Crawl, walk, run: A practical AI adoption roadmap for finance teams

For finance teams, the idea of adopting AI can make it sound like you need to rethink the entire function at once. But it doesn’t have to happen that way.

The teams furthest along aren’t making one big AI bet. They’re putting individual capabilities to work, learning what delivers value and building from there.

Most finance teams are still early in that process. According to PEX’s 2026 State of Finance report, 91% either haven’t started with AI or are experimenting or in early adoption. Just 9% have reached the scaling or advanced stage.

Getting from one stage to the next doesn’t require a dramatic transformation. Finance teams just need to begin with one problem worth solving.

What finance teams can gain from AI

The payoff from AI isn’t just doing the same work with new technology. It’s reducing the manual work that consumes your team’s time and giving them more capacity for work that requires their expertise.

The State of Finance data shows that those benefits become more pronounced as teams put AI into practice. Among teams experimenting with or beginning to adopt AI, 61% say it has reduced manual review and 42% say it has shortened close times. Among teams scaling their use of AI, both figures rise to 82%. 

That can mean fewer transactions to review by hand, less work piling up before month-end and more time for your team to focus on the financial decisions that need human judgment.

Teams further along are also far more likely to see a return. Reported ROI rises from 3% of teams that haven’t started with AI to 25% of teams experimenting or in early adoption and 59% of teams scaling it. The teams furthest along also run an average of 3.7 live AI capabilities, compared with 1.6 and 0.5 at earlier stages, reinforcing the case for building on what works rather than perfecting a single use case.

The question, then, is how to get there without trying to change everything at once. A crawl, walk, run approach gives finance teams a practical way to start small, prove what works, operationalize it and expand from there. 

Crawl: Start with the low-hanging fruit

If you’re just getting started, look for a painful, repetitive task where AI could make an immediate difference. You don’t need to tackle your most complex finance process first.

Routine transaction work is a natural place to look. Tasks like categorizing transactions, matching receipts and invoices and reviewing documentation happen over and over, often according to clear rules.

And they’re already consuming a significant amount of finance teams’ time. Thirty-one percent spend more than 20 hours a month collecting receipts, coding transactions and reconciling expenses. Receipts and reconciliation are also the most commonly cited finance operations challenge.

The best first AI use cases also have clear boundaries. Look for repetitive work with predictable inputs and outputs, where AI’s work can be checked before it affects the books. That could mean matching receipts with transactions or suggesting GL codes while a person reviews the output when needed. Keeping a person in the review loop matters because trust in AI accuracy is the top barrier to expanding AI, named by 36% of finance leaders.  

You may be here if: Your team hasn’t put AI into regular use yet or is testing individual use cases.

What to do next: Choose one repetitive, well-defined task where AI’s output can be reviewed and put it into everyday use.

Walk: Build on what’s working

Once you’ve seen results from your first AI use case, look at the workflow around it. You don’t need to perfect the first use case before moving forward.

At this stage, the question changes from “Can AI handle this task?” to “Where else in this workflow could AI help?” Look immediately before and after the capability that’s already working and identify another task AI can handle or another capability that can strengthen the workflow.

For example, a team that starts with AI-powered transaction categorization might connect it with capabilities such as fraud and anomaly detection or real-time spend controls. Instead of collecting disconnected AI pilots, you’re building outward from something that has already proved useful.

As AI becomes part of a larger workflow, give someone responsibility for overseeing that workflow and its results. Start measuring things like time spent, whether required information is complete and how many exceptions still need human attention. Compare results before and after each change so the case for expanding is easy to make. 

This is also an opportunity to move controls earlier in the process. Instead of identifying problems during reconciliation or after money has been spent, look for ways to catch them closer to the transaction. Today, only 18% of finance teams enforce policy at the point of transaction, while 40% review or flag spending after the fact.

You may be here if: AI is working for individual tasks, but your team still handles much of the surrounding workflow manually.

What to do next: Choose one workflow and build outward from the AI capability that’s already working. Connect the capabilities that can remove the most friction and assign someone to own the results.

Moving from one AI capability to a connected workflow

Logan Services offers an example of what that kind of end-to-end expense process can accomplish.

When a cardholder makes a purchase, they capture the receipt and add the appropriate business-unit information. That transaction data then syncs nightly to Sage Intacct, eliminating the need for finance to download files, add details and code expenses manually. Meanwhile, finance can adjust card limits in real time and restrict which tags each team sees.

By connecting those steps, Logan Services cut the time spent getting card data into Sage Intacct by 75% and eliminated expense reports for managers, executives and directors.

Connecting individual capabilities can reduce work across a process. As AI becomes part of more processes, the next challenge is connecting them without losing visibility or control.

Run: Connect AI across the finance function

Once AI is working across several finance processes, the next step is to connect more of the data and systems those processes depend on. That can open the door to more advanced applications, such as cash-flow forecasting, predictive spend controls and analytics that draw on information across the finance function. The State of Finance report identifies forecasting and predictive controls as potential areas for teams at this stage.

As AI gains access to more data and takes on more responsibility, its boundaries need to become more explicit. Define what data AI can access, what actions it can take, who has permission to use different capabilities and where human review is required. Permissions, audit trails and clear ownership help finance maintain visibility and give people defined points to review or intervene. 

The data also shows greater comfort with AI among teams further along in adoption. Comfort with AI making routine financial decisions rises from 17% of teams that haven’t started to 35% of teams experimenting or in early adoption and 51% of teams scaling AI. Over the same progression, the share naming trust as their top barrier falls from 42% to 25%. 

The goal isn’t to remove people from the process but to shift their attention toward the work and decisions that require judgment.

You may be here if: AI is already part of several everyday workflows and you’re ready to connect those workflows or expand into more advanced use cases.

What to do next: Identify where disconnected data or systems are limiting what AI can do, then establish the permissions, governance and human oversight needed to expand its role.

How PEX helps finance teams put AI to work

PEX gives finance teams ways to start with AI in everyday expense management, then expand how they use it as their needs and comfort level grow.

• Start with easy wins. PEX uses AI to match receipts and invoices with transactions and offers AI-powered GL code suggestions to categorize expenses. These capabilities can reduce the manual work involved in collecting documentation, coding transactions and preparing for reconciliation.
• Give AI more to do. The PEX agent can answer questions about your account and help admins take actions from a prompt. Live capabilities include creating spend rules and completing multiple account-management actions from a single request.
• Connect AI to more of your PEX data. For teams further along, the PEX MCP server can connect PEX with outside AI assistants such as Claude and Copilot. That gives admins a way to interact with their PEX account in natural language from the AI tools they already use.
• Add oversight as AI’s role expands. PEX maintains an AI audit history of account changes, giving admins visibility into actions such as changes to card status, permissions and approval policies. As AI moves from answering questions to taking actions, that record helps finance see what changed and maintain oversight.

Put AI to work where it counts

The measure of successful AI adoption isn’t simply how many use cases you’ve deployed. It’s whether they’re taking meaningful work off your team’s plate while giving you the visibility and control finance requires.

As AI becomes part of more everyday finance work, the opportunity is to spend less time on repetitive tasks and more time on the decisions that require your team’s expertise.

Download PEX’s State of Finance 2026 report to see how finance teams are using AI today and where they’re seeing the biggest gains.