The Forschungszulage is the only federal funding instrument where a refund arrives without a prior departmental decision: no committee, no quota, no race for the money. What that becomes in bootstrap reality is an entitlement that regularly sits unclaimed in closed financial years, because the projects have long been running but nobody ever wrote the application. § 169 AO reopens those years after the fact. Four full financial years are recoverable in typical situations, and the sum behind that is rarely trivial for a young company.[1][5]
How the retroactivity works
Under § 10 FZulG the Forschungszulage is credited against the income or corporation tax of the financial year in which the eligible expenditure arose, not the year of application. The application to the tax office isn't a grant application in the sense of grant law but a request to amend the tax assessment. That brings the procedural law of the Fiscal Code into play, with its deadlines rather than a funding programme's. The distinction is legally fine-grained and consequential in practice: grant funding has application deadlines, budget caps, first-come-first-served logic and departmental decisions; the Forschungszulage has assessment periods, a legal entitlement to a correct assessment and a procedure open all twelve months of the year.[5][6]
The procedure has two stages. First, an application to the BSFZ for a certificate that a project meets the definition of qualifying research and development within the meaning of §§ 2 and 3 FZulG. That certificate is a basic assessment notice under § 171 (10) AO and binds the tax office. Second, on the basis of the certificate the company applies to the tax office for assessment of the Forschungszulage. Neither step is tied to an annual deadline; both are tied to the tax assessment period.[3][10]
That two-stage structure is the lever. Because the BSFZ certificate establishes the position for each financial year separately, a company can submit several years in parallel: one ongoing project gives rise to several annual entitlements. The sum of the entitlements doesn't arrive as a lump; it arrives year by year as a credit against corporation tax or, in loss years or where the tax liability is exhausted, as a payment. For a loss-making startup the tax credit therefore becomes a liquid refund in practice: under § 10 (1) FZulG the part of the assessed allowance that can't be offset has to be refunded. That is the difference between a tax reduction in the classic sense (which lapses if there is no tax liability) and a genuine allowance.[5][8]
§ 169 AO and the four-year deadline
The legal basis is in § 169 (2) no. 2 AO: the assessment period is four years for all taxes other than excise duties. The FZulG expressly frames the Forschungszulage as a "tax incentive" (§ 10 FZulG) and it takes part in the ordinary assessment period.[1]
"The assessment period is (…) four years for taxes and tax refunds that are not taxes or tax refunds within the meaning of number 1." § 169 (2) no. 2 AO
What is decisive in practice is when the period starts. Under § 170 (2) no. 1 AO the assessment period begins not at the end of the financial year but at the end of the calendar year in which the tax return for that year was filed, and at the latest at the end of the third calendar year after the year in which the tax arose (the suspension of the start). That rule extends the window considerably in typical bootstrap situations: if you file your 2022 corporation tax return only at the end of 2023, the assessment period for 2022 begins on 31 December 2023 and ends on 31 December 2027.[2]
§ 171 AO extends the period further: as long as a basic assessment notice hasn't been issued, the assessment period doesn't expire. Because the BSFZ certificate is a basic assessment notice, a pending BSFZ application suspends the limitation for that year. Anyone filing a BSFZ application in good time before the deadline thereby secures the retroactive credit even if the BSFZ itself decides months later. The BSFZ's own application portal names no annual deadline but refers to the tax assessment period, which in tactical terms means: the BSFZ application may be filed as late as the relevant financial year's assessment period still allows, but it shouldn't be, because the suspension only runs until the decision and not beyond.[3][10]
§ 173 AO also applies where facts become known to the tax office after the fact that lead to a lower tax, and that is exactly the Forschungszulage situation: the fact of "a qualifying R&D project with expenditure of X" wasn't known to the tax office at the original assessment and becomes known through the BSFZ decision. The amending provision is therefore not only § 175 (1) sentence 1 no. 1 AO (basic assessment notice) but, alongside it, § 173 (1) no. 2 AO.[4]
Which financial years are still open
The Forschungszulage has been in force since 1 January 2020. The FZulG applies to eligible expenditure arising after 1 January 2020. Expenditure from 2020 and 2021 can in principle still be claimed, provided the assessment period hasn't already expired. In practice that means: anyone who files tax returns late and uses the suspension under § 170 AO can regularly recover four full years in which the expenditure arose, and in borderline cases five.[5][8]
As at 19 April 2026, the typical window for a company with a calendar financial year looks like this. Financial year 2021: open if the 2021 corporation tax return was filed after 31 December 2022, with the deadline expiring on 31 December 2026 at the earliest. Financial year 2022: open, expiring 31 December 2027 at the earliest. Financial year 2023: open, expiring 31 December 2028 at the earliest. Financial year 2024: open in any case; in many cases the tax return hasn't even been filed. For financial year 2020, whether it is open depends on whether the 2020 return was filed late enough; with ordinary filing in 2021, the deadline expired on 31 December 2025.[1][2]
The pattern that shows up in many young technology companies: the tax return for the year of incorporation filed late, the following years also delayed, deadlines therefore pushed far back, but a BSFZ certificate never applied for. The Forschungszulage for those years isn't lost, it simply hasn't been claimed. It remains available until the assessment period expires.
A second pattern concerns companies with a non-calendar financial year (1 April to 31 March, for instance). What governs the assessment period isn't the divergent financial year but the assessment period for income or corporation tax. The Forschungszulage is assessed in the assessment period in whose financial year the expenditure arose. The application decree makes clear that the assessment base covers only expenditure of the relevant financial year; shifting it into another year afterwards isn't permitted, even where that would be more advantageous for tax.[6]
A worked example: recovering four years
A bootstrapped software company based in Berlin, founded in 2021, has employed four developers continuously since 2022 on a compiler project of recognisably experimental character (new optimisation methods, unclear technical feasibility). The four developers' annual salaries are €75,000 gross each; plus the employer's social security contribution of around 20%. Under § 3 (1) FZulG, wages and provisions for employees' future security count towards the assessment base to the extent the employees work on R&D.[5]
A simplified, purely illustrative calculation, with no deduction for non-R&D shares: around €90,000 in claimable wage costs per developer, giving around €360,000 a year. At a rate of 25% (or 35% for small and medium-sized companies since the Growth Opportunities Act version), that gives a Forschungszulage of between €90,000 and €126,000 per financial year. Across four recoverable years (2022–2025), the order of magnitude is between €360,000 and €504,000, paid out or offset and not repayable. The actual figure depends on the R&D share of the staff time, any external service elements (up to 27 March 2024, 60% of the contract volume; from 28 March 2024, 70% under § 3 (4) FZulG as amended by the Growth Opportunities Act), the de minimis and state aid cumulation rules, and the far from trivial line between product development and qualifying experimental development.[5][6]
Operationally, recovery works like this: the company files a separate BSFZ application for each affected financial year, since the BSFZ issues certificates by financial year. Once the certificate arrives, the company applies to the tax office for assessment of the Forschungszulage for each year. The tax office amends the tax assessments already issued under § 175 (1) sentence 1 no. 1 AO (basic assessment notice) and credits the allowance.[3][10]
Sequence and tactics matter here. The BSFZ application should be filed for several years in parallel where the project is substantively connected across those years: the BSFZ assesses a project consistently across the years applied for and so avoids contradictory findings. The later application to the tax office, by contrast, is made year by year as each certificate arrives; a bundled submission isn't provided for procedurally because each assessment stands alone. The tax office assesses the assessment base (staff costs, own work, external services) independently of the BSFZ's finding, and that is the point at which recovery fails in practice when the derivation of staff costs isn't robust.[6][7]
Reconstructing the documentation
The real operational difficulty of recovery isn't legal but documentary. § 8 FZulG and the application decree require the funded staff expenditure to be evidenced per employee and per project. In bootstrap reality that is exactly the problem: often there are no timesheets, project plans are only rough, and lab or development journals are missing. The application decree does accept reconstruction after the fact in principle, but requires a robust derivation, not a plausible estimate.[6][7]
What can be used for reconstruction: Git history with commit messages and authors, a Jira or Linear export with ticket attribution, calendar entries, Slack or Teams archives, internal wiki pages with version history, pull request review data, meeting minutes. A combination of several of those artefacts supports a derivation of R&D shares. What matters is evidence that an employee worked predominantly on a project recognised as qualifying during a given period. The BSFZ FAQ stresses that the certification body assesses whether the R&D project meets the definition, while evidence of staff costs is a matter for the assessment procedure at the tax office.[7]
For external services the burden of evidence is stricter, because § 3 (4) FZulG counts only part of the contract volume — 60% for contracts up to 27 March 2024 and 70% for contracts from 28 March 2024 under the rate raised by the Growth Opportunities Act — and requires invoices plus contracts with a clear link to the R&D project. Invoices with no project-related description of the work are, in experience, problematic under scrutiny; a supplementary agreement with the contractor after the fact can document the attribution but doesn't substitute for the fact of the exchange of services.[5]
For shareholder-managing directors — the rule rather than the exception in bootstrap GmbHs — a separate mechanism applies: under § 3 (3) FZulG the shareholder-managing director can claim notional own work, capped at 40 hours a week. The hourly rate follows the time of the R&D work, not the year of application: for R&D hours up to 31 December 2025 the rate is €70 an hour, and for hours from 1 January 2026 it is €100 an hour under § 3 (3) FZulG as amended by the 2025 immediate tax investment programme. In recovering financial years 2021–2025, therefore, the €70 rate applies throughout; for 2026 hours open in parallel, the new rate applies. In recovery, the provision requires an hour-by-hour record per shareholder-managing director and per project, reconstructable from a calendar, commit history and customer communication, but not estimable on the spot. That reconstruction is what decides in practice whether recovery succeeds in full or only in part.[5][6]
Risks and limits
Three risks deserve to be named expressly. First: limitation runs year by year. Anyone relying on the apparent calm of the assessment period regularly loses the oldest year at the turn of the year. Anyone wanting to apply in January 2026 for financial year 2020 generally has no chance if the 2020 tax return was filed before 1 January 2022; the deadline expired on 31 December 2025.[1][9]
Second: state aid cumulation. The Forschungszulage isn't de minimis aid; it runs as notified aid under the General Block Exemption Regulation (GBER). For the same qualifying project, double funding is ruled out under § 7 FZulG where it would exceed the permitted aid intensity. Undertakings in difficulty within the meaning of the GBER are in principle excluded. Anyone who received grants (ZIM, state programmes) or EXIST funding for the same staff costs in the recovery years has to adjust the eligible expenditure accordingly.[5][6]
Third: the risk on the merits. The BSFZ rejects a significant share of applications on the ground that the project isn't experimental development within the meaning of the Frascati Manual and § 2 FZulG but product development, market adaptation or routine integration. The description of the project in the BSFZ application is therefore the lever: not the technical subject as such, but the argument along the criteria of novelty, creativity, uncertainty, systematic method and transferability. A badly written retroactive filing compromises all four years at once.[7][6]
A closing word on where this sits. The retroactive Forschungszulage isn't an insider tip but a sober consequence of the Fiscal Code and the FZulG. Nor is it a simple reflex: it presupposes deadline-accurate assessment bookkeeping, reconstructable R&D documentation and a project description that meets the statutory definition. For a bootstrapped company that has been working on a substantially experimental project for several years, the sum of the open years is structurally the largest single cash flow lever in the German funding catalogue, and one of the least often claimed. upsmart maps the application route and the BSFZ and tax office interaction as a platform, so that recovery stays operationally manageable year by year and the deadlines are in view before they pass.
- [1]Fiscal Code § 169 — limitation of assessmentFederal Ministry of Justice · gesetze-im-internet.de · 2024Open source
- [2]Fiscal Code § 170 — when the assessment period begins (suspension of the start)Federal Ministry of Justice · gesetze-im-internet.de · 2024Open source
- [3]Fiscal Code § 175 — amendment because of a basic assessment noticeFederal Ministry of Justice · gesetze-im-internet.de · 2024Open source
- [4]Fiscal Code § 173 — annulment or amendment because of new factsFederal Ministry of Justice · gesetze-im-internet.de · 2024Open source
- [5]Research Allowance Act (FZulG) — consolidated versionFederal Ministry of Justice · gesetze-im-internet.de · 2024Open source
- [6]Ministry of Finance circular of 7 February 2023 on granting the Forschungszulage (AEFZulG)Federal Ministry of Finance · 2023Open source
- [7]Frequently asked questions on the certificate under the FZulGBescheinigungsstelle Forschungszulage (BSFZ) · 2024Open source
- [8]Monthly report August 2025 — Germany's immediate investment programme (including the Forschungszulage)Federal Ministry of Finance · monthly report · 2025Open source
- [9]Fiscal Code § 171 — suspension of the limitation period (including for basic assessment notices)Federal Ministry of Justice · gesetze-im-internet.de · 2024Open source
- [10]Bescheinigungsstelle Forschungszulage — procedure and attribution to financial yearsBescheinigungsstelle Forschungszulage · 2026Open source
Forschungszulage 2026 for startups: what the €4.2m ceiling really means
A €12m assessment base, a 35% SME rate on application, plus 20% overheads: a worked example for an eight-person engineering team.
Forschungszulage 2026: a €12m assessment base, €4.2m for SMEs — what applicants need to know now
The immediate tax investment programme raises the cap again. Which documentation mistakes are surfacing in the first BSFZ wave, and how to be ready for them.
The de minimis trap: when one funding programme blocks the next
The €300,000 threshold counts across three tax years, and across group boundaries. How to avoid application B devaluing application A.
From analysis into the application.
We'll show you the platform on your actual case.