Where AI fits in budget vs actual

Today. Export the P&L from QuickBooks or NetSuite, paste it into the budget workbook, and fix the formulas that broke.

With AI. Pulls actuals by account, department and entity once the period closes, and checks the totals against the ledger. In: the closed trial balance in your ledger. Out: actuals in the model that tie to the ledger.

Who decides. AI-assisted: Actuals load themselves, an analyst checks the tie-out. AI-run: Actuals load when the period closes, a failed tie-out goes to a person.

Control. Actuals load only from a closed period, and the totals tie to the ledger's trial balance.

Tools that do this. Datarails (score 72): actuals land on a schedule and flow into your existing Excel reports with a Refresh button. Cube (score 76): analysts fetch live actuals into the ranges of their existing Excel or Google Sheets workbooks. Fathom (score 71): data flows one way from the ledger, through native Xero and QuickBooks connectors.

Today. Keep a mapping tab that matches ledger accounts to budget lines, and fix it by hand when someone adds an account.

With AI. Maps each account, department and entity to its budget line, and flags new or unmapped accounts. In: actuals by GL account and the budget's line structure. Out: actuals and budget on the same lines, with unmapped accounts listed.

Who decides. AI-assisted: AI suggests mappings, an analyst approves each change. AI-run: Known accounts map themselves, new accounts go to a person.

Control. Every mapping change is logged, and an unmapped account holds the report until a person maps it.

Tools that do this. Datarails (score 72): consolidates entities that run on different charts of accounts and currencies. Prophix (score 76): an Architect Agent maps source data into the planning model. CFO.ai (score 70): its assistant, Ari, builds the model from your connected QuickBooks Online or Xero data.

Today. Build the variance columns in Excel, department by department, and hope nobody sorted the rows.

With AI. Calculates dollar and percent variances by line, department and entity, and links each one to the transactions behind it. In: mapped actuals, the budget and the latest forecast. Out: a budget vs actual view where every number drills to its transactions.

Who decides. AI-assisted: AI builds the view, an analyst checks it. AI-run: The view builds itself when actuals land.

Control. Every variance drills to the ledger transactions behind it.

Tools that do this. Cube (score 76): pulls budget versus actual by department and drills to the NetSuite transactions behind a variance. Vena (score 78): pulls plan versus actual report data and drills into the transaction rows behind a number. Datarails (score 72): runs grouped totals such as actuals versus budget by department, under each user's own permissions. Agicap (score 73): shows cash variance against plan by entity and category.

Today. Scan hundreds of lines for anything that looks big, and miss the ones that are small but wrong.

With AI. Checks each variance against dollar and percent thresholds, separates timing from real overspend, and flags lines drifting off plan. In: every variance and your thresholds by line. Out: a short list of variances worth a question, each with the reason it was flagged.

Who decides. AI-assisted: AI flags lines, an analyst confirms the list. AI-run: Lines inside thresholds pass, flagged lines go to a person.

Control. Finance sets the thresholds, and every change to them is logged, so nobody quietly raises one to hide a miss.

Tools that do this. Planful (score 76): its Planner Assistant projects each account from closed actuals and flags accounts drifting off plan. Fathom (score 71): for accounting firms, its Insights Dashboard flags clients that miss thresholds. Cube (score 76): its Smart Variance feature handles variance explanation.

Today. Email each budget owner their lines, chase the replies, and paste the answers into the deck.

With AI. Drafts a first explanation for each flagged line from its transactions, sends it to the owner to confirm or correct, and chases the rest. In: the flagged variances and the transactions behind them. Out: owner-confirmed commentary linked to the numbers.

Who decides. AI-assisted: AI drafts the commentary, each owner confirms it. AI-run: AI drafts and chases, owners confirm their own lines.

Control. Every comment links to the transactions it explains, and its owner confirms it before it goes in the pack.

Tools that do this. Fathom (score 71): its Commentary Writer drafts report commentary and links each figure to its source. Prophix (score 76): its Reporting Agent builds tables and variance commentary. Cube (score 76): budget owners can ask its assistant, Charlie, questions in Slack or Teams instead of opening a workbook.

Today. Roll the forecast forward a month, retype the assumptions, and save a new copy called forecast-FINAL.

With AI. Projects each line forward from closed actuals, copies the forecast into base and downside scenarios, and shows what changed and why. In: closed actuals, owner commentary and your drivers. Out: a draft forecast and scenarios, ready for a person to approve.

Who decides. AI-assisted: AI drafts the forecast, finance adjusts and approves it. AI-run: AI drafts it, the CFO or controller approves it, always.

Control. The approved forecast is version-locked, and AI writes only to draft scenarios.

Tools that do this. Cube (score 76): AI copies the budget into a new scenario, while Budget, Forecast and Actuals stay protected unless a person opens them. Planful (score 76): each forecast can start from a baseline projected from closed actuals, with the drivers behind each projection shown. CFO.ai (score 70): snapshots pin a budget or board baseline, though a Snapshot can still be edited. Workday Adaptive Planning (score 78): rolling forecasts pull actuals in on a schedule, and analysts test changes in personal what-if scenarios first.

Today. Copy charts into PowerPoint, retype the numbers, and check every slide against the model the night before.

With AI. Refreshes every chart and table from the approved numbers, drafts the summary, and checks the pack against the model. In: the approved actuals, forecast and commentary. Out: a board pack that ties to the model, ready for the CFO to send.

Who decides. AI-assisted: AI refreshes the deck, finance reviews every page. AI-run: AI refreshes it, the CFO reviews and sends it, always.

Control. The pack is built only from the approved forecast version, and a person sends it.

Tools that do this. Cube (score 76): its Cube Decks feature puts refreshable visuals into PowerPoint and Google Slides. Datarails (score 72): its Storyboards feature is built for board decks. CFO.ai (score 70): one view-only link opens the board summary, P&L and headcount pages in a chosen scenario. Fathom (score 71): schedules delivery of one branded report template with commentary, KPIs and charts.

“How much of my travel budget is left for the year?” Reads the department's budget and closed actuals and answers from the numbers the owner is allowed to see. AI-run sends the answer. AI-assisted drafts it for the analyst.

“What does the forecast look like if the two Q4 hires slip to January?” Copies the forecast into a new scenario, moves the start dates and shows the change in cost and cash. A person checks any scenario before it goes to leadership.

“Can you move $25k of my September spend into October so I come in on budget?” Never changes closed actuals. Flags the request and routes it to the controller. The controller, always.

Tools that do this. Cube (score 76): its assistant, Charlie, answers questions in Slack and Teams. Datarails (score 72): its Claude connector answers actuals versus budget questions, and it can only read. Planful (score 76): its Analyst Assistant answers variance questions in plain language. Pigment (score 80): through its MCP server, Claude answers why a line moved from live plan numbers the user can already see.