Accountants for
tech startups

Tech startup revenue recognition is the single area where non-SaaS tech companies diverge most sharply from the subscription playbook. A marketplace earning a take rate on transactions between buyers and sellers needs to decide whether it reports revenue gross (the full transaction value, with the portion paid out to sellers as a cost of sales) or net (only the take rate as revenue). Under IFRS 15, the question turns on whether the marketplace is the principal (controls the service or goods before transfer to the customer) or the agent (arranges for the transfer). For most UK marketplaces the correct treatment is net, but the facts pattern matters: control over pricing, inventory risk, customer obligation, and who bears default risk all feed into the assessment. Getting gross wrong at first filing typically requires restatement at Series A due diligence, which is an avoidable but costly event.

Transaction fees, payment processing revenue, and interchange-based models have similar principal/agent questions. A fintech startup earning a spread on payment flow or a platform earning API transaction fees needs to identify who bears the risk at each point of the chain. API usage credits sold upfront are a variable consideration question: credits purchased but not yet consumed sit on the balance sheet as deferred revenue and recognise as the customer consumes them. Expired unused credits that are non-refundable under the terms recognise on expiry. Licensing revenue splits into point-in-time licences (a distinct, transferable right granted at a point) and over-time licences (access to IP that evolves over the licence period), with different recognition patterns. For a tech startup with a mix of revenue streams, the accounting policy document at first year-end should specify the treatment of each revenue type, because it becomes the reference investors diligence against.

A substantial share of UK tech product companies started as agencies or consultancies that built a tool for an internal client, realised it had commercial potential, and pivoted. The accounting during the transition period is genuinely difficult because the business carries two revenue streams (project-based agency work and product licences or subscriptions) with very different economics, two cost structures, and two valuation methodologies. Investors valuing the company for a Series A almost always want to strip out the agency revenue and value only the product ARR, which requires management accounts that show both revenue streams cleanly separated at the gross margin line, not blended at the P&L total.

The practical questions are about cost allocation. Engineers split across client projects and product development need timesheet or ticket-based allocation, which matters both for R&D claims (only product-directed qualifying work counts for R&D relief on the product) and for product unit economics (client work subsidising product development needs to be visible as such). A chart of accounts that creates separate service lines for agency and product work from the first accounting period, with consistent cost allocation rules, is substantially cheaper than rebuilding retrospectively. The transition typically ends with either the agency work winding down as product revenue scales, or the agency spinning out as a separate entity, and the structural work at first accounting period should accommodate both outcomes.

R&D credit claim methodology for AI and ML companies has specific features that differ from general software claims. Under the merged scheme, qualifying work needs to resolve scientific or technological uncertainty that a competent professional in the field could not readily deduce. For AI companies, this typically covers novel model architecture work, training methodology that goes beyond applying existing approaches, data preparation and augmentation at meaningful scale, specialised inference optimisation, and infrastructure work that enables model training at useful scale. What does not qualify is standard model fine-tuning against existing architectures, routine API wrapping of third-party model providers, prompt engineering against commercially available LLMs, and typical application-layer integration work.

The cost categories that matter most for AI and infrastructure-heavy tech companies are cloud compute and data costs, which are specifically qualifying under the merged scheme where directly consumed in R&D. For a company spending six-figure monthly cloud bills on GPU training runs, this is often the largest single line of qualifying expenditure, and the allocation between qualifying R&D use (training, experimentation, research-phase inference) and non-qualifying use (production inference serving paying customers, development tooling, operations) needs to be documentable at invoice level. Cloud billing line-item tagging, set up at first major infrastructure spend, is substantially cheaper than retrospective allocation at claim time. Externally provided workers and specialist subcontractors (ML researchers on short contracts, data engineering consultancies, domain-specific advisors) also qualify at the sixty-five per cent SME rate where the work addresses scientific or technological uncertainty.

UK tech startups routinely raise from a mix of UK and non-UK investors, and the SEIS or EIS route has specific constraints that interact with international investor expectations. SEIS shares cannot be issued to investors who already hold shares in the company, which catches out founder-and-friend rounds where an early contributor has been issued nominal equity before the SEIS round opens. EIS has a three-year qualifying trade requirement that covers the post-investment period, meaning acquihires or asset sales in the three years after an EIS round can claw back investor relief if the structure is not managed carefully.

US investor participation adds a layer. US VCs rarely accept standard UK SEIS/EIS share classes as-is, and a side letter or convertible structure often sits alongside the EIS-qualifying share issue. The specific form of any pre-emption, information rights, or protective provisions needs to remain compatible with EIS qualifying share rules, because a class of shares with unusual rights can break qualifying status. Advance assurance from HMRC is ordinarily obtained before share issue and covers the specific share structure proposed, and a material change to the rights of EIS shares after advance assurance requires fresh review. For UK tech startups preparing a mixed-syndicate round with US lead investors, the specialist accounting work is handled alongside the legal work, with the cap table and share class design consistent across both.

Open-source tech companies have a specific set of revenue and accounting questions that generalist accountants rarely encounter. The commercial model typically combines freely available open-source software (OSS) with one or more paid components: a managed hosted version of the OSS, an enterprise licence with additional features, paid support contracts, professional services, or dual-licence arrangements. Each paid component has its own recognition pattern. Hosted offerings typically follow SaaS-style subscription accounting. Enterprise licences split between point-in-time term licences and over-time licences depending on the rights granted. Paid support contracts are recognised over the service period. Professional services are recognised as work is performed.

For R&D credit purposes, the qualifying work on the underlying OSS project often does qualify even where the software is publicly available, provided the company is the economic owner of the development and the work addresses scientific or technological uncertainty. The documentation needs to identify the company as the commissioner and beneficial owner of the R&D, not merely a contributor alongside an unpaid community. Where the OSS project has external contributors who are not paid by the company, the claim covers only the work done by the company’s own employees and paid contractors, and the cost allocation between qualifying internal work and unpaid external contribution needs to be documented at first claim. Dual-licence arrangements (where the OSS is under a copyleft licence and a parallel commercial licence is sold to customers who want different terms) create an IP structuring question that matters at exit, because the commercial licence route frequently becomes the primary acquirer interest.

EMI option schemes for tech startups follow the standard structural pattern, but the specifics of pool sizing, grant allocation, and exercise price agreement differ by stack and team composition. An early-stage AI company with a small team of specialist researchers often needs individual grants at the higher end of the typical range (2 to 4 per cent of fully diluted for the most senior technical hires), whereas a marketplace or platform company with a broader engineering team typically distributes grants more evenly through an option budget. The pool size at Series A should accommodate both the next twelve to eighteen months of hiring and a top-up before Series B, because pool refreshes after a priced round dilute the founders disproportionately.

Exercise prices need to be agreed with HMRC using a defensible valuation methodology at each grant round, not self-assessed. For a tech startup that has raised an SEIS or EIS round with investor-agreed pricing, the pre-money valuation at the round provides a reference point for the option exercise price, typically with a discount to reflect the preferential rights of the investor shares versus the ordinary shares the options will vest into. The HMRC valuation opinion process is worth completing formally, because unagreed exercise prices create exposure at exercise or exit, particularly in the three years after a round where HMRC can revisit the opinion. The £250,000 per-employee unexercised option limit and £3 million scheme-wide cap apply throughout the life of the scheme, and both matter more at tech scale-ups with fast-growing valuations than founders typically anticipate.

Exit routes for UK tech startups divide into four main structures, each with distinct tax and structural consequences that depend on decisions made years before the exit. A share sale (acquirer buys shares in the UK company) is the simplest from the founder perspective, qualifying for Business Asset Disposal Relief at 10 per cent on the first £1 million of gain per founder where the founder has held 5 per cent or more of ordinary share capital and voting rights for at least two years and the company has been a trading company throughout. An asset sale (acquirer buys specific assets, typically IP and customer contracts) leaves the original UK company in the hands of shareholders, usually wound down after the sale, and the tax treatment depends on how the sale proceeds are extracted. An acquihire (acquirer primarily wants the team rather than the assets) is often structured partly as a share sale and partly as compensation to the founders and employees through the acquirer, with the compensation element taxed as employment income rather than capital gains.

The dual-track approach (running a fundraising process and a sale process in parallel) requires clean diligence-ready financials, clean cap table records, clean IP assignment from all past contributors and contractors, and clean R&D claim history through the full period. Contractors who worked on the product without formal IP assignment agreements are a recurring diligence issue that surfaces at exit and can significantly delay or reduce the sale price. Accountants in our network who work with UK tech startups flag the exit-diligence items at first accounting period and keep the record current year-on-year, rather than having to reconstruct it retrospectively at the point of acquirer interest.