FedTally — federal-expenditure tally for US nonprofits

1. One-liner

Flags which of your state and foundation grants are actually federal dollars, before the auditor finds the ones you missed.

2. Trend signal — why now?

Three things moved in the last eighteen months, and they moved in opposite directions, which is exactly what creates a gap.

The threshold changed, and then it didn’t change for everybody. The Single Audit trigger rose from $750,000 to $1,000,000 in federal funds expended — the first change since 1997. But the effective-date rule is the trap: the revised Uniform Guidance applies to federal awards issued on or after October 1, 2024. Awards issued before that date stay under the old $750K threshold unless the awarding agency amends them. Any nonprofit holding multi-year grants that straddle the date is now running two thresholds against two pools of money in the same fiscal year. As one CPA firm put it, a nonprofit that spent $900,000 in federal funds can still be required to obtain a Single Audit if the older awards apply. That dual position persists through 2026 and beyond.

The thing that trips people up isn’t the arithmetic — it’s knowing which dollars count. Federal funds routinely arrive wearing a state agency’s name. A state Department of Education grant labelled “Title I” is federal money. A state Medicaid managed-care payment can include federal pass-through dollars. Subrecipients who don’t recognise the federal origin of a payment omit it from the SEFA, then get caught when the pass-through entity’s audit confirmation arrives. Incomplete SEFA is among the five most common Single Audit findings identified by GAO. An incomplete SEFA is a material weakness in the financial statement component of the audit — and a material weakness disqualifies you as a low-risk auditee, which means more testing and a bigger audit bill next year.

The data to check this became free and queryable. The USAspending.gov API v2 is public and requires no API key. It exposes prime awards and — with a subawards: true flag on the spending_by_award endpoint — subaward records searchable by the recipient’s UEI. It is not complete (prime recipients must report subawards over $30,000 through FSRS, and compliance is uneven), but that incompleteness is the point: the diff between what the federal record shows and what the nonprofit has booked is exactly the list of things a finance director needs to go ask about.

Meanwhile 2 CFR 200.332 already requires the pass-through entity to identify each subaward as a subaward and supply the Federal award identification, including the ALN and the amount of federal funds obligated. The duty exists. It is honoured unevenly. And when it isn’t, the subrecipient — not the pass-through — eats the finding.

Provenance:
  - Signal 1 (Demand): Incomplete SEFA is among the five most common Single Audit findings identified by GAO; it is treated as a material weakness and costs low-risk-auditee status, raising the following year's audit cost — https://www.smith-howard.com/6-common-issues-in-single-audits-from-an-auditors-perspective/ — 2026-09-05
  - Signal 2 (Feasibility): USAspending.gov API v2 is public, needs no API key, and supports subaward search by recipient UEI via `spending_by_award` with `subawards: true` — https://govconapi.com/usaspending-api-guide — 2026-09-05
  - Signal 3 (Economic): ~40,000 Single Audits were submitted to the Federal Audit Clearinghouse for FY2023 against $1.1 trillion in awards; a financial statement audit for a nonprofit spending $800K in federal funds runs $15,000–$25,000 — https://www.gao.gov/products/gao-24-106173 and https://grantedai.com/blog/single-audit-threshold-1-million-nonprofit-compliance-uniform-guidance-strategy-2026 — 2026-09-05
  - Signal 4 (Regulatory): 2 CFR 200.332 obliges pass-through entities to identify each subaward with its Federal award identification and ALN; SEFA instructions provide a "U" (unidentified) placeholder for when the ALN extension is unknown — https://www.ecfr.gov/current/title-2/subtitle-A/chapter-II/part-200/subpart-D/subject-group-ECFR031321e29ac5bbd/section-200.332 — 2026-09-05
  Category: Regulatory arbitrage

3. The opportunity

Every existing tool in this space assumes you already know which grants are federal.

Grant management software — Instrumentl at ~$162/mo, Grantseeker at $57/mo, Submittable from $14.99/mo, Good Grants at $400/mo — manages applications, deadlines and reporting. It is prospecting and workflow software. It does not classify your revenue by federal origin, and it does not run a Single Audit threshold forecast. Fund accounting systems (Sage Intacct, Blackbaud) will track a grant against whatever fund code a bookkeeper assigned it. If the bookkeeper coded the state DOE money as “State — Education,” the general ledger will happily report it as state money forever.

The audit firms know the answer, but they sell it once a year, after the fiscal year has closed, at audit-planning time — which is precisely the moment when it is too late to do anything except pay for the audit. Every CPA blog on this subject says some version of the same thing: for nonprofits near the threshold, the planning conversation should happen before year-end, not after the engagement letter arrives. Nobody sells the thing that makes that conversation possible in month four.

So the gap is a specific, narrow one. Not “grant management.” Not “audit defense” — the evidence-vault problem for organisations already in an audit is a different product. This is the classification-and-forecast layer that sits between the general ledger and the SEFA: which of my money is federal, whose pool does it belong to, and am I going to cross a line this year?

The wedge is that the federal government publishes enough data to answer most of this for free, in a format no nonprofit finance director will ever query by hand.

4. Target market

• Primary customer: Finance Director, Controller, or Director of Finance & Administration at a US nonprofit with $2M–$25M total revenue and $400K–$1.5M in annual federal expenditures, receiving money from three or more sources including at least one state or local pass-through. Typically a finance team of one to three people. Community health centres, workforce development orgs, Head Start grantees, housing and homelessness services, behavioural health, community action agencies, and school-adjacent nonprofits are the densest pockets.
• Why they buy: They are in the band where the answer is genuinely uncertain. Above $2M in federal spend, you are obviously getting a Single Audit and you have staff for it. Below $300K, you obviously aren’t. In the $400K–$1.5M band the answer depends on classification judgments they are not equipped to make — and the cost of being wrong in either direction is real. Wrong low: a material weakness finding, loss of low-risk-auditee status, more testing and higher fees next year, and a finding visible to their federal agency that can affect future award decisions. Wrong high: they budgeted and paid for a Single Audit they didn’t need.
• Rough TAM reasoning: About 40,000 Single Audits were submitted to the FAC for FY2023. The population that should be watching the threshold is meaningfully larger than the population that crosses it — call it 60,000–80,000 US nonprofits and small local entities with material federal exposure. If 15% of those are in the uncertain band and reachable, that’s roughly 10,000 organisations. At $149/mo that’s a ~$18M ceiling. I only need a low single-digit percentage of it.
• Why now for them: The dual-threshold period is live right now and will persist for as long as pre-October-2024 multi-year awards are still being spent down. This is a window, not a permanent condition — which is a risk I name in section 13, and also why the product has to be worth keeping after the window closes.

5. Product sketch (MVP)

• Federal-origin classifier. Upload or connect the chart of accounts and the grant revenue detail. Every funding source gets a verdict: Federal (direct), Federal (pass-through), Not federal, or Unresolved — with a confidence level and the reason.
• USAspending cross-check. Looks up the organisation by UEI against prime awards and subaward records, and shows the diff: federal money the public record says you received that does not appear anywhere in your books as federal.
• Dual-pool threshold forecast. Splits expenditures into the pre-October-2024 pool ($750K threshold) and the on-or-after pool ($1M threshold), and projects year-end position for each against its own line. Shows the gap to each threshold in dollars and in weeks of current burn rate.
• The ask-your-funder queue. For every Unresolved source, generates the specific written request to the pass-through entity citing 2 CFR 200.332 and naming exactly what they are obliged to provide — federal award identification, ALN, and amount of federal funds obligated. Tracks who answered.
• Draft SEFA, maintained continuously. A running Schedule of Expenditures of Federal Awards that is current as of last month’s close, not assembled in a panic in month fourteen. Exports to the format the auditor wants.
• Threshold-crossing alert. Email when the forecast crosses either line, or when a newly classified source moves the projection materially.
• Audit-readiness snapshot. A one-page PDF the finance director takes to the board or the audit committee: where we are, which pool, what’s unresolved, what it costs us if we cross.

6. AI angle — what’s load-bearing

Remove the AI and this product does not exist, because the core task is a judgment call on unstructured text.

A nonprofit’s federal exposure is buried in grant award letters, contracts, notices of award, state agency agreements, and reimbursement schedules — PDFs and Word documents with no standard structure. The question “is this federal money, and if so under which Assistance Listing” is answered by reading the award document, spotting the tells (a programme name that maps to a federal Assistance Listing, a citation to a federal statute, an ALN buried in an attachment, flow-down clauses referencing 2 CFR 200), and reasoning about whether the pass-through’s language actually establishes federal origin. That is document comprehension plus domain reasoning, and it is the whole product.

The second load-bearing use is reconciliation. Matching a nonprofit’s ledger descriptions (“WIOA Youth — Q3,” “State Workforce Bd”) against USAspending’s award and subaward records requires fuzzy entity and programme matching across two vocabularies that don’t agree. Deterministic string matching fails here; that’s why nobody has bothered.

What is not AI: the threshold arithmetic, the pool split, the alerting. That’s plain code, and it should be — I don’t want a language model deciding whether $940,000 is more than $750,000.

The honest caveat, which goes in the product and not just in this document: the classifier proposes, the finance director disposes. Every verdict is reviewable, every one carries its evidence, and the tool never files anything. A wrong confident answer here is worse than no answer.

7. Localization angle

N/A — this is a US-only play by construction. The entire product is a function of 2 CFR Part 200, the Assistance Listings taxonomy, the Federal Audit Clearinghouse, and USAspending.gov. There is no version of this for another country; the analogous products elsewhere would share zero code and zero domain model. The correct expansion is deeper into US adjacent entities — small municipalities, school districts, tribal organisations, and community colleges all face the identical SEFA problem — not sideways into other geographies.

8. Business model — path to $1M–$5M ARR

• Pricing: $149/month for the core tier (single entity, up to 25 funding sources, threshold forecast, SEFA draft). $349/month for multi-entity or 25+ sources, which is where affiliated corporations and fiscal sponsors land. Annual billing at ten months’ price, because nonprofit budget cycles want it that way.
• ACV: ~$1,900 blended, assuming a mix weighted toward the core tier and a healthy share on annual.
• Rough math to $1M ARR: 525 customers at a $1,900 blended ACV. Against a reachable population in the tens of thousands, that is a low-single-digit penetration rate. Reasonable.
• Rough math to $5M ARR: ~2,600 customers, which almost certainly requires the CPA-firm channel carrying most of the volume (see section 9) plus the adjacent-entity expansion — municipalities and school districts — since the pure nonprofit band gets tight around $3M.
• Expansion path: Per-entity pricing for fiscal sponsors and multi-affiliate structures is the cleanest expansion — those organisations have 5–40 entities each and the problem multiplies rather than adding. A second lever is a firm-tier seat licence for CPA practices who want it across their whole nonprofit book. I would resist usage-based pricing; finance directors hate variable software bills more than they hate high ones.

The unit economics are unremarkable in a good way: near-zero marginal cost per customer (the expensive input is document classification, which is bounded by the number of grant documents an org has, which is small), no hardware, no per-seat infrastructure. The constraint on this business is distribution, not margin.

9. Go-to-market wedge — first 100 customers

• Mine the FAC for near-threshold organisations. The Federal Audit Clearinghouse is public and searchable, and it publishes who filed, what they spent, and what findings they got. Two lists fall out of it directly. First: every organisation whose most recent submission carries a SEFA-related or material-weakness finding — they have already been burned, by name, in a public document. Second: organisations whose reported expenditures sit within 20% of either threshold. Both lists are hundreds to low thousands of names with the entity, the amount, and the auditor. Cold email that references their actual reported number and their actual finding is a completely different email from a generic pitch.
• Go through the auditors, not around them. The CPA firms doing nonprofit Single Audits are a concentrated, listed population — the FAC records name the audit firm on every submission, so the list of firms doing the most nonprofit single audits builds itself. These firms have a specific commercial pain: SEFA preparation is the part of the engagement that blows the budget because the client shows up unprepared. A tool that makes the client arrive with a clean draft SEFA is margin for the firm. Pitch it as a client-readiness tool the firm recommends, not as software that competes with them. Ten firms each recommending it to fifteen clients is 150 customers.
• State association conferences and the pass-through agencies themselves. Every state has a nonprofit association and most run an annual finance/compliance track. The talk writes itself and it is not a sales pitch: “The two thresholds you’re running this year, and the state grants you think aren’t federal.” State pass-through agencies also have a genuine interest here — subrecipient monitoring is their obligation under 2 CFR 200.332 too, and a subrecipient with a clean SEFA is less work for them.
• The free classifier as the top of the funnel. Give away the single-question version: enter your UEI, get the list of federal awards and subawards the public record says you received. That is a genuinely useful five-second answer, it is cheap to serve, and the follow-on question — “three of these don’t appear in your books as federal; want to find out why?” — is the product. This is the one channel I would spend on paid distribution for.
• Direct outreach to the multi-affiliate structures. Fiscal sponsors and multi-entity nonprofits are a small, identifiable, high-ACV list. There are maybe 300 fiscal sponsors of consequence in the US. They can be worked one by one.

10. Build complexity — justification

Low. The data sources are public and free: USAspending API v2 (no key), the Assistance Listings taxonomy, the FAC’s published submissions. The document classification is off-the-shelf model work over PDFs — no fine-tuning, no proprietary corpus, no training pipeline. The threshold and pool logic is arithmetic with a date rule. The genuinely fiddly parts are entity resolution between ledger descriptions and federal records, and getting the ALN taxonomy mapping right — both are grind, not research.

A technical founder with a competent nonprofit-finance advisor ships a credible v1 in 8–10 weeks. The advisor is not optional: the classification rules encode real domain judgment, and getting them wrong in a confident-sounding way is the fastest path to killing the product’s credibility.

The accounting-system integrations (QuickBooks, Sage Intacct, Blackbaud) are the one place scope can run away. V1 should accept a CSV export and nothing else.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Reads public federal data and the customer’s own documents. Files nothing, certifies nothing. Explicitly not an audit service and not legal advice — that disclaimer is load-bearing and belongs in the product.
Ethical — no harm / dark patterns	✅	Helps organisations comply more accurately, not evade. The threshold forecast is as likely to tell a customer they do need an audit as that they don’t.
Market exists (evidence above)	✅	~40,000 annual Single Audits; incomplete SEFA is a GAO top-five finding; audit costs of $15K–$25K at the relevant size.
1–5 person team can build this	✅	One technical founder plus a domain advisor. 8–10 weeks to v1.
Launchable with <$50K / ₹40L	✅	Free data sources, standard web stack, inference costs bounded by document volume. Realistically $8–15K to first revenue, most of it the advisor’s time.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Real money and real consequences — material weakness, lost low-risk-auditee status, higher fees, findings visible to the funding agency. Docked because the pain is annual rather than weekly. It sharpens hard at fiscal year-end and audit planning, and goes quiet in between. The threshold-crossing alert is what has to carry the other nine months, and I’m not certain it will.
Demand evidence	15	11/15	Strong indirect evidence: GAO names incomplete SEFA a top-five finding, every nonprofit CPA firm publishes on the dual-threshold problem, audit costs are documented. What I do not have is verbatim customer voice — I could not surface a single finance director saying this in their own words on a public forum, and I am not going to invent one. That absence is exactly why this is an 11 and not a 14.
Build feasibility	15	13/15	Free public APIs, off-the-shelf document classification, arithmetic. Entity resolution is the only genuinely hard part.
Distribution clarity	15	12/15	The FAC gives me a named list with dollar amounts and prior findings — that is a real list, not a demographic. The auditor channel is concrete and the incentive is aligned. Docked because nonprofit finance directors are slow-moving buyers with committee-shaped purchasing, and because the free-lookup funnel is unproven.
Revenue mechanics	15	11/15	$149/mo is well inside what this segment already pays for grant software, and trivially justified against a $15K–$25K audit. Docked for churn risk: an organisation that drops decisively below the threshold has a rational reason to cancel, and one that goes well above it graduates to something heavier. The sweet spot is a band, and bands leak at both edges.
Time to first revenue	10	8/10	The FAC list plus the auditor channel can produce paying pilots inside 6–8 weeks of a working v1. Not instant — nonprofits do not buy on the first call — but fast.
Defensibility	10	4/10	Honest score. The data is public, the logic is knowable, and any of the grant-management incumbents could bolt this on if it works. The only real moats are the accumulated classification rules for specific state pass-through programmes (which compound with every customer) and the CPA-firm relationships. Neither is strong at month three. This is an execution-and-focus play, and I’m scoring it as one.
Total	100	75/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

The technical work is unremarkable; the domain judgment is not. A builder without a real nonprofit-finance or Single-Audit advisor will produce a confident classifier that is wrong in ways they cannot detect, and one bad classification that costs a customer a finding ends the business. This is the rare idea where I would insist the advisor is locked in before a line of code.

Key assumptions to validate (3–5)

1. Assumption: Finance directors in the $400K–$1.5M band genuinely cannot answer “which of my grants are federal” without their auditor. How to test: Interview 20 of them. Ask them to name their federal sources from memory, then check against the FAC and USAspending record. The falsifiable version: if more than 12 of 20 get it substantially right unaided, the core premise is weaker than I think.
2. Assumption: The USAspending cross-check actually surfaces material discrepancies — federal money in the public record that the nonprofit has not booked as federal. How to test: Run the diff for 30 organisations from the FAC list before talking to any of them. If the median org has zero unexplained federal records, the headline feature is theatre.
3. Assumption: CPA firms will recommend the tool rather than treat it as encroachment. How to test: Pitch 10 firms that appear frequently as auditors in FAC records. The signal I want is a firm offering to introduce clients, not a firm saying “interesting.”
4. Assumption: $149/mo clears the bar for an annual-feeling pain. How to test: Pre-sell. Offer an annual plan at a founding-customer discount before v1 is finished and see who pays. Anything less than 5 prepayments out of 40 serious conversations is a pricing or urgency problem.
5. Assumption: The dual-threshold window is wide enough to matter. How to test: Sample 20 organisations’ award portfolios for pre- vs post-October-2024 issue dates. If most portfolios have already rolled over entirely, the “two pools” hook is dead and the product is a plainer threshold-tracker — still viable, less sharp.

Risk flags

1. Market timing / decaying hook: The dual-threshold complication is transitional. As pre-October-2024 awards spend down, the two-pool problem disappears and the product reverts to a single-threshold forecaster plus a classifier. The classifier is the durable half; the founder should treat the dual-pool feature as the wedge, not the product, and should not build the company’s identity around it.
2. Annual pain, monthly price: This is the risk I’d worry about most. The acute moment comes once a year. Subscription software sold against an annual pain churns unless it earns attention in between. The threshold alert and the continuously-maintained SEFA are the intended answer; whether they are enough is an open question that only retention data settles.
3. Accuracy liability: A confidently wrong classification could contribute to a real audit finding. The product must be positioned as decision support with visible evidence and human review, never as an authority. Terms of service and product copy both need to carry this, and the temptation to over-claim in marketing will be constant.
4. Incumbent bolt-on: Instrumentl, Submittable, or a fund-accounting vendor could add a federal-classification module. Defensibility is 4/10 for a reason. The counter is speed plus the CPA channel plus accumulated state-programme classification rules — all real, none decisive.
5. Data dependency: USAspending subaward coverage is acknowledged as incomplete, since FSRS reporting compliance is uneven. The product must present the cross-check as “here’s what the federal record shows” rather than “here’s everything,” or it will over-promise and get caught.

14. Structured verdict

Score:                  75/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder paired with a nonprofit-finance or Single-Audit
                        domain advisor. The advisor is a hard requirement, not a nice-to-have.
Time to revenue:        10-14 weeks (8-10 weeks to v1, then 4-6 weeks through a slow buyer)
Capital to launch:      $8-15K (mostly advisor time; data and infrastructure are near-free)
Top 3 assumptions to validate first:
  1. Finance directors in the target band cannot self-identify their federal sources —
     20 interviews, compare their recall against the FAC and USAspending record
  2. The USAspending diff surfaces real discrepancies — run it cold for 30 FAC-listed
     organisations before any customer conversation
  3. CPA firms will channel rather than block — pitch 10 frequent nonprofit auditors,
     count offers of client introductions
Kill criteria:
  - Abandon if the USAspending cross-check produces zero unexplained federal records
    for more than 20 of 30 sampled organisations (the headline feature is empty)
  - Abandon if fewer than 5 of 40 serious conversations will prepay an annual plan
  - Abandon if 3 or more of 10 CPA firms treat the tool as competitive rather than
    complementary (the primary scalable channel is closed)
  - Abandon if 12-month logo retention on the first cohort falls below 70% (confirms
    the annual-pain / monthly-price problem is fatal)

15. Next step — 1-week validation sprint

• Day 1–2: Pull the FAC data. Build two lists: organisations with SEFA-related or material-weakness findings in their most recent submission, and organisations whose reported expenditures sit within 20% of either threshold. Separately, tally which audit firms appear most often across nonprofit submissions. No product, just lists — and if these lists cannot be built cleanly in two days, that is itself a finding about how hard the data is.
• Day 3–4: Run the USAspending diff, by hand and unpolished, for 30 organisations drawn from those lists. Count how many have federal awards or subawards in the public record that look inconsistent with their reported SEFA. This is the single most informative thing in the week: it either proves the discrepancy exists at scale or kills the idea outright, before a line of product code.
• Day 5: Take the 10 most striking discrepancies to their finance directors directly — specific, personalised, “here is what the federal record shows for your organisation, does this match your books?” Not a pitch. A question. In parallel, call 5 of the frequently-appearing audit firms and ask whether client-side SEFA preparation is a budget problem for them.

Falsifiable outcome: Go if the Day 3–4 diff finds material unexplained federal records in at least 10 of 30 organisations, and at least 4 of the 10 finance directors contacted respond wanting to understand the discrepancy. No-go if the diff comes back mostly empty — that means the public record and nonprofit books already agree, and the product is solving a problem that resolves itself.