Magnetic

Buying preparation capacity, not software, is what is on offer here. A firm sends in the client's documents, an agentic AI system reads them and applies the figures to the return, an automated check compares the entered values back to the sources, and a credentialed US-based CPA or Enrolled Agent reviews every return before it goes back. The output is a prepared return inside the firm's own tax engine plus workpapers carrying flags and notes on anything that looks missing. Accepted inputs are deliberately messy: PDFs, spreadsheets, TIFFs, photographs, handwritten organizers and receipts, along with multi-state K-1s and profit and loss schedules.

On the two axes that separate these tools, Magnetic sits at one extreme. The human verification queue is not a sample, it is every return, and the reviewers are named as US-credentialed with active PTINs, which is a stronger claim than the offshore review most outsourcing arrangements run on. Data goes into the tax engine rather than into a bookmarked PDF, so the preparer starts at review. Security is stated as AES-256 in transit and at rest, US-only servers, and no model training on taxpayer data. Setup is described as five minutes through a portal, and the accuracy claim is 99.9 percent at field level with a guarantee that inaccurate work is free.

What cannot be checked is most of the rest. The site says data is entered directly into five of the leading tax engines and does not name one of them, which is the single most important fact for a firm deciding whether this works with Drake or UltraTax. No price appears anywhere, in any unit, so the difference between 200 returns and 1,500 cannot be modelled at all. No turnaround time is committed to, which for a service model in March is the constraint that matters most. There is no SOC 2 claim, no trust centre, no security page, no named leadership, no stated headquarters and no blog or changelog. The whole public presence is a single-page site.

Scale is early by the vendor's own account: 60 or so firms onboarded, against 350 for Soraban and thousands for SafeSend. Time saved is put at 77 percent per return. Those numbers are consistent with a young company doing real work, and nothing here reads as vapour, but a firm cannot do due diligence on a one-page website before handing over a season of client data.

The comparison worth drawing is with Juno, which is the mirror image. Juno publishes its per-return price, names its write-back targets, and states explicitly that no one on its side ever reviews the data, so the firm supplies the second pair of eyes. Magnetic supplies the reviewer and publishes neither the price nor the engines. Firms short on staff rather than short on time should prefer the Magnetic shape; firms with reviewers but no data entry capacity should prefer Juno's. Against Soraban and the fuller workflow tools, Magnetic covers only preparation, with no intake, delivery, 8879 signing or K-1 distribution.

One to watch is the right tier and stays. The proposition is well constructed and the human review model is genuinely differentiated, but sixty firms, an unnamed integration list, no published price and no security documentation is not enough to commit a busy season to.