What it actually takes to win non-dilutive funding for an AI company

What was broken

An AI-native startup needed early non-dilutive capital to take its platform to market, and had to justify the ask against a credible utilisation plan.

What we did

CapEasy structured the application — a fund-utilisation plan tied to product and platform milestones, market sizing, and multi-year financial projections — and aligned the proposal to the scheme’s eligibility and permitted-spend rules.

A seed grant approval is the reviewing committee’s sanction; actual disbursement follows the programme’s due-diligence and compliance milestones and is not guaranteed.

Where it landed

The startup secured approval from a public seed grant programme.

The R&DTI is the ongoing programme; grants are the moving target

For an Australian AI company, the R&D Tax Incentive (R&DTI) is the piece of non-dilutive funding that runs every year rather than opening and closing on a round. Under the current rules, a company with aggregated turnover under $20 million gets a refundable offset equal to its corporate tax rate plus an 18.5% premium; a company at or above that turnover gets a non-refundable offset, at a lower premium for R&D spend up to 2% of total expenditure and a higher premium above that. Either way the claimable expenditure caps at $150 million a year and the minimum R&D expenditure to qualify is $20,000. The registered tax agent files the claim on your return — the offset rate and eligibility call are theirs and the ATO's to make.

Sector and stage-specific grants sit on top of that and behave differently: they open in named rounds, close on a date, and get replaced by the next thing. The Cooperative Research Centres Projects (CRC-P) programme is the clearest current example for an AI company doing applied R&D with an industry partner — matched funding from $100,000 to $3 million for projects up to three years, requiring at least two industry organisations and one research organisation, with rounds opening and closing on the Minister's schedule (Round 19 closed 12 May 2026, with up to $20 million of that round earmarked for AI systems work; the next round date was not yet announced as of this writing). That earmark is a useful signal of what "AI-specific" federal funding currently looks like — the discipline is to check the business.gov.au grants and programs finder for what is open now rather than to plan a raise around a round you read about once.

Software R&D has its own eligibility test — and its own paper trail

business.gov.au publishes a software development sector guide for the R&DTI specifically because software is the sector the Department sees the most eligibility disputes in. The line it draws: an activity is core R&D only when the team hits a genuine technical hurdle that "no existing knowledge, method, or solution can resolve," tested through a hypothesis-driven experiment — not when the team is configuring documented options, integrating a third-party API against its published docs, or migrating data with known tools. Building a new orchestration layer for AI agents that nobody has published a working method for can clear that bar; wiring an existing LLM API into a workflow tool the standard way generally does not, on its own.

The records the guide asks for are specific to that test: what the technical hurdle was and why known approaches didn't resolve it, what alternatives were researched, how the experiment was planned, and what the test results and conclusions were. It explicitly accepts the artefacts an engineering team already produces — spike tickets, architecture decision records, chat logs, performance graphs, meeting notes — rather than asking for a tax-specific document to be written after the fact. One exclusion matters for an automation company specifically: software built for the "dominant purpose" of your own or a related entity's internal administration is not eligible core R&D, though it may still qualify as supporting R&D if it is directly connected to eligible work.

• The technical hurdle, stated in writing — why documented methods and existing tools could not resolve it
• Research into alternative approaches, and why they were rejected
• The experiment plan and the preparations made before running it
• Test results, data analysis and the conclusion drawn — kept as they happened, not reconstructed
• Architecture decision records, spike tickets and technical logs, filed where they survive a tool migration

Project accounting is what makes a claim or an application defensible

business.gov.au's record-keeping guidance for the R&DTI is explicit that records "should be created at the time" the R&D activity happens, not assembled once a year, and must be kept for five years after the expenditure is claimed. The information it wants tied together is the same either way you're funding the work: what was done, how it was performed, who did it, what resources it consumed, and — critically — a cost allocation showing which R&D activities the expenditure relates to. That last point is a bookkeeping requirement as much as a technical one: a chart of accounts that already separates R&D-coded engineer time and cloud spend from ordinary operating expense, reconciled monthly, is what turns a registration or a grant application into an export instead of a reconstruction project the week before the 10-months-after-year-end registration deadline.

The same fund-utilisation-plan discipline from this engagement — milestones the spend is tied to, market sizing that supports the ask, multi-year projections that reconcile back to the ledger — is what a CRC-P application or an R&DTI advance finding request is built from. CapEasy's part is the plan, the schedule, and the ledger tied to the activity: which engineer-hours and which cloud spend sit against which project, month by month, ready for whoever reviews the claim. The eligibility determination, the registration with the Department of Industry, Science and Resources, and the lodgment itself are prepared for your registered BAS or tax agent — R&DTI is a self-assessment programme, and business.gov.au is explicit that registration confirmation is not an eligibility finding.