How CFOs Are Using AI to Prepare Earnings Calls: The Financial Reporting Risks Nobody Wants to Explain

AI is helping CFOs prepare earnings calls, anticipate investor questions and explain financial results. I like the efficiency. I have a few concerns about what happens when the explanation sounds better than the underlying analysis.

Apparently, we’ve reached the point where the CFO can have an AI version of himself.

I suppose it was inevitable.

We gave AI the expense reports, the reconciliations, the forecasts and half the PowerPoint presentations. Eventually, somebody was going to hand it the executive’s personality.

According to an October 13 Bloomberg Law report, Expensify CFO Ryan Schaffer uses an internally developed tool called “Ryan Bot” to help draft his earnings-call remarks.

The bot has been trained on his previous public statements so it can imitate his communication style.

Schaffer still reviews and edits the material. Sometimes extensively.

That distinction matters.

He’s using AI to get past the blank page, which seems perfectly reasonable to me.

I’ve spent enough time around financial reporting to appreciate anything that eliminates the ceremonial staring contest between a finance professional and an empty document.

But the growing use of AI in CFO earnings call preparation raises a question that goes well beyond writing efficiency.

An earnings call is where management explains what happened to the business, why it happened and what investors should expect next.

Those explanations depend on financial judgment.

And financial judgment is where things get interesting.

Because I’ve seen perfectly accurate numbers accompanied by explanations that were complete nonsense.

Now we have technology capable of making the nonsense sound considerably more professional.

Progress, apparently. For the governance side, see forecast assumption ownership.

How CFOs Are Using AI to Prepare Earnings Calls

Let me establish something before somebody accuses me of wanting finance departments to return to paper ledgers and pneumatic tubes.

I use AI. I think it has legitimate applications throughout corporate finance and FP&A.

Drafting earnings-call remarks is one of them.

A CFO preparing for an investor presentation has an enormous amount of information to organize. Financial results, operational performance, guidance, market conditions, competitor disclosures and whatever unexpected development occurred approximately six minutes after the presentation was supposedly finalized.

If AI can organize that information into a workable first draft, wonderful.

I can think of better uses for a CFO’s time than rewriting the same sentence about operating margins fourteen times.

And there’s another use I find particularly valuable.

AI-powered earnings call preparation can help finance teams challenge the presentation before investors do.

Give an AI assistant the financial results, historical disclosures and an approved set of assumptions. Ask it to identify inconsistencies, generate difficult investor questions and compare management’s explanation against the underlying numbers.

That could improve preparation substantially.

The Bloomberg report describes finance leaders doing exactly this kind of work.

Latitude Financial Services CFO Guillaume Leger discussed using AI to anticipate analyst questions. Other executives described using the technology to review how competitors explain developments such as foreign exchange movements.

I see the value.

What interests me is what happens after the first draft arrives.

Because that’s where the finance work begins.

The Biggest Risk of AI in Financial Reporting Isn’t Always an Incorrect Number

This is one of the more frustrating realities of financial analysis.

You can reconcile every number in a report and still reach the wrong conclusion.

Imagine a company reports revenue growth of 12%.

The AI drafts a beautiful explanation:

“Revenue increased because of strong customer demand, successful pricing initiatives and continued market expansion.”

Very polished.

Management could probably read it aloud without spilling coffee.

But what actually happened?

Suppose the increase came primarily from an acquisition completed halfway through the prior year.

Organic revenue grew 2%.

Pricing contributed another percentage point, while customer volumes declined in the core business.

Suddenly, that cheerful explanation needs some work.

The reported 12% growth can be mathematically correct. The underlying performance tells a different story.

Now consider a margin improvement.

Operating margin increased from 14% to 17%.

AI might describe improved operating efficiency and disciplined cost management.

Maybe that’s true.

Or perhaps the company delayed hiring, reduced discretionary spending and benefited from a one-time supplier credit.

Those aren’t necessarily bad developments.

But they’re different from sustainable productivity improvements.

If management presents temporary benefits as structural improvement, investors may build expectations the business cannot meet.

That’s the distinction I care about when evaluating AI-generated financial reports and earnings commentary.

AI can describe a variance.

It needs verified evidence to explain the cause.

And a polished sentence is a remarkably effective way to conceal the difference.

Why AI Earnings Call Tools Need Reliable FP&A Data

I’m genuinely curious about what happens when an AI assistant encounters conflicting financial information.

Not specifically because of Expensify. Bloomberg’s reporting describes Schaffer reviewing and editing his bot’s drafts, which is exactly the kind of human involvement I’d want.

My question is about the broader direction of AI automation in FP&A and corporate financial reporting.

Suppose the approved earnings presentation reports customer retention of 91%.

The operational dashboard shows 89%.

The investor relations team has a previous-quarter definition that excludes a particular customer segment.

Meanwhile, somebody in FP&A has a reconciliation explaining why the numbers differ.

That reconciliation, naturally, is in a spreadsheet called Final_Final_Updated_v7.

Finance has its traditions.

Which number does the AI use?

More importantly, does it recognize that the definitions differ?

An AI system trained to imitate a CFO’s writing style doesn’t automatically know which financial definition management has approved.

That’s a separate problem.

And it’s one I would solve before worrying about whether the bot sounds sufficiently executive.

I’d want every material metric connected to an approved definition, a reporting period, a controlled data source and someone responsible for validating it.

I’d also want the system to recognize when information is missing.

Sometimes the most useful answer in finance is:

“We haven’t established that yet.”

It’s not an exciting sentence.

It is, however, considerably cheaper than explaining a misleading statement after the earnings call.

Why financial data governance matters before AI automation

Companies considering AI tools for financial reporting should examine the underlying data governance process before introducing automation.

Who owns the definition of revenue retention?

Which forecast is approved for external guidance?

How are non-GAAP adjustments reconciled?

Who verifies that the financial commentary uses the same definitions as the published results?

These are ordinary finance questions.

They become considerably more urgent when a system can turn inconsistent information into a convincing explanation in seconds.

Automating an unresolved definition doesn’t resolve it.

It just makes the disagreement harder to notice.

How AI Is Changing Investor Questions and Earnings Call Preparation

Here’s another development that deserves attention.

CFOs aren’t the only people bringing AI to earnings season.

Analysts and investors are using it too.

They can compare transcripts, examine historical guidance, identify inconsistencies and generate questions across enormous amounts of financial information.

According to Gartner research reported by CFO Dive earlier this year, 43% of surveyed CFOs said they were spending more time preparing for earnings calls. Another 34% reported an increase in unexpected investor questions.

That changes the preparation process.

An analyst who previously reviewed several quarters of financial statements manually can now use AI to examine a much longer history.

Imagine an investor asking:

“Last year, management attributed margin pressure to temporary implementation costs. Those costs were expected to decline by the second quarter. Why are they still affecting margins in the fourth quarter?”

That’s a reasonable question.

And potentially an uncomfortable one.

Especially if the company’s current explanation was generated without examining what management said previously.

I think this creates an interesting arms race.

Companies use AI to prepare better answers.

Investors use AI to formulate better questions.

Companies respond by investing in more sophisticated preparation tools.

Eventually, we may have two AI systems interrogating each other’s assumptions while the humans sit nearby wondering whether anyone ordered lunch.

But there’s a serious implication.

The quality of financial explanations will become easier to test against historical statements.

That’s good for transparency.

It’s also a reason finance teams need to pay closer attention to consistency.

If management describes a cost as temporary, someone should track when that cost is expected to disappear.

If a productivity initiative is supposed to improve margins, the forecast should reflect the expected benefit.

If the benefit doesn’t materialize, the explanation needs to change.

AI may make those inconsistencies easier to discover.

I consider that a feature.

Why AI Financial Reporting Automation Won’t Fix a Weak Finance Process

When I read stories about AI transforming executive reporting, my mind usually goes somewhere less glamorous.

The underlying finance process.

Before anyone drafts an earnings-call script, somebody has to close the books, consolidate the results, reconcile management reporting and explain the variances.

Then the team needs to determine which developments matter.

Was the revenue increase driven by price, volume, mix or acquisitions?

Did operating expenses decline because the business became more efficient, or because spending moved into the next quarter?

Did working capital improve because collections accelerated, or because the company delayed paying suppliers?

Those are different economic stories.

An AI assistant might be excellent at organizing the answers.

But if the finance team hasn’t resolved the questions, the assistant has a problem.

Or, more accurately, the company has a problem and the assistant now has an opportunity to describe it elegantly.

This is why I would put more attention into the quality of the financial analysis than the sophistication of the drafting tool.

I’d rather have an ordinary-looking report with defensible conclusions than a magnificent presentation supported by assumptions nobody can explain.

Finance departments have spent decades proving that PowerPoint can make almost anything look intentional.

We don’t need AI to improve that particular skill.

Practical Ways FP&A Teams Can Use AI for Earnings Call Preparation

If I were designing an AI-supported earnings preparation process, I’d start with the questions management needs to answer.

And I’d build backward from there.

1. Use AI for financial variance analysis

First, I’d want the system to compare the current results against the approved forecast, prior year and previous quarter.

Not simply calculate differences.

I’d want it to identify which variances are material enough to require explanation.

A $200,000 expense variance might matter enormously to one company and barely register at another.

Context matters.

2. Test management’s financial explanations

Then I’d want it to test management’s proposed explanations against the supporting information.

Let’s say the finance team attributes a gross margin decline to unfavorable product mix.

Fine.

Show me the mix.

Which products increased as a percentage of sales?

What happened to their contribution margins?

How much of the overall margin change does the mix shift actually explain?

If the explanation accounts for only half the variance, I want to know that.

3. Compare current earnings commentary with historical disclosures

I’d also want AI to compare the current narrative with previous public statements.

Did management promise savings from a restructuring?

Were those savings included in the forecast?

Have they materialized?

Is the company discussing the same metric using the same definition?

4. Use AI to simulate difficult investor questions

Finally, I’d ask the system to behave like a skeptical analyst.

Find the weakest explanation.

Identify the assumptions most likely to fail.

Ask the question management hopes nobody asks.

I find that last exercise particularly valuable.

It’s easy to prepare for questions you already know how to answer.

The difficult questions are usually the ones that expose something the company hasn’t fully examined.

AI could help uncover those gaps before an investor does.

That would be a meaningful improvement.

Calculating the ROI of AI in CFO Earnings Call Preparation

Because this is FP&A, I eventually want to see a calculation.

Suppose a company spends 160 combined staff hours preparing the narrative portions of its quarterly earnings materials.

That includes drafting, reviewing, reconciling explanations and revising presentations.

Assume the blended fully loaded labor cost is $125 per hour.

That’s $20,000 per quarter.

Across four quarters, $80,000.

Now suppose AI reduces that workload by 25%.

The arithmetic suggests 40 hours saved per quarter, or $20,000 annually in labor capacity.

That’s useful.

But I wouldn’t automatically call it $20,000 in savings.

If the same employees remain on payroll and spend the recovered time on other work, the company hasn’t reduced its expenses.

It has created capacity.

There may be real economic value in that capacity, particularly if the team uses it to improve forecasting, support decisions or avoid additional hiring.

But I would measure those benefits separately.

And I’d subtract the additional costs.

Software licenses.

Implementation.

Data preparation.

Security reviews.

Human verification.

Any additional controls needed because the system is handling sensitive financial information.

Imagine the AI tool saves 40 hours but requires 25 hours of additional review.

The net benefit is now 15 hours.

At our assumed labor rate, that’s $1,875 of quarterly capacity.

Still potentially worthwhile.

But the business case looks different.

I’d also avoid treating the cost of an incorrect earnings explanation as a neat little probability multiplied by a hypothetical stock-price impact.

That kind of precision can become fiction remarkably quickly.

I’d assess the exposure through control failures, disclosure corrections, remediation costs and the consequences of misleading investors.

Some risks deserve controls even when the expected dollar loss cannot be estimated reliably.

A CFO’s credibility would be one of them.

Can AI Replace CFO Judgment in Financial Reporting?

This is the part of the story I keep returning to.

Ryan Bot has learned how its human counterpart communicates.

That’s interesting technology.

But communication style and financial judgment are separate capabilities.

A CFO makes decisions using information that may never appear in a published financial statement.

They know which customer contract is under negotiation.

Which operational improvement is running behind schedule.

Which expense reduction is sustainable.

Which forecast assumption makes everyone in the room suddenly interested in their coffee.

A model trained on historical public remarks won’t necessarily have that context.

And even if a company gives an AI assistant access to internal information, access doesn’t guarantee sound interpretation.

The assistant still needs to distinguish facts from estimates, approved guidance from preliminary forecasts, and management expectations from established results.

Consider a company expecting $8 million in annual savings from an automation program.

The implementation is behind schedule.

Only $1 million has been realized so far.

The remaining savings depend on system integration, process redesign and staffing decisions that haven’t been finalized.

An AI-generated script might describe the company as making progress toward its $8 million savings target.

Technically, that could be accurate.

But I’d want a stronger explanation.

How much has been realized?

How much is contractually or operationally committed?

What remains dependent on future actions?

When should investors expect the savings to appear in operating results?

Those are the questions that turn a corporate statement into useful financial information.

A good CFO should be able to answer them.

So should the finance team supporting the CFO.

The bot can help prepare the explanation.

I wouldn’t ask it to decide whether the explanation is defensible.

AI Financial Reporting Risks: What Controls Should CFOs Put in Place?

Bloomberg also cited a Workiva survey in which one in four executives reported that an AI-generated error had reached investors or the board.

That deserves attention.

The figure comes from a survey, so I wouldn’t interpret it as the probability that any particular earnings-call script contains an error.

Still, it indicates that AI mistakes aren’t necessarily being caught before information reaches important audiences.

My response wouldn’t be to ban AI from financial communications.

I’d establish a review process appropriate to the risk.

For earnings materials, I’d want a controlled source for reported results, traceable support for significant claims and documented approval of forward-looking statements.

I’d separate three activities.

AI can draft language.

Finance validates the financial explanations.

The appropriate executives and disclosure reviewers approve the final communication.

Those responsibilities shouldn’t become interchangeable simply because the software can perform several tasks in the same window.

I’d also pay attention to version control.

A statement that was accurate on Monday may be outdated by Thursday if guidance changes or a material development occurs.

The AI needs current, authorized information.

And the review process needs to confirm that the final script matches the approved disclosures.

This sounds like ordinary financial control work because that’s exactly what it is.

We already understand the importance of reconciliations, approvals and evidence.

AI hasn’t made those principles obsolete.

It’s given us another reason to apply them carefully.

The Future of AI in FP&A Is Better Financial Analysis

For all the attention surrounding AI-generated executive remarks, I suspect the greater value may come from improving the analysis behind them.

Think about the amount of time finance teams spend preparing information that never makes it into an earnings presentation.

Investigating variances.

Reconciling operational and financial metrics.

Comparing forecasts.

Reviewing business-unit explanations.

Testing assumptions.

Trying to determine whether a favorable result reflects genuine improvement or simply a timing difference.

Those activities influence the quality of management’s decisions.

They also determine whether the CFO has something meaningful to say.

I’d be far more interested in an AI system that identifies an unexplained $4 million margin variance than one that produces six attractive paragraphs about operational excellence.

Especially if those paragraphs arrive before anyone has explained the variance.

A finance team that understands its numbers can produce a credible earnings narrative with relatively simple tools.

A finance team that doesn’t understand its numbers won’t solve the problem by purchasing a more sophisticated writing assistant.

It may simply become faster at publishing explanations that need correcting.

And speed is a questionable achievement when you’re headed in the wrong direction.

Should CFOs Use AI to Write Earnings Calls?

I like the direction this technology is taking.

I like the idea of reducing repetitive preparation work.

I like using AI to challenge assumptions, compare disclosures and identify difficult questions before investors ask them.

I particularly like the possibility of giving finance professionals more time to investigate the business instead of formatting the explanation.

But I wouldn’t measure success by how closely an AI assistant imitates the CFO.

I’d measure it by whether the company produces clearer, better-supported financial explanations.

Did the tool help identify an inconsistency?

Did it uncover a weak assumption?

Did it improve the quality of management’s preparation?

Did it reduce unnecessary work without weakening controls?

Those are outcomes I can evaluate.

The ability to reproduce an executive’s favorite phrases is entertaining.

It might even save time.

But investors aren’t buying the CFO’s writing style.

They’re evaluating management’s understanding of the business and its ability to make credible decisions.

So by all means, let Ryan Bot prepare the first draft.

Let AI examine the numbers and generate the difficult questions.

Let it help the finance team work through a complicated quarter.

Just make sure somebody can explain the results without asking the bot what they mean.

Because when an investor asks why margins improved, “the AI wrote that part” is probably not the answer anyone wants to hear.

And if we ever reach the point where the CFO’s bot is answering questions from an analyst’s bot, I’d still like one actual finance professional in the room.

Preferably someone who knows where the numbers came from.

And has access to the spreadsheet.