Unrelated Business Income: Screening Your Revenue Lines for UBIT Risk

The Exclusions and Modifications That Do Most of the Work

Clearing the three-part test does not mean the income is taxable. A long list of statutory exceptions and modifications sits behind it, and in practice these are where most revenue lines come to rest. The IRS maintains a reference page on UBIT exceptions and exclusions, and your screening criteria should track it rather than a summary written by somebody else.

Passive investment income is modified out under section 512(b). Dividends, interest, annuities, royalties, rents from real property, and most gains on the sale of property are excluded from unrelated business taxable income. This is why an endowment does not create a tax problem simply by earning a return, and it is the reason the word royalty carries so much weight in nonprofit licensing arrangements. Two large caveats apply. Income from debt-financed property is pulled back in under section 514, and payments from a controlled subsidiary are treated differently under section 512(b)(13). Both are covered below.

Qualified sponsorship payments are excluded under section 513(i), which is the provision that makes conventional corporate sponsorship workable for nonprofits. It is narrow enough to deserve its own section, and it has one.

The convenience exception covers a trade or business carried on by a 501(c)(3) primarily for the convenience of its members, students, patients, officers, or employees. The classic application is the school cafeteria or the hospital gift shop serving patients and visitors. The exception is about who is served, not what is sold, and the moment an operation is oriented toward the general public rather than the served population, the analysis shifts. A cafe open to passing foot traffic on a public street is a different thing from a cafe inside a members-only facility, even if the menu is identical.

The volunteer labor exception removes any trade or business in which substantially all the work is performed by unpaid volunteers. The donated goods exception removes any trade or business that consists of selling merchandise substantially all of which was received as gifts or contributions, which is the basis on which thrift store operations generally sit outside the tax regardless of whether staff are paid. In both cases substantially all is conventionally read as roughly 85 percent, and in both cases the practical difficulty is evidentiary rather than conceptual. If you intend to rely on either exception, you need contemporaneous records: volunteer hours against paid hours for the specific activity, or donated inventory against purchased inventory. Organizations lose these exceptions not because they fail the standard but because they cannot demonstrate that they met it.

Sponsorship or Advertising: The Line Drawn in the Language

A qualified sponsorship payment under section 513(i) is a payment by a business for which the business receives no substantial return benefit other than the use or acknowledgment of its name, logo, or product lines in connection with the organization's activities. That payment is excluded from unrelated business income. The moment the organization promotes, markets, endorses, or otherwise advertises the sponsor, the payment stops being a qualified sponsorship payment and becomes advertising revenue, which is taxable. The IRS discussion of advertising versus qualified sponsorship payments is short, readable, and worth putting in front of your development team directly.

The striking thing about this rule in practice is that the determination often turns on the wording of a few sentences. A banner that reads "Supported by Harbor Bank" alongside the bank's logo is acknowledgment. A banner that reads "Harbor Bank, the lowest mortgage rates in the county, apply today" is advertising, and the payment behind it is now presumptively taxable. Nothing about the sponsor, the amount, the event, or the organization's purpose changed. Only the copy changed. Qualitative or comparative claims, price information, indications of savings or value, endorsements, and calls to action are the specific features that cross the line. A neutral display of name, logo, locations, contact information, and a value-neutral description of product lines generally does not.

This has two consequences for how a nonprofit should operate. First, the sponsorship agreement and the deliverables it promises are the primary evidence, so vague benefit language in a sponsorship package is a liability. A tier that promises "prominent recognition across our channels" has not told anyone whether what gets delivered is acknowledgment or advertising, and the answer will be determined after the fact by whatever a staff member actually wrote. Second, where a package genuinely mixes benefits, the payment may need to be allocated between the qualified sponsorship portion and the advertising or other substantial return benefit portion rather than treated as all one thing. Organizations designing tiers will find the structural side of this in our guide to building corporate sponsorship packages, and the tax treatment should be settled while the tiers are being written, not after the invoices go out.

One more feature of section 513(i) catches organizations out. If a sponsorship payment is contingent on the level of attendance at an event, broadcast ratings, or other indicators of public exposure to the sponsor, it does not qualify as a qualified sponsorship payment at all. Agreements that tie payment to audience numbers are therefore worth flagging on sight. The fix is usually simple and commercial: pay a fixed amount, acknowledge the sponsor neutrally, and keep the exposure guarantees out of the contract.

Form 990-T, the 990 Itself, and the Paper Trail Behind Both

Filing obligations begin at a low level. Gross income of $1,000 or more from an unrelated business generally requires Form 990-T, and because the trigger is gross rather than net, an activity that loses money can still create a return. There is a specific deduction of $1,000 that keeps small amounts from producing actual tax, which is a different thing from keeping them from producing a filing requirement. For calendar-year 501(c)(3) organizations the return is generally due on the fifteenth day of the fifth month after year end, the same mid-May date as the Form 990, and electronic filing is required. Many states impose their own tax on unrelated business income as well, often keyed to the federal figures, so a federal 990-T frequently implies a state return nobody has thought about.

Before any of that, the Form 990 itself already demands the classification work this article describes, which is the fact most useful for getting internal attention. Part VIII, the statement of revenue, asks the organization to split every revenue line across columns: related or exempt function revenue, unrelated business revenue, and revenue excluded from tax under sections 512, 513, or 514, with exclusion codes identifying the basis. An organization filing a 990 is therefore asserting a UBIT classification for every dollar it earned, every year, whether or not anyone did the analysis behind the assertion. The Form 990 instructions set out how those columns work, and reading them against your own filed return is a fast way to discover whether your classifications were reasoned or inherited from last year's file.

The second half of the 990-T problem is expenses, and it is the half that draws scrutiny. Only expenses directly connected with carrying on the unrelated business are deductible against it, and where facilities or personnel serve both exempt and unrelated purposes, the costs must be allocated between the two on a reasonable basis. The regulations say reasonable without prescribing a method, which cuts both ways. It allows you to choose a sensible approach, and it puts the burden of defending that approach on you. Conventional practice allocates facility costs such as rent, utilities, insurance, and security by square footage and time of use, and personnel costs by time actually spent, with the method documented, applied consistently year over year, and constructed so no expense is counted twice. An allocation that moves conveniently each year is worse than a slightly rough one that holds steady.

Underneath both halves sits the documentation that decides how an examination goes. For each revenue line you need a written statement of what the activity is, which test each part of the analysis turned on, which exclusion you relied on if any, who reached that conclusion and when, and what evidence supports it. For each taxable activity you need the allocation method and its basis. This is the same discipline that makes an external audit straightforward rather than exploratory, and the practices in our article on audit preparation apply almost unchanged. Organizations already running a tight monthly process will find the natural home for revenue classification there, as discussed in the AI-assisted monthly close, because a classification captured when a new revenue stream first posts is worth ten reconstructed in May.

Step Two With AI: A First-Pass Screen, a Rationale File, and an Allocation Method

With the inventory built, the screen itself becomes a structured classification exercise. Give the model your criteria in writing rather than relying on what it already knows: the three-part test, the exclusions you might rely on, and the specific traps relevant to your operation. Then ask it to work through each inventory entry and produce, for every one, an answer to each part of the test, the exclusion it believes applies if any, the facts it relied on, the facts it would need and does not have, and a confidence level. The useful output is not a verdict. It is a sorted list: items that are clearly contributed or clearly mission-delivery and need no further thought, items that are clearly outside the test on frequency or exclusion grounds, and items that need a professional. Expect the third pile to be small, which is the whole point. Turning thirty revenue lines into four questions for a CPA is a service worth paying for, and you can generate it in an afternoon.

Contract and copy review is where the second big gain sits, because the sponsorship distinction lives in language and reviewing language at volume is a model's strength. Feed in your sponsorship agreements, benefit tier descriptions, program book pages, event signage copy, website recognition pages, and acknowledgment emails, and ask for every instance of qualitative or comparative claims, pricing or savings language, calls to action, endorsement phrasing, and payments contingent on attendance or exposure. Ask for it as a table with the exact quoted text and its location, which makes the output verifiable rather than assertive. The same pass should flag benefit descriptions too vague to classify, since those are the ones that will be resolved by whatever a staff member improvises. Every flag is then a human review item, not a conclusion, and the person reviewing is deciding both a tax question and a relationship question about what you promised a sponsor.

The rationale file is the deliverable that matters most in two years and gets built last, if at all. Ask the model to draft, for each revenue line, a short memorandum stating the activity, the analysis under each part of the test, the exclusion relied upon with its statutory reference, the evidence supporting it, the open questions, and spaces for the reviewer's name and date. Then have a human review and sign each one, because an unreviewed AI memo is not documentation of a conclusion, it is documentation that nobody reached one. Keep the file in a shared location with a plain naming convention, add a review date to the finance calendar, and record what changed each year, since the most common failure is not a bad conclusion but a correct conclusion that quietly stopped being true when the program changed.

Expense allocation is the last piece and benefits from the same treatment. For each activity your advisor confirms as taxable, a model can help design the allocation: identify which costs are directly connected, propose a basis for shared facility and personnel costs such as square footage, hours of use, or time records, check that no expense is being counted in two places, and document the method in a form that can be applied the same way next year. It can also build the supporting schedule and populate it from your own usage data. What it must not do is pick a method because that method produces a smaller number. The test is reasonableness and consistency, and an allocation chosen for its result fails both. Organizations with the broader finance picture in hand will find the budgeting context in AI-assisted budget management useful here, since a taxable venture needs its tax modeled in the budget rather than discovered in it.

A Screen You Can Actually Run This Quarter

Sequencing matters more than ambition here, because a UBIT screen that tries to be comprehensive on the first attempt tends not to finish. Week one is the inventory: export the chart of accounts and a year of revenue detail, have a model group it into activities, and list the gaps. Week two is the interviews, filling the operational fields for each activity by talking to the people who run them, with a model structuring what they tell you. Week three is the first-pass screen and the contract extraction, producing the three piles and the flagged language table. Week four is the rationale file and the brief, then one conversation with your tax advisor about the handful of items that survived triage.

Three operating principles keep it from becoming shelfware. Write your criteria down before you screen anything, because a classification that cites a statutory reference can be checked and one that cites a model's opinion cannot. Have a named human review and date every conclusion, since the documentation is the asset and an unsigned memo is not documentation. And connect the screen to the calendar, with the rationale file reviewed annually alongside the Form 990 preparation and a classification question embedded in the approval of any new revenue activity. Organizations that handle their broader information discipline well will find the mechanics familiar from our coverage of the Form 990 process with AI, where the same principle holds: the work is easier when the structure exists before the deadline.