and working capital optimization at 46 percent. Those are execution problems, and the market is heading the same way. Microsoft’s Satya Nadella says AI experiences are evolving from answering questions toward executing multi-step tasks with clear user control points. ServiceNow CEO Bill McDermott puts it more simply: “AI that thinks and workflows that act.” Different language, same destination.
Analytical AI vs. Executional AI
That’s the move from analytical AI to executional AI, and it changes the value equation. If AI gives you a wrong answer, you had a bad meeting or maybe a missed sale. If it acts on wrong information, you’ve created a bad transaction. Executional AI acts inside the transaction itself, working from live business rules and data, so the wrong order is never created. For most established companies, the ERP is where decades of business rules meet the transactions that determine revenue, margin and cash. If AI is going to touch those transactions, it must work from that reality as it exists right now, not a copy of it from last night’s sync.
Most companies aren’t there yet. Accenture found only 7 percent of the 2,000 companies it studied have the data readiness to scale advanced AI. I’m not surprised that Deloitte’s newest Finance Trends research found 95 percent of finance leaders are comfortable with agentic workflows somewhere in finance, but only 14 percent support full autonomy for critical decisions. A CFO asking what happens when it’s wrong is asking exactly the right question.
And increasingly, that CFO decides what gets funded. The same Deloitte research found 54 percent of finance leaders lead cross-enterprise AI and technology capital allocation. If I were in that seat, here’s how I’d spend it.
How CFOs should fund AI: start with broken work
I wouldn’t start with a list of AI use cases. I’d start with the drivers of business value, meaning growth, margin, working capital and asset use, and go find the work getting in the way. Where does margin leak between the quote and the invoice? What are we spending every quarter on credits, rebills and rework? Where is cash stuck in disputes? Which assets sit underused because the front of the business can’t see what the back can deliver? Where could AI help us win business we’re missing?
Pick one. Baseline it against a number your board already tracks: growth, DSO, inventory turns, order cycle time, asset utilization or margin. Then fix it, using AI where AI is the best way to do it. With what frontier models can do now, the potential is enormous.
That’s the difference between funding an AI project and funding broken work. An AI project gets you a demo and an adoption chart. Fixing broken work gets you growth, margin and cash flow. Point AI at the right work, give it governed access to the right data and business rules, and I think the results are going to surprise a lot of people. That’s why I’m so bullish.
Every technology cycle I’ve lived through eventually faced the same test:
- Did it change the economics?
- Did we sell more?
- Did we use our assets better?
- Did we get the order right the first time?
- Did we collect faster?
That’s basic finance, and it has driven valuations for more than a century. If the answer is yes, keep writing checks. If it’s no, we’re working on the wrong problem.
What’s next: arnold® AI for SAP
That test is exactly what arnold® AI is designed to pass. It’s a trusted execution layer that connects SAP to frontier AI platforms such as Claude and enterprise AI platforms such as Microsoft 365 Copilot and Copilot Cowork, Salesforce and ServiceNow. It gives AI secure, governed, role-based access to live SAP data and the business rules behind it. We’re putting it to work first in high-value sales and customer-service workflows, where so much of the broken work in this article begins.
In my next piece, I’ll show what executional AI looks like when you put it inside the work that drives the business.
About Nick Fera
Nick Fera is Chief Executive Officer of enosix and Managing Director of enosix GmbH, leading the company’s strategic, operational and financial activities globally. He brings nearly four decades of experience in B2B enterprise software and SaaS, from early-stage companies to public enterprises. Before enosix, Nick served as CEO of Firm58 and earlier as CEO of Parlano, where he led the company’s 2007 sale to Microsoft. He holds a BS in Finance from the University of Illinois and an MBA from Northwestern University’s Kellogg School of Management.
Trademark notice: arnold® is a registered trademark of enosix, Inc. enosix is a trademark of enosix, Inc. SAP and other SAP products and services mentioned herein are trademarks or registered trademarks of SAP SE or its affiliates in Germany and other countries. Microsoft, Microsoft Copilot and related Microsoft products and services are trademarks or registered trademarks of Microsoft Corporation in the United States and/or other countries. ServiceNow and related ServiceNow marks are trademarks and/or registered trademarks of ServiceNow, Inc. in the United States and/or other countries. Salesforce and related Salesforce marks are trademarks of Salesforce, Inc. Claude and Anthropic are trademarks of Anthropic, PBC. All other company, product and service names referenced are the property of their respective owners.
Frequently asked questions
Why isn’t AI improving EBIT for most companies?
Most of the first wave of enterprise AI has been parked next to the business instead of inside it: search, summarize, forecast, write the deck. McKinsey’s State of AI 2026 survey found 80 percent of respondents say AI has improved their individual productivity, but only 37 percent report that AI has contributed positively to their organization’s EBIT. Making one task faster only goes so far when the process around it still breaks.
What is executional AI?
Analytical AI answers questions and reports on what happened. Executional AI acts inside the transaction itself, working from live business rules and data, so the wrong order is never created. Because it acts, it must work from the ERP as it exists right now, not a copy of it from last night’s sync.
What should CFOs fund instead of AI projects?
Start with the drivers of business value, meaning growth, margin, working capital and asset use, and find the broken work getting in the way. Pick one. Baseline it against a number your board already tracks, such as DSO, inventory turns, order cycle time, asset utilization or margin. Then fix it, using AI where AI is the best way to do it.
How do you know if an AI investment is working?
Apply the same test every technology cycle eventually faces:
- Did it change the economics?
- Did we sell more?
- Did we use our assets better?
- Did we get the order right the first time?
- Did we collect faster?
If the answer is yes, keep writing checks. If it’s no, you’re working on the wrong problem.