Accrual

The clearest thing about Accrual is what it refuses to do. It does not calculate tax liability, does not generate government forms and does not file. It takes the client's documents, splits combined PDFs into individual forms, classifies and rotates pages, decides which worksheets and state returns a return needs, then runs those worksheets in parallel and produces a set of preparer-reviewable schedules with every field cited back to the page it came from. Those worksheets are inputs. The firm's existing tax engine still calculates the return, runs diagnostics and produces the forms. When a late 1099 arrives, a checkpoint surfaces only the fields that moved rather than regenerating the whole thing.

Volume evidence exists and is unusually specific for a company this young. Accrual reports 15,000 returns in the 2026 season with 99.7% of worksheet fields generated by the model, and publishes a case study with Armanino, a Top 20 firm, covering six offices, 74% client portal acceptance and preparation hours down as much as 60% on complex returns. Security is well documented: SOC 2 Type II, AES-256 at rest, US-only storage, Microsoft or Google SSO, passwordless email login, scoped API tokens and an explicit no-training commitment on customer data.

The problem for a firm of 2 to 50 staff is that none of this was built for it. The integration story is API-first and the Model Context Protocol, not a list of prebuilt connectors, and no tax engine is named anywhere on the public site. A firm running UltraTax or Drake cannot tell from the marketing whether the export is a mapped import file, a spreadsheet, or a project. No price is published and there is no self-serve path. Nothing addresses e-signature, so Form 8879 with knowledge-based authentication and the fallback when a client fails KBA still comes from elsewhere. Coverage beyond the 1040 is described as a frontier problem in the company's own writing on trust and estate K-1s, which is candid but is not a shipped feature list.

Set against Black Ore, which publishes nine named tax programs it exports to, Accrual is less legible on the one question that decides whether the tool saves any time. Set against Filed, which publishes a firm-level flat pricing model and names small practices as customers, Accrual is aimed a long way upmarket. The Armanino result is the strongest single reference in this category, and it is also the reason a smaller firm should be careful about reading it as applicable.

One-to-watch stands, and the reason is fit rather than doubt about the technology. General Catalyst funding, a Top 20 firm in production and a published SOC 2 Type II are real. What is absent is any sign that a ten-person practice can buy this, price it, connect it to the tax software it already runs, and get through a busy season without an internal project. Until named connectors and a price appear, this belongs on a watch list rather than a shortlist.