Ask most auditors and they will tell you: AI already lives in their spreadsheet. Copilot ships inside Microsoft 365, ChatGPT and Claude are a tab away, and a growing number of Excel AI agents chain several steps together from one instruction.

But is the same tool you'd ask to draft an email or generate a picture the right one for a workpaper someone else has to sign off on? And how confident are you uploading confidential financials to get that answer?

This piece looks at where AI in Excel actually helps with audit work, where it hits a wall, and which tool fits the task in front of you.

How is AI transforming the audit process?

AI is useful for the everyday parts of audit work: drafting, summarizing, a fast answer on an unfamiliar term. And it is being adopted faster than most firms' governance has caught up with. According to Wolters Kluwer’s report in 2026, AI adoption is now a strategic leadership responsibility as 87% of high-growth firms plan to increase their investment in AI-enablement tech. Half of them already use advanced AI on a daily basis, and audit teams are no exception.

Why audit teams are adopting AI tools

The potential benefit is real. In a study published in the Journal of Accounting Research, accounting teams already using AI daily can reallocate about 8.5% of their time from routine data entry to higher-value work and close their books roughly a week faster. When working on complex inventory-valuation tasks, auditors complete the task faster and produced more formally correct documentation, with the largest gains among the people who used it most.

For any team stretched thin in busy season, that is a real reason to consider where AI could reduce manual work. A junior can get a first answer on an unfamiliar account without waiting for a manager. A senior can sanity-check a number before it goes into review. AI saves real time and effort, which is exactly why adoption keeps climbing.

How does ChatGPT work on audit tasks?

How can auditors use AI to automate tasks in Excel? Paste a table or upload a file, then ask it to summarize, extract a value, or write a formula. Copilot works the same way inside Microsoft 365. Claude does it through its own chat or a connected app. All three read whatever you give them and answer with one general model built to handle almost any task, not just financial ones.

Point one at an invoice or a contract and it does a reasonable job of AI-based data extraction: a date, an amount, a counterparty name, roughly where it found it. A newer wave of Excel AI agents chains a few of these steps together, so one instruction replaces five.

Adoption is climbing to match. Reported by AFM, the share of audit firms in the Netherlands using data analysis tools and cloud solutions for statutory audits rose from 74% in 2022 to 91% in 2025, leaving them more time for more complex audit work, which can directly and indirectly improve the quality of their audits. However, adoption of advanced tools only went from 4% to 9% within the same period. Most firms have adopted the easy layer of AI. Few have moved past it, and that’s where general-purpose tool starts to show its limits.

Where Excel AI falls short

For management in audit firms, the major challenge in using AI for audit lies in the gap between how fast firms adopt general AI tools and how slowly the governance around them catches up. Three specific pain points show up once an auditor tries to use general Excel AI for real audit work.

AI tools not built for audit tasks: lack of audit context, traceability, and explainability.

A general AI tool can describe what it finds in a document, but it can't link a cell back to the exact spot in the source, and it can't show why it landed on an answer. That's the "black box" problem often seen in AI tools. A general model can tell you the figure is on page 4, but cannot bring you to page 4 when you review, and cannot show its work well enough for a reviewer to challenge it. A workpaper is only finished when a reviewer can follow it back to the evidence, and the reasoning, without asking anyone.

Falling short for governance and compliance standards.

Speedy output using AI tools does not guarantee consistent, uniform results across a team. The FRC's review of the six largest UK audit firms found no formal monitoring in place to quantify the audit quality impact of the automated tools they had already deployed. Firms have bought the tools. Few can say, with evidence, whether they are working as intended, and a general AI assistant's terms of service were written for a mass market, not for the confidentiality expectations a firm signs up to on an audit engagement.

Limited customization to firms’ requirements.

Additionally, a general AI tool has no concept of your firm's template. Every answer starts from a blank chat, only controlled by your prompts and formatted however the model decides that day, so a preparer still copies, reformats, and pastes the result into the standard structure by hand, for every file, every time.

Flex Agent: audit automation built for the workbook

One agent, two functions: answer and execute

Flex Agent is the AI agent inside RobotX Flex, the audit workspace that runs inside Excel. Built by auditors who have worked on countless workpapers themselves, it's purpose-built to handle audit tasks: procedure, evidence, result, and conclusion, not just a data dump. Upload your documents, ask a question, or tell it to run Flex's extraction and matching. Whatever it produces stays linked to the source document, ready for review.

Ask it a question and it answers from what is in your engagement: an account variance, whether an invoice matches its purchase order, what is still missing on a sheet. Give it an instruction and it routes the task to the right Flex tool, the same one you'd otherwise select and run yourself, across a full population in one pass.

Why Flex Agent fits audit work where general AI doesn't

Flex Agent is built to help auditors produce work a reviewer can follow without redoing the task. General AI tools are useful for simple, self-contained work such as drafting, summarizing, and quick lookups, but they are not built around audit context or the checks required in a working paper.

With Flex Agent, an auditor can run a specific task, like extracting and matching a batch of invoices or testing journal entries against a risk filter, by asking in plain language instead of doing each step manually. Flex Agent works with the information already in the workbook and the documents behind it, so the auditor can give it the task and review the results, already linked to source, before they go into the file.

A general AI tool can't do the same thing as reliably. It has no library of audit-specific actions to route to, so it reasons through the instruction itself, the same way it would answer any other prompt, rather than picking the specific check a working paper calls for. That difference comes down to what each tool was built around: audit context, an overview of the documents on your file, a traceable result, and a review step, the things a general assistant was not built around.

It's also what makes Flex Agent explainable AI rather than a black box. Because it routes execution through a defined process instead of reasoning freely, what lands in your file traces back to a specific cell in the source document, using the same procedure every time. Traceability does not remove the need for professional judgment and human review, which is why Flex Agent still lands every result as something a preparer reviews rather than something taken on faith.

Choose the purpose-built AI agent for audit automation

Most auditors are skeptical of AI for a specific reason: these tools cannot show the evidence behind an answer or provide such evidence to the reviewer before sign-off. That's one of the gaps Flex Agent is built to answer.

Flex Agent is purpose-built to support extraction, matching, and review inside the workbook auditors already use. Reviewers always stay in the loop: every result lands as an AI-review cell, and a preparer has to approve it before it counts. That’s the same standard regulators are asking audit software to meet, built in from the start, not added on afterward.

We run the checks. You keep the judgment. That is the every RobotX product runs on.

See what Flex Agent can do →