Magnetic

Documents come in by email, by portal, by paper — W-2s, 1099s, K-1s, Schedule Cs, brokerage statements, handwritten notes. An associate opens the doc, opens UltraTax, alt-tabs back and forth, and types numbers from one window into another. At a 5–50 person firm doing thousands of 1040s a season, that's the entire job.

The U.S. has 340,000 fewer accountants than it did five years ago.^[12] Accounting enrollment is down. Retirements are accelerating. Private equity is rolling up the firms whose owners would rather sell than recruit. The remaining staff log 80-hour weeks for four months a year. Firms now turn away clients because they cannot find the bandwidth.^{[5] [11]}

Thomson Reuters, Wolters Kluwer, Intuit, and Drake sell tax engines, not tax preparers.^{[1] [2] [3] [4]} Their "automation" is template-based imports and OCR overlays that hit ~50% field accuracy on anything past a clean W-2. Magnetic's proprietary vision stack hits 90%+ across W-2, 1099, K-1, Schedule C — including handwritten notes and messy scans — and 99%+ on W-2s.^[5]

The labor model collapsed. 340,000 fewer accountants than five years ago.^[12] 43% of financial decision-makers now invest more in automated accounting and AI tools, up from a rounding error two years ago.^[12] The AICPA's own pipeline report calls the trend "structural," not "cyclical." ^[10] Firm owners describe associate hiring as the bottleneck on revenue — they have more clients than they can take.

Computer-use crossed the line. Vision models now read tax documents at human accuracy and operate Windows applications at human speed. The same primitive that powers consumer AI agents now drives UltraTax and Lacerte field-by-field, keystroke-by-keystroke. Twelve months ago this didn't work; today, Magnetic is shipping returns into production at 8 paying firms.^[5]

PE consolidation forced firms to pick a side. Private equity has been rolling up CPA firms for two years. Owners either modernize or sell. The ones staying independent need an AI partner, and the buyer is the managing partner — not the procurement team, not the CTO. Sales cycles measured in weeks. Budget already exists; it just used to be called "associate compensation."

90%+ vs. ~50%. Magnetic's vertical vision model hits 90%+ field-level accuracy across the most common forms; legacy OCR baselines sit around 50%.^[5] The delta isn't a benchmark — it's the difference between an AI that saves the reviewer time and an AI that costs the reviewer time. SurePrep, AutoScan, and Document Intelligence sit on the wrong side of that line.^{[8] [13] [14]}

Reasoning over the tax code. Multi-state tax-exempt interest allocation. K-1 cascades. Schedule C income smoothing. These are not OCR problems — they're judgment calls that require reading the code and reasoning over the document. Vision transformers extract; the LLM allocates.

Reviewer-first UX, not preparer-first. The customer is the reviewer, not the AI. Workpapers are bookmarked and cross-linked to source documents so the reviewer can audit any field in two clicks. Trust comes from auditability — and the workpaper is what marquee firms switched on.

Pricing matches the labor model. ~$125 per return. ~$20k initial ACV per firm. Expansion paths to $50k–$100k ACV as coverage grows from 1040s into business returns and advisory work.^[5] The economic unit is the return — the same unit firms already track, bill, and staff against.

The buyer is the managing partner. Not the CTO. Not procurement. Not a champion who has to socialize the deal across the firm. The partner with hiring authority is the one who has spent the last two seasons trying and failing to hire associates. Magnetic is the answer that didn't exist in 2023, and the budget is already allocated — it just used to be called payroll.

The expansion path is the rest of accounting. Magnetic starts at ingestion and data entry. The natural next steps — review automation, planning & advisory, eventually engine — compound the product surface area and the data moat with every season. The same agents that drive UltraTax for prep can drive it for amended returns, K-1 cascades, and multi-entity allocation.

~80M U.S. individual returns are prepared by professionals every year.^[9] At ~$100 per return in addressable TAM, that's an ~$8B initial market on 1040s alone. The buyer pool overlaps almost perfectly with the install base of UltraTax, Lacerte, and CCH Axcess.^{[1] [2] [3]}

The expansion ladder runs from 1040s into business returns (1065, 1120, 1120-S), then into review automation, planning, and advisory — the full AI accounting OS. Each step compounds: the data flywheel from validated 1040s makes business returns easier; business returns make planning easier; planning expands the ACV per logo from $20k to $100k+ without ever needing a new buyer.