Can AI financial analysis explain why your margins changed?

Sales are up 11%. Gross margin is down five and a half points. When those two numbers show up together, someone always asks the same thing: what happened?

AI-powered financial analysis can answer most of that question, and fast. It splits the change into price, volume, mix, and cost, then points to the clients or product lines behind it. What AI can't tell you is why a driver moved, or whether the data underneath agrees with it. The risk AI brings to the table is an incorrect or partially correct explanation built on bad data that sounds super confident.

This article is for owners, finance operators, and finance leaders who want AI's speed without taking its word for everything. You'll learn how a margin bridge works, what data each driver needs, and five checks to run before you act on the insights. Then we'll walk through two worked examples.

What can AI-powered financial analysis explain?

AI is good at telling us "the what and the where". The "why" still takes human judgment.

Every margin analysis answers three questions. What changed is the calculation used to show how much each driver moved gross profit. Where it changed is a drill-down into clients, products, or service lines. Why it changed is context, like a discount the sales team committed to or a supplier's new tariff surcharge. AI models handle the first two really well. The third isn't something AI usually finds readily available unless you inform it in detail.

AI improves financial analysis by combining structured data, like ledgers and invoices, with unstructured financial text, like contracts, which natural language processing can analyze. AI models monitor transactions in real time to flag anomalies, the core of fraud detection, and machine learning spots patterns linked to credit risk. Forecasting models analyze historical sales and economic indicators, and some tools claim they can predict future financial outcomes with remarkable accuracy. Real-time cash flow forecasting helps with liquidity management. Investors and analysts even use AI to summarize companies' earnings calls and catch shifts in management tone.

Most companies are asking themselves, "What happened to our margin, and is the AI explanation right?" (For the bigger picture, see what an AI finance assistant does.)

Being correct is critical. In Thomson Reuters' 2026 Future of Professionals report, 90% of professionals said they "need outputs they can explain and defend." Accuracy matters more than confidence, and AI-generated insights you can't trace back to the data don't clear that bar.

How does a margin bridge work?

Start by picking the margin. Gross margin is revenue minus the direct cost of delivering it, divided by revenue.

In SPI Research's 2026 benchmark, professional services project margins rose to 37.7%, while EBITDA held at 9.9%, "still well below the five-year average of 13.8%."

A margin bridge is a simple model. It tracks from last period's gross profit to this period's, one driver at a time: volume, then mix, then price, then each cost. Each step starts where the last one ended, so the steps add up to the total. At this stage it's just math, and it's the mechanism behind most automated variance analysis. Larger companies pay financial analysts to build these manually, but AI builds them in seconds, which speeds up financial reporting cycles.

Two rules keep a bridge honest: it has to match the books (check 2 below), and the order has to be stated, because a different order can shift a few thousand dollars the wrong way. If an AI tool hands you a leftover labeled "interaction effect," check the math. Some published examples use that label to cover an arithmetic slip, which can lead you to the wrong fix.

Dollars and points tell different stories

Selling more at last year's prices and mix adds gross profit dollars but leaves the margin percentage exactly where it was. In both examples below, volume adds real money and zero points.

So when the percentage falls, the cause is price, mix, or cost. Mix is sneaky because every client can hold its own margin while the total continues to fall, simply because of a shift toward lower-margin work. And always work out a period's margin from total dollars, not an average of monthly percentages.

What data does each driver need?

Good financial analysis comes from evidence, and each driver leaves it in a specific place:

• Product sales: invoice lines (units, prices, discounts, and credit notes), supplier bills and item costs, carrier invoices, and inventory adjustments
• Service work: agreements, time entries (hours per client, project hours, and write-offs), payroll, and vendor bills for tools and licenses
• Both: the general ledger and its supporting documentation, for one-offs and for how costs are classified

Many firms keep that data in separate systems: a professional services automation (PSA) system for agreements and time, accounting software for the books, and payroll in a third system. Manually combining it all can take finance teams weeks. AI handles it without the copy-paste, while real-time analytics keep it current. That's where AI is most valuable, and where bad data leads to bad insights. The challenge is that each system has its own data model, so the same client or cost can look different in each one. As Deltek's summary of the SPI research puts it, "Firms running disconnected point solutions continue to report weaker visibility, more manual workarounds."

That's the problem Alti, the financial intelligence inside Alternative Payments, is built around. In Analytics, Alti's overview puts profitability from an accounting software like QuickBooks Online next to MRR and operational efficiency from your PSA (Professional Services Automation), the two systems where a service business's margin evidence usually lives. Alti reviews money in, money out, and answers questions using your specific software data as its source without making assumptions like common LLMs.

Five checks before you trust an AI explanation

AI-powered tools can produce a clean-looking bridge from messy financial data. The risk usually isn't bad math. It's bad inputs. Run these checks before relying on the analysis or reporting it to leadership.

1. Compare like with like

Best practice is to compare the same quarter last year or a rolling twelve months of data, with the same definitions. Otherwise, you risk having AI read seasonal swings as trends.

2. Make the bridge tie to the books

This is the most critical check. The drivers should add up to the total, and the total should match the reported P&L. Give any gap its own row, labeled unexplained, until someone in finance explains it. Then look at the individual values, not just the net. In the service example below, a $3,000 gap hides $27,000 of movement.

3. Don't let timing skew financial reporting

Put revenue and its costs in the same period before you compare. Supplier bills entered weeks late break that rule, and so do projects finished in one month and invoiced in another: one period looks worse, the next looks better, and any forecast built on either one is off. Both examples below have a timing row.

4. Pull one-offs out of the trend

Write-downs, write-offs, true-ups, and bonuses are real costs, but they don't repeat. Give them their own rows, with a comment saying what happened, so nobody builds a pricing strategy or a forecast on them.

5. Check that the data is consistent and complete

Each cost should sit in the same function in both periods, whether that's cost of sales or overhead. Every bill should be entered, every hour logged, and cost rates should reflect current pay. If you use Analytics, the Client hours report asks, "Which clients' recorded hours dropped compared with their usual?" A drop can mean less work or work nobody logged. AI can flag it.

Finally, decide up front which drivers deserve a why, say anything over half a point or $10,000. That keeps the focus on the rows that matter.

Two margin bridges, side by side

Both examples below are built from a step-by-step model so the values in every column add up. Each compares Q3 with Q3 last year. Rows above the subtotal are what an AI tool would compute from operations data, and rows below it are what the checks found in the ledger.

A product-sale business

A distributor sells equipment, parts, and consumables. Last year's Q3 brought in $1,000,000 of revenue and $320,000 of gross profit, a 32.0% margin. This year, equipment units rose 15% on a 2% promotional discount, parts rose 12.5% with a 4% price increase, and consumables were flat. Tariff surcharges pushed equipment costs up 5%, and fuel surcharges added 15% to freight.

Those two costs aren't hypothetical. ISM's Steve Miller said in October, "Tariffs and fuel cost impacts were the most cited issues impacting respondents' supply chains."

Volume added $36,800 without moving the margin. The damage was tariffs on one line: equipment sold 15% more units, yet its gross profit fell $21,550 while parts gained $17,650. A third of the reported drop (1.8 of 5.5 points) came from a write-down and late bills that won't repeat. An AI bridge built from the ledger alone would have folded them into the trend. Pull them out, and the outlook changes: the underlying trend is 3.7 points, which is still a problem, just a smaller one. It's also the number your forecasts and plan should start from.

A service business

This one could be an MSP, an agency, or an accounting practice: 40 clients on monthly agreements at $3,000 a month, plus project work at $150 an hour. Last year's Q3 brought in $480,000 of revenue and $240,000 of gross profit, a 50.0% margin. The team logged 70% of its paid hours to clients at $63 an hour in pay and benefits, so an hour of client work cost $90.

This year brought four new clients and 10% more project hours. The new clients pay $2,400 a month and need 13 hours, against last year's average of 10. Existing clients took a 3% renewal increase but need 10.5 hours. A fixed-fee project ran weeks over, and 5% of project hours were written off. Pay rose 3%, close to the 3.1% private-industry wage growth the Bureau of Labor Statistics reported for the year to June, and utilization slipped to 68%.

Pay rose 3%, and utilization slipped two points, so an hour of client work went from $90 to $95.43, up 6%. That's the biggest driver in this scenario, and it's an efficiency problem as much as a pay problem. Growth hid some of it, too: the four new clients run at a 26% margin, against 53% for last year's clients.

On paper, and in any AI summary of the ledger alone, gross profit looks almost flat, because a $15,000 milestone invoiced late and a $12,000 salary reclassification pull in opposite directions. In practice, operations lost $3,400.

What no amount of analysis can tell you is whether the equipment discount strategy won customers who'll come back for consumables or just cost pricing power, or whether the new clients' extra hours will fade after onboarding. That's where people come in.

If your data looks like either table, bring last quarter's numbers to a demo. We'll show you which rows of a service bridge Alti can fill from your PSA and accounting software data, and which ones stay with you.

What AI does well, and where human judgment comes in

AI handles the heavy lifting that used to fall to analysts. It does the margin bridge math, checks every client and product line without a manual search through invoices, flags the unexplained gap, turns the bridge into plain-language insights, and reruns forecasts and scenario analysis whenever an assumption changes. AI agents can repeat the job every month and report any driver that crosses your threshold. That's intelligent automation at its most useful.

People should focus on and make decisions on the analysis and work that carries consequences, like the definitions, the threshold, whether a one-off really is one, and what to do next. In the utilization row, a two-point drop can be a new hire ramping up, which is an investment, or hours nobody logged, which is leakage.

Once you know the why, the fixes are familiar: pricing strategy, cost control, and mix (see these three profitability levers) or the staffing model (see this approach to resource allocation). If you want outside help, here's how AI CFOs, virtual CFOs, and finance assistants compare.

As AI adoption grows, keep AI governance simple

Human oversight is always important when it comes to your business finances, but good AI governance doesn't have to add complexity. Someone with enough data literacy to read the bridge, and approval before an agent changes anything, is sufficient. Thomson Reuters CEO Steve Hasker put it simply: "'almost right' isn't good enough." With Alti, any action that changes your account needs an on-screen confirmation from you first, and each confirmed action is logged with who approved it. Our guide to AI agents in finance covers how agents should handle approvals.

Test it on a period you already understand

Before relying on any AI tool, run the quickest accuracy test there is: hand it a period your finance team already knows well. If the analysis doesn't tell the story you know, tie it to your P&L, and highlight what it can't explain, it isn't ready.

Where Alti fits into your business

Alti is the financial intelligence inside Alternative Payments, an AI-powered financial operations platform for service businesses that combines accounts receivable, accounts payable, and financial reporting and analytics. For finance teams on ConnectWise Manage, HaloPSA, or Autotask with QuickBooks Online Analytics covers the "where" for most of the service bridge.

Alti's Client Trends shows each client's profit margin, using labor costs from PSA time entries and service costs from agreements and invoice lines, so changes in price, hours, and service costs show up client by client. Utilization adds team and individual utilization, plus Labor Cost Impact, "a relative signal of how labor affects client profitability."

The Client profitability report asks, "Which clients' margins moved most this month, and which are below my average?" Each report is sent to your inbox with headline numbers, what changed, charts, and downloadable tables. Describe the dashboard you want in a request, and Alti proposes the KPIs first. Nothing is built until you approve. With the Benchmarking add-on, Analytics also shows how your numbers compare with industry peers.

With Analytics, recurring agreements, projects, equipment sales, and T&M views and reports are in the overview dashboard.

AI for the math, people for the why

AI-powered financial analysis can tell you which drivers moved your margin almost as fast as you can ask. However, AI-generated insights are only as good as the checks behind them, and the why is still a conversation with the people on your team. Put the two together, and you get faster decisions you can stand behind.

Want to see it on your own data? Bring last quarter's numbers when you request a demo of Alti.

Common questions

What is AI-powered financial analysis?

It's software that uses AI to read financial and operational data, run the calculations, spot patterns and outliers, and explain the results in plain language. Common uses include variance analysis, financial reporting, forecasts, scenario planning, and fraud detection. Its accuracy depends on the data and definitions behind it.

Can AI explain why profit margins changed?

Partly. AI can split a margin change into drivers like price, volume, mix, and cost, and show where each one happened. AI can't explain why a driver moved unless the reason is in the data, such as a discount note, a contract change, or a comment on the deal. In practice, that part needs the people involved.

How does Alti help explain margin changes?

For service businesses on ConnectWise Manage, HaloPSA, or Autotask with QuickBooks Online, Analytics shows per-client margins, utilization, and a monthly report of which clients' margins moved most. Alti drafts dashboards around your questions. It shows where margins moved, and the reasons stay with your team.