AI Tax Preparation for Form 990 Nonprofits

Form 990 exists because tax exemption is a privilege granted in exchange for transparency. An organization that pays no federal income tax on its exempt activities owes the public a detailed account of what it did with the money instead. That account runs to twelve parts on the core form before a single schedule is attached: a summary page, a statement of program service accomplishments, a governance and management section, compensation detail, revenue and functional expense statements, a balance sheet, and reconciliations. The return is not filed in confidence. The moment it is accepted, most of it becomes a public document, indexed and searchable, sitting alongside every prior year and every peer organization.

That audience is wider and more attentive than most 1040 or 1120 filers ever face. A foundation deciding whether to fund the organization reads the 990 to size up program efficiency and governance. A charity watchdog scores it. A journalist mines it for executive pay and related-party deals. A state attorney general's charity bureau cross-references it against the state registration. A prospective board member reads it before saying yes. Preparers who have worked only on private returns are sometimes surprised at how forensically the 990 gets read — and at how a technically correct but carelessly worded answer can create a reputational problem for a client that never rises to an IRS issue at all.

Because of that, preparing a 990 well is as much about clarity and consistency as it is about arithmetic. The numbers have to tie — Part VIII to the audited statements, Part IX to Part VIII, the balance sheet to the ledger — but they also have to tell a coherent story year over year. A program that was 78% of expenses last year and 61% this year invites a question, and the return should answer it before it is asked. UpTax is designed with that reader in mind: it doesn't just fill lines, it flags the places where a swing, an outlier, or a blank will draw an eye, so your preparer can decide how to handle it rather than discovering it after the return is public.

If there is one task that defines 990 preparation, it is functional expense allocation. Part IX takes the organization's total expenses and demands they be split across three columns — program services, management and general, and fundraising — line by line, account by account. Some expenses assign directly: a grant paid out is program, an audit fee is management, the cost of a gala is fundraising. But most real expenses are shared. Rent, utilities, the executive director's salary, the accounting software, the insurance — these serve all three functions and have to be spread across them on a reasonable, documented basis. Get the basis wrong and the organization's program-efficiency ratio, the number donors and watchdogs fixate on, comes out wrong with it.

The honest difficulty is that there is no single correct allocation. There are reasonable methods — allocating by headcount, by square footage occupied, by a documented time study, by relative direct costs — and the preparer's job is to pick a method that fits each pool, apply it consistently, and be able to explain it if asked. Done by hand, this is hours of spreadsheet work that has to be rebuilt or at least re-checked every year, and the documentation of why each account was split the way it was often lives only in the preparer's head. When that preparer leaves, so does the reasoning, and next year's engagement starts from a weaker footing.

UpTax approaches allocation as a method to be proposed and reviewed, not a black box. It reads the ledger, groups expenses into logical pools, and proposes an allocation basis for each — carrying forward the prior year's approach where the facts are stable and flagging where they've changed. Crucially, it writes down the basis for every allocated line: this salary spread 70/20/10 by a time study, this occupancy cost spread by square footage, this shared IT cost spread by headcount. Your reviewer sees the whole method laid out, adjusts any pool where professional judgment differs, and the documentation persists into next year's file instead of evaporating. The AI does the mechanical spreading and the record-keeping; the firm owns the judgment about what is reasonable.

Part VI of the core form asks about governance — not because the IRS mandates a particular board structure, but because it has learned that weak governance correlates with the problems it cares about. It asks whether the organization has a conflict-of-interest policy, a whistleblower policy, a document-retention policy; whether the board reviewed the 990 before filing; how many voting members are independent; whether minutes are contemporaneously kept. None of these answers changes a dollar of tax. All of them signal to a reader how seriously the organization takes its stewardship, and a string of "No" answers is a quiet invitation to look closer. Preparers who treat Part VI as a checkbox exercise do their clients a disservice; the section rewards a client conversation and careful, accurate answers.

Compensation is where the public reads most closely, and where the mechanics are most unforgiving. Part VII lists officers, directors, trustees, key employees, and the five highest-compensated employees over the threshold, with their reportable and other compensation. Cross that threshold or hit certain roles and Schedule J opens up, demanding a breakdown into base, bonus and incentive, other reportable, deferred, and nontaxable benefits, plus pointed questions about first-class travel, companion travel, tax gross-ups, and discretionary spending. The reporting convention compounds the difficulty: compensation is generally reported on a calendar-year basis from the W-2 and 1099, even when the organization files on a fiscal year, so the figures on the return deliberately don't match the fiscal-year books. Reconciling all of it by hand — Part VII to Schedule J, both to the payroll records, across a calendar/fiscal mismatch — is slow and error-prone.

UpTax pulls the compensation picture together and reconciles it before your preparer touches it. It reads the W-2 and 1099 data, applies the calendar-year convention against the fiscal-year return, populates Part VII and Schedule J consistently, and surfaces the mismatches that usually surface only in review — a key employee missing from one schedule, a deferred-comp figure that doesn't tie, a benefit that triggers a Schedule J question the preliminary answers missed. For Part VI, it maps the organization's actual policies and board data to the governance questions and drafts the Schedule O explanations they require. Your reviewer confirms the policy answers with the client, edits the compensation narrative, and signs — starting from a reconciled draft rather than a pile of source documents.

The core form is only the beginning. The 990's real complexity lives in its schedules, sixteen of them lettered A through R, each triggered by specific facts in the organization's year. The danger is not the schedules an experienced preparer expects — Schedule A for public charity status is nearly universal, Schedule O is required for everyone — but the ones that hinge on a single transaction buried in the ledger. One grant to one organization over $5,000 pulls in Schedule I. One foreign wire or overseas program triggers Schedule F. A board member whose company the organization bought services from opens Schedule L. A loan to an officer, an endowment, a conservation easement, a donor-advised fund, a fundraising event with more than $15,000 of gross receipts — each quietly obligates its own schedule, and a missing one is precisely the gap a trained examiner or a diligent grant officer spots.

Schedule A deserves its own note because it carries existential weight. It establishes and defends public charity status through the public support test — demonstrating that the organization draws support from a broad base rather than a handful of sources, measured over a rolling five-year window under either the 33⅓% mechanical test or the 10% facts-and-circumstances test. Fall below the line and the organization risks being reclassified as a private foundation, with a heavier tax and disclosure regime attached. The computation is unforgiving, it depends on correctly characterizing every revenue source across five years, and it is not the kind of thing to eyeball in the last hour of an engagement.

UpTax treats schedule determination as a reasoning step, not an afterthought. It reads the transactional detail — the grants list, the contributor data, the related-entity structure, the foreign activity, the fundraising events, the interested-person transactions — and produces a triggered-schedule map: here are the schedules the facts require, here is the specific fact that triggers each, and here are the ones we evaluated and set aside, with the reason. Schedule B contributor thresholds get applied with the public-inspection redaction handled so donor identities don't leak into the public copy. Schedule A's support test is computed across the full five-year window from the characterized revenue. Your preparer reviews the map, agrees or overrides, and knows that nothing quietly fell through — which is a materially different starting point than a blank schedule list and a hope that memory holds.

The engagement starts with whatever the client actually has. There is no prescribed template to force the books into first. Upload the trial balance and the general ledger export, add the audited or reviewed financial statements, the prior year's return, the grant schedules, the payroll summary, and the board minutes or policy documents, and UpTax reads them as they are. It recognizes fund-accounting structures, net-asset classifications, and grant tracking, and it builds the crosswalk from the client's chart of accounts to the 990's revenue types and functional expense lines — the crosswalk that, done manually, gets reconstructed from memory every filing season.

From there it drafts. It populates Part VIII revenue and Part IX functional expenses, proposes and documents the functional allocation, ties the balance sheet to the ledger, and reconciles the return to the audited statements through Schedule D where the two diverge. It runs the schedule-trigger analysis over the transactional data and assembles every schedule the facts require, computing Schedule A's support test and applying Schedule B's contributor rules. It reconciles Part VII and Schedule J compensation across the calendar/fiscal convention. And it drafts Schedule O — the program accomplishments, the governance answers, the explanations behind every 'Yes' that needs one — in clean, public-facing prose. What comes out is a complete, internally consistent, disclosure-ready draft, with the AI's reasoning visible at every allocation and every triggered schedule.

Then the return leaves the AI and goes to your firm. A CPA or EA reviews the draft the way they would review a competent junior's work: checking the allocation method, confirming the governance answers with the client, testing the support-test result, editing the Schedule O voice, and applying the professional judgment that the engagement letter — and the signature block — require. UpTax never files. It produces a review-ready draft and hands it over. Your firm reviews it, corrects it, signs it as preparer, and transmits it. The tool compresses the hours between raw documents and a reviewable return; it does not, and is not meant to, replace the licensed professional who takes responsibility for what gets filed.

A draft is only useful if it is trustworthy, and on a public return trustworthiness means more than "the math adds up." It means every figure ties to a source you can point to, every allocation carries a documented basis, and every schedule that appears has a stated reason for appearing. UpTax is built to be auditable in exactly that way. Behind each material number is a trace back to the ledger account or source document it came from; behind each functional split is the method used; behind each triggered schedule is the fact that triggered it. Your reviewer is never asked to trust an unexplained output — they are checking work that shows itself, which is what makes a fast draft also a safe one.

The review workflow assumes a professional at the end of it, and it is designed to make that professional efficient rather than to work around them. The draft flags its own soft spots: year-over-year swings in program ratio, blanks where the source data was thin, compensation figures that didn't reconcile cleanly, governance answers that depend on a client confirmation rather than a document. Those flags direct the reviewer's attention to the handful of places that need judgment instead of forcing a line-by-line re-creation of the whole return. The result is a shorter review that lands on the right questions, which is the only kind of time saving that survives contact with a partner's standards.

Finally, the data. A 990 file contains donor identities, executive compensation, board rosters, and financial detail — some of it destined to be public, much of it, like unredacted Schedule B contributor names, emphatically not. UpTax handles it accordingly: source documents and drafts are encrypted in transit and at rest, access is scoped to the firm's engagement, and the public-inspection redactions are applied so that the copy meant for disclosure and the copy retained by the organization stay properly distinct. The client's data is used to prepare the client's return and for nothing else. On a document whose whole purpose is disclosure, drawing the line precisely between what is public and what is protected is not a feature bolted on at the end — it is part of preparing the return correctly.