R&D Tax Credits for Ecommerce: Six Figures Hiding in COGS

R&D Tax Credits for Ecommerce: Six Figures Hiding in COGS

A skincare founder I worked with last year had been running her business for four years. Revenue around $4M. Profitable. Her accountant filed taxes on time, every year, with no issue. She thought she had her financial house in order.

She had also left $120,000 in R&D tax credits sitting on the table across two financial years. Her formulation lab ran monthly batch trials, three of which failed every quarter. Every dollar of that work, contractor wages, ingredient costs, packaging prototypes, was bundled into Cost of Goods Sold and treated as ordinary operating expense. It never occurred to her accountant that physical product iteration was qualifying R&D activity. Her accountant assumed credits were for software companies and biotechs, not skincare.

When a R&D Tax Incentive specialist ran the numbers, she recovered $120,000 across the prior two years and built a forward-looking process to claim another $80,000 the following year. That money funded her next product line.

This is not an outlier. It is the default state of physical product brands between $1M and $10M.

The Skincare Founder Who Found $120K in Failed Batches

I will call her Maya, because the details of her business are real but the name is not. Her brand sells acid-based exfoliants and serums to a customer base she has built mostly through Instagram and TikTok. The product line started with three SKUs and now has fourteen. Most of those fourteen launched after at least four formulation trials. Some launched after six.

The lab notebook was always there. So were the contractor invoices, the ingredient COAs, the packaging prototypes that broke in transit testing, the failed stability studies. None of it was tagged. None of it was tracked against any qualifying framework. It was filed under "R&D" only in the loosest sense, as a label her ops manager used to organise PDFs in Google Drive.

The IRS Section 41 test qualifies activity that meets four conditions: the work must be technological in nature, eliminate uncertainty, involve a process of experimentation, and serve a permitted purpose. Maya's formulation trials cleared all four. So did her packaging prototypes. So did the Shopify code customisations her contractor built to handle subscription cancellation logic. None of it was claimed.

The accountant's assumption was the silent killer. R&D credits sit in popular memory as something for biotech founders and deep tech startups. The reality is broader. The AusIndustry R&DTI program in Australia explicitly covers experimental activities that resolve technical uncertainty, with no industry exclusion. The HMRC R&D relief regime in the UK covers any project that seeks an advance in science or technology. The US Section 41 framework is even broader. Yet most ecommerce accountants never raise the topic.

Maya's lost two years are not unusual. Most physical product founders running formulation, packaging, hardware, or platform iteration generate qualifying activity every quarter and claim none of it. The reason is structural, not lazy.

Why the Math Doesn't Work: The COGS Burial

Here is the structural problem. Most ecommerce P&Ls bury R&D activity inside COGS. Contractor costs for a developer who customised a Shopify checkout flow get coded as "platform fees" or "agency expense." Formulation chemist hours get coded as "product development" but rolled into COGS at a unit level. Packaging prototype runs get coded as "samples" or "wastage." The activity exists on the invoices. The activity does not exist in any queryable, audit-ready format.

When tax season rolls around, the accountant looks at the chart of accounts. There is no R&D line. There is COGS, marketing, payroll, SGA. The accountant produces a return that reflects what the chart of accounts shows. Nothing in the chart of accounts says "qualifying experimental activity," so nothing gets claimed.

This burial pattern destroys claim potential in two ways. First, by the time someone tries to reconstruct the activity, the supporting evidence is scattered across email threads, Slack DMs, Notion pages, and physical notebooks. Reconstructing it costs more than the recovered credit for some operators. Second, tax authorities increasingly demand contemporaneous documentation. The KPMG R&D credits commentary on Section 41 audits notes that retroactive documentation is the single biggest reason claims get reduced or rejected.

The financial gap is real and large. A typical $5M physical product brand running ongoing Shopify customisation, packaging prototyping, and formulation or hardware iteration leaves between $40,000 and $200,000 of annual R&D credits unclaimed. That range comes from the gap between the activity that qualifies under the four-part test and the activity that gets claimed when no one is tracking against the test.

The damage compounds. R&D credits are usually claimable for the prior financial year, with some jurisdictions allowing two prior years. Every year an operator does not run a structured claim, that year's potential credit drops off the back end. Three years of unclaimed activity at $80,000 per year is a $240,000 hole in retained earnings that will never come back.

Worse, the operators most likely to qualify are the ones least likely to claim. The skincare founder running formulation trials. The hardware brand iterating on injection mould tooling. The apparel brand testing fabric blends. The supplements brand running stability studies. The Shopify Plus brand whose contractor wrote a custom shipping logic engine. All of them qualify. Most of them never claim.

There is a second compounding problem worth naming. When an operator finally engages a R&D specialist mid-financial-year, the specialist often refuses to look backward. Some firms only work on a forward basis because the documentation gap on prior years is too thin to defend under audit. Operators discover this in the worst possible moment, after they have decided to claim and after they have signed an engagement letter. The window to recover prior-year credits closes quietly and without warning. The Blueprint exists in part to prevent this scenario.

The Qualified Activity Blueprint

I call the framework that fixes this The Qualified Activity Blueprint. It is a documentation and classification system that maps every iteration cycle inside a physical product business against the four-part qualifying test before the activity gets buried in COGS. It works on the principle that the documentation must exist at the moment the work happens, not at year-end when an accountant tries to reconstruct twelve months of contractor invoices.

The Blueprint has four moving parts.

The first part is the activity register. This is a single source of truth, owned by the founder or COO, that lists every iteration project the business runs. Formulation batches. Packaging prototypes. Code customisations. Hardware iterations. Stability studies. Supply chain process redesigns. Each project gets a one-line entry: project name, start date, technical uncertainty being resolved, expected outcome, owner, contractor.

The second part is the four-part test mapping. For each project on the register, the activity register includes a column for each of the four qualifying conditions: technological in nature, eliminates uncertainty, process of experimentation, permitted purpose. A project that cannot answer all four does not qualify and does not need to be tracked further. A project that answers all four becomes a claim candidate.

The third part is the failed experiment log. This is the highest-yield component and the one operators always overlook. Failed batches, failed code deployments, packaging prototypes that broke in transit, stability studies that crashed the formulation, supplier trials that produced unusable inputs. All of it qualifies. The qualifying conditions specify a process of experimentation, and an experiment that fails is still an experiment. Tax authorities treat failed experiments as some of the strongest evidence that genuine technical uncertainty existed. The BDO R&D credits operator guidance is explicit on this point: failed work is often the cleanest qualifying evidence. Operators who throw failed batches in the bin are also throwing the audit trail in the bin.

The fourth part is timing alignment. Every jurisdiction has cut-off dates and registration windows. Australia's R&DTI requires registration within ten months after the end of the income year. The UK requires the claim to be filed within two years of the end of the relevant accounting period. The US allows amended returns going back three years for missed credits. The Blueprint's timing layer ensures that the activity register and failed experiment log get reviewed and packaged for filing inside each jurisdiction's window.

I have run this framework with operators across formulation-heavy DTC brands, hardware startups iterating on injection moulds, and Shopify Plus brands running ongoing platform customisation. The pattern repeats. Operators who do not run the Blueprint discover, when they finally do, that they qualified for between two and five years of unclaimed credits. The first claim is usually the largest, because it captures backdated activity that suddenly becomes documentable.

Execution: Day 0 to Day 90

The Qualified Activity Blueprint runs in two phases. Phase 1 is the backward audit. Phase 2 is the forward documentation cadence.

Phase 1 (Days 1-30): Backward Audit. Start by pulling the last 18 months of contractor invoices, ingredient and packaging supplier invoices, lab notebooks, Slack threads tagged #product-dev, and Google Drive folders containing prototypes. The goal is a single spreadsheet listing every iteration project, with a one-line technical uncertainty statement for each.

Day 1 to Day 7 is data extraction. Pull the invoices. Filter for any contractor whose work touched product, packaging, or platform. Most operators find ten to thirty distinct projects in this pass. Day 8 to Day 15 is the four-part test mapping. Sit with each project and write the technical uncertainty being resolved. If you cannot articulate a technical uncertainty, the project does not qualify and gets struck from the register. If you can articulate it, mark all four columns and gather the supporting evidence.

Day 16 to Day 25 is the failed experiment recovery. This is the highest-yield work in Phase 1. Email the lab. Email the dev contractor. Pull failed batch records. Pull stability study failures. Pull deployments that got rolled back. Operators who do this systematically usually find 30% to 60% more qualifying spend than they recovered from invoice review alone. Day 26 to Day 30 is engagement of a R&D specialist, not a generalist accountant, to validate the register and prepare the claim.

Cost of Phase 1: typically $5,000 to $15,000 in specialist fees, plus 40 to 80 hours of internal time. Recovery: usually $40,000 to $200,000 in credits for a $1M-$10M brand with two years of unclaimed activity. The ROI math is rarely in dispute.

Phase 2 (Days 31-90): Forward Cadence. Phase 2 builds the system that makes Phase 1 unnecessary going forward. The activity register becomes a living document, owned by a named person, reviewed monthly. Failed experiments get logged within seven days of occurring. Contractor invoices get coded with a project identifier at the moment they are paid, not at year-end.

Day 31 to Day 45 is the chart of accounts work. Add an R&D classification layer. This does not have to be a separate GL account. It can be a project tag in your accounting system that applies across COGS, contractors, and SGA. Xero, QuickBooks, NetSuite, and most other accounting platforms support tagging that survives reporting.

Day 46 to Day 60 is the contractor agreement update. Every contractor invoice, going forward, must include a project tag that maps to the activity register. This is a one-line addition to the invoice template and a one-line update to your contractor onboarding doc. Do it once and it runs for years.

Day 61 to Day 90 is the monthly review cadence. Whoever owns the activity register sits with the lab lead, the dev lead, and the ops lead once a month for thirty minutes. New projects added. Failed experiments logged. Closed projects archived with their final outcome and supporting evidence. The whole meeting fits in a Google Doc with three sections.

Phase 2 ongoing cost: about three to four hours of internal time per month and a one-day annual review with the R&D specialist before filing.

A note on which jurisdictions to prioritise. Operators with multi-entity structures (an Australian Pty Ltd, a US LLC, a UK Ltd) often qualify in more than one jurisdiction for the same activity, but the rules differ on which entity captures the credit. An Australian-headquartered brand running a US Shopify build through a US contractor may be able to claim the Shopify work under R&DTI if the IP is owned by the Australian entity and the work is performed under contract. Get this wrong and you double-claim, which triggers an audit. Get it right and you amplify the recovery without doubling the workload. The R&D specialist engaged in Phase 1 should map this entity-by-entity before any claim is filed.

The Deloitte tax outlook commentary on R&D credit policy notes that documentation expectations are rising in every major jurisdiction, particularly around the interaction between Section 174 capitalisation and Section 41 credits in the US. Operators who run a contemporaneous activity register are positioned well for this trend. Operators who rely on year-end reconstruction will find the gap between potential and recovered credits widening.

From Buried Cost to Funded Trial

Maya's brand looks different now. The activity register is owned by her ops manager. Failed batch logs go into a Notion database within a week of the failure. Her formulation chemist tags every contractor invoice with a project ID. Her dev contractor does the same. The annual filing takes one day with her R&D specialist instead of three weeks of frantic reconstruction.

Her recovered credit funded her next product line. The new line launched profitably because the credit absorbed the launch cost. The failed batches that historically hit her P&L as wastage now hit her tax position as qualifying activity. Same activity, different financial outcome.

The shift is not about new revenue. It is about converting buried operating cost into funded trial budget. Every physical product brand running iteration is generating qualifying activity right now, today, in this quarter. The question is whether that activity is being captured against the four-part test or buried in COGS where it dies quietly.

The Qualified Activity Blueprint is not complicated. It is one spreadsheet, one chart-of-accounts update, one monthly meeting, and one specialist engagement per year. The reason most operators do not run it is not difficulty. It is the assumption their accountant would have raised it if it mattered. That assumption is wrong, and it is costing the average $5M physical product brand somewhere between $40,000 and $200,000 a year.

Pull your last 18 months of contractor invoices this week. Tag the ones that touched product or platform. Write a one-line technical uncertainty statement for each. You will know within an afternoon whether you have a six-figure recovery sitting in your inbox.