Agentic AI in Finance Could Cut Tech Firms’ Costs by 40 Percent

Tech companies ship agentic AI to their customers and put it inside their products, but inside their own finance functions many are still running the piecemeal automation that the rest of corporate America has used for a decade. An analysis of agentic AI in tech finance from EY calls the situation Robotic Process Automation 2.0, AI automating parts of a process while the rest of the function runs on the old stitching-together.

EY’s piece argues the gap is large enough that a move to an AI-first finance operating model can cut costs by up to 40 percent and reshape the function within months. Whether a tech company captures those numbers depends on which finance processes it builds the AI for and which it lets vendors carry.

Why Most Tech Firms Are Still Stuck at Piecemeal Automation

Adam Blaylock, who leads EY Americas Financial Accounting Advisory Services for the technology, media, and telecommunications sector, puts the shortfall in three places finance teams keep running into: siloed data, unclear ownership of AI inside the company, and the change-management drag that comes with handing judgment to a model. Tech leaders can describe the future of AI in finance, Blaylock says, but moving from today’s stitched-together automation to that future is the work most have not yet done.

The drag is concentrated where finance work actually happens. Order-to-cash, the path from a signed contract to a collected payment, runs on multiple systems that do not speak to each other cleanly. Financial planning and analysis pulls from those same systems by hand, every analyst repeating the gather step before doing any actual analysis. Compliance, tax, and record-to-report sit on top, doing their own reconciliation against the numbers the upstream work has not yet aligned. Reimagining the function means rebuilding those handoffs so a single agent can move a transaction through them, not bolting another dashboard on the wall.

What Design for Zero Looks Like in a Real Finance Function

EY’s prescription for the move is a phrase its finance consulting practice now uses inside every redesign engagement: design for zero. The phrase means the team reimagine the process from the outcome backward, asking which steps could be removed entirely if humans were taken out of the loop, then putting people back in only where they are required. Amanda Donohue, a principal in Ernst & Young LLP’s finance consulting practice, frames the test as whether a company is willing to redesign its critical and complex processes itself rather than wait for a software vendor to do it.

The contrast with the current state is sharp. Most finance AI today is invoice extraction here, anomaly detection there, a model that flags what a human will still reconcile.

An end-to-end redesign changes the shape of the work. Routine operational tasks, the matching, posting, reconciling, and first-draft reporting, get handled by AI agents with human oversight. Higher-level tasks, the judgments about pricing, customer credit, working-capital allocation, and the closing-period narrative, stay with people. The two categories do not overlap much, and the operating model is built around that split.

Inside an FP&A team, the immediate change is what disappears. An analyst who once spent the first half of the day pulling data from multiple systems opens a workspace where the data is already gathered, and the first draft of the analysis is already there for review. What the analyst then does with their day is the work the operating model is supposed to protect: partnering with the business, pressure-testing the forecast, and shaping the decisions the company is about to make.

Where to Build First and What to Outsource to Vendors

Not every finance process should be tackled first, and that is one of the more useful pieces of sequencing in the EY playbook. Order-to-cash and FP&A, the two areas the firm flags as priorities for internal development, are also the two where a tech company’s product surface shows up most directly in the numbers. OTC is what makes a tech company different in the marketplace, Blaylock argues, and the AI the company builds for that process will not arrive on a vendor shelf.

Procure-to-pay and record-to-report are the flip side, more standardized across companies and closer to commodities. Major software providers for these processes are already embedding AI into their products. EY’s counsel is to wait until those features land, then buy them, and redirect the internal build capacity into the two areas where the company has to own the work.

The discipline is to draw the line clearly. Build the AI for processes that define how the company makes money, and let the vendor carry the rest.

• Up to 40 percent cost reduction tied to end-to-end agentic AI deployment
• Finance function transformation in months, not years with an AI-first delivery model
• Two categories of finance work in an agentic operating model: routine tasks managed by agents, higher-level tasks by people
• Three common concerns slowing AI adoption in tech finance: tool choice, workforce readiness, enterprise software integration
• More than 20 distinct ERP systems observed across a single set of enterprise customers in a 150-day agentic reconciliation deployment

How Work Splits When Agents Run the Routine Tasks

The full picture of an agentic finance function goes wider than FP&A. Across the lead-to-cash cycle, the path from a generated marketing lead through contract, credit, invoicing, fulfillment, and customer service, agentic AI is doing the connective tissue between systems that have historically not talked to each other. Predictive analytics on customer behavior shape the campaign and the credit decision before the contract is written. After the contract, agents handle invoicing and collections against the contract’s terms, flag renewal or reorder timing, and personalize the follow-up communication without a human drafting it.

The measurable effect on cash flow is what the operating model is built around. Faster, cleaner invoicing compresses days sales outstanding. Better credit decisions reduce the rate at which invoices become write-offs.

The point of the operating model is what happens to the human role. The human stops doing keystroke work and starts doing the work the keystrokes were hiding. The cash application team that used to spend hours matching payments to invoices now spends those hours on the exceptions the agent could not resolve and on the policy calls those exceptions force. The work does not get smaller, but it gets more valuable, and the headcount it needs is smaller than the headcount that did the keystrokes. That pattern is what enterprise deployments of agentic reconciliation, such as OneCap’s 150-day catch across enterprise customers, have begun to surface in production.

The 40 Percent Savings and the Months-Long Transformation

The headline number from the EY argument is up to 40 percent. The savings are tied directly to end-to-end agentic deployment, with most of them coming from labor substitution inside the routine-operations category. Current costs remain high because companies are still in the exploration phase, paying for token use and pilot projects that do not yet touch the operating model.

The second half of the claim is the timeframe. An AI-first delivery model combined with an outcomes-based strategy, Blaylock says, can transform the finance function within months.

End-to-end agentic AI can reduce costs by up to 40 percent and improve financial insights. Current costs remain high due to widespread investment in AI exploration and token costs.

Blaylock is the EY Americas leader for Financial Accounting Advisory Services in the technology, media, and telecommunications sector, a role documented in his professional biography on the firm’s people page. The 40 percent figure is a forecast, not an audited result, and it carries the usual caveats of any cost-savings claim made before a transformation is finished. What makes it worth taking seriously is the mechanism behind it: an operating model where routine work moves to agents and human capacity moves to judgment work that compounds over time.

The Three Hurdles Slowing Adoption

In EY’s framing, each hurdle belongs inside the transformation itself. Each one forces a finance leader to decide what the function will look like once the work is divided between humans and agents. Tech finance officers consistently name the same three.

• Tool choice: deciding which AI tools to invest in, given that each model carries its own data residency, its own pricing curve, and its own behavior under audit
• Workforce readiness: assessing whether the team is ready to work with AI and figuring out how to upskill or incentivize employees to embrace the change
• Enterprise software upgrades: deciding how to integrate AI with major enterprise software upgrades that change the surface the agents were trained against

The first hurdle is a vendor-strategy decision disguised as a technology one. Each model carries its own data residency, its own pricing curve, and its own behavior under audit, and the choice locks in for years once it is wired into the agent stack. For most tech companies, a small portfolio matched to the work each part of the function does is the right shape. That is a different conversation than the one the procurement team is used to having about a single enterprise contract.

The second hurdle is the change-management load on the people side. Upskilling analysts to work with agents is a different problem than reskilling them, and the EY framing implies the former is the entire job. The third hurdle is the one most companies discover after the agents are already running. Enterprise software upgrades change the surface the agent was trained against, and designing for that breakage is the price of admission for an AI-first finance function.

Why a Validation Layer Is Becoming Non-Negotiable

The other half of the EY argument is one most AI-in-finance stories skip past. Agentic AI makes it easy for any employee to write code and stand up an automated process, and there is no built-in guarantee the tool is delivering accurate data or accurate responses. The EY view is that finance functions need to add a validation layer that monitors the proprietary models running inside them, checking outputs against expected behavior and flagging drift. Without that layer, the cost savings inside the operating model are paid for by an accuracy tax that compounds quietly across every closing cycle.

The validation layer is what turns AI in finance from a leap of faith into something a controller, an auditor, and a regulator can each verify on their own. It is also where most companies are thinnest, because it is a category of work the cloud vendors do not provide. Building it is a build-versus-buy decision with no clean buy answer, and that is why it tends to be the part of the redesign that gets under-funded.

Frequently Asked Questions

What is agentic AI in finance?

Agentic AI in finance refers to AI systems that execute end-to-end finance workflows such as invoice matching, reconciliation, and first-draft reporting with human oversight on the routine layer and human judgment on the higher-level layer. The defining feature is that the agent chains multiple steps toward an outcome rather than executing one rule at a time.

What does “design for zero” mean in a finance redesign?

“Design for zero” is an EY finance consulting framing that means reimagining a process from the desired outcome backward, removing steps that exist only because humans used to do them, and inserting human review only where it is genuinely required. The point is to redesign the work around the absence of human keystrokes before deciding where humans belong.

How much can agentic AI cut tech companies’ finance costs?

Adam Blaylock, EY Americas FAAS TMT Industry and Technology Sector Leader, has said end-to-end agentic AI can reduce tech company finance costs by up to 40 percent. The figure is EY’s framing, not an audited result, and Blaylock ties most of the savings to labor substitution inside the routine-operations category.

Which finance processes should tech companies automate first?

EY’s sequence is to build internally for order-to-cash and FP&A first, because those are the areas where a tech company’s product shows up most directly in the numbers and where off-the-shelf AI is unlikely to fit. Procure-to-pay and record-to-report are typically more standardized and can wait for AI features to land inside the software providers’ products.

What are the biggest barriers to adopting AI in tech finance?

EY identifies three: choosing which AI tools to invest in, assessing whether the workforce is ready to work with AI and figuring out how to upskill or incentivize them, and deciding how to integrate AI with major enterprise software upgrades. Each barrier forces a decision about what the function will look like once work is divided between humans and agents.