Education Matters – March 2025

By Geraint Lewis, VAT Director, Moore Kingston Smith

After the flurry of activity in late 2024, schools have dealt with initial challenges of VAT registration and charging VAT to parents. Now, a few weeks into 2025, it’s a great opportunity to look at some of the challenges schools will be facing in the coming months.

Upcoming deadlines

The first deadline on the horizon is the submission of VAT returns to HMRC. For some schools, this will be as soon as early March, but for most, the first VAT return will cover the period up to 31 March, with submission and payment by 7 May. The VAT return itself is deceptively simple, typically requiring schools to report on five numbers:

1. VAT on income;
2. recoverable VAT on costs;
3. total income;
4. total expenditure; and
5. the amount due to (or from) HMRC.

But producing these numbers accurately and submitting the return can be far from simple.

The shift to digital

Sadly, gone are the days when a VAT return was a green form you posted to HMRC. Submission is now electronic, and the introduction of Making Tax Digital (MTD) a few years ago means that schools may need to invest in new accounting software or upgrade current packages to become MTD compliant. Another complication is that schools will need to complete partial exemption and/or business non-business calculations to determine recoverable VAT on costs. Setting aside the complexity of these calculations, standard accounting packages typically lack the functionality to perform these calculations or to include them within their standard environment. Therefore, these calculations often need to be completed manually and the VAT figures within the accounting system adjusted.

In most cases, this will mean that schools will need to obtain additional bridging software to comply with the MTD rules and to submit VAT returns successfully to HMRC.

Partial exemption and business non-business calculations

There has been much talk about partial exemption and non-business apportionment and the impact these will have on VAT recovery. I don’t propose to revisit these matters in this article, but rather to highlight a point of detail which appears to have slipped by unnoticed.

I expect most schools to adopt the standard income-based calculation. While it’s not the only option available, this approach is simple and, unlike special methods, does not require prior agreement from HMRC. But, for schools who received pre-payments before the law changed, a simple standard method calculation may not be seen by HMRC as ‘fair and reasonable’. Using figures for income received during the period (including January fees paid in 2024 on which VAT has been accounted for) would assign a value of £0 to pupils whose fees were prepaid, thereby artificially increasing the level of VAT recovery.

There are several ways to address this issue. For example, the standard income-based calculation could be amended to include an allocation of prepaid fees which relate to the current period. Alternatively, an extra step could be inserted to allocate a portion of general costs to prepaid students before any income-based calculation was undertaken. While both may be relatively straightforward, they still represent a deviation from the standard method which would require HMRC’s agreement.

Preparing for the first annual audit

The summer will see many schools undergoing their first annual audit since the VAT changes.  Auditors will require additional information about the VAT system and assurance that these systems are appropriate and robust. This will make the 2025 audit more complex and time consuming, and therefore potentially more expensive.

Dealing with HMRC

As the year rolls on, schools will face the reality of dealing with HMRC. Despite its adoption of terminology like ‘customer’ rather than the more accurate ’taxpayer’, HMRC exists not to serve taxpayers but to ensure that any tax revenue due is collected. Without seeking to comment on HMRC performance, there are a few guiding principles which schools need to have in their minds when dealing with HMRC. Very little of HMRC’s published guidance or notices have the force of law. HMRC’s various publications are simply a summary of their views.

HMRC can and do often argue in court that their public guidance cannot be relied on by taxpayers and is meant as general guidance rather than binding advice to individual taxpayers.

While individual HMRC officers are almost always fair and reasonable professionals, as an organisation, HMRC does not care about the impact their decisions may have on your school.

Progressing any matter with HMRC is a long, complex, and frustrating process. Nothing is ever resolved quickly, and don’t assume the person dealing with your case has any knowledge of matters previously agreed or discussed with their colleagues.

Prepayments

Finally, a word on prepayments, I believe it’s almost inevitable that HMRC will start enquires into prepayments in 2025, and that initially schools will be asked to provide all manner of additional information on this subject. Regardless of the technical merits of the point, I believe they will try and argue that some prepayment arrangements did not shelter fees from VAT and will issue assessments. How schools react to these is something for the future, but there are some important practical considerations.

A final resolution of any dispute with HMRC on prepayments will almost certainly involve complex and protracted litigation. Looking back at the last time prepayments were widely used to avoid VAT changes, this means it may well be 2032 or beyond before this matter is finally concluded.

Planning ahead for a smooth operation

In the meantime, the strategy for handling the VAT issue involves several considerations and raises the following questions:

1. What’s the strategy?
2. Are schools going to use funds on hand to pay the VAT, thereby reducing the value of the prepayments, and undertake to repay any overpaid VAT to families in five or six years?
3. Do prepayment structures, especially those which have been in place for several years, allow for this?
   Alternatively, is the plan to dispute the liability with HMRC and demand additional sums from the parents when the matter is resolved?
4. If so, will schools be able to recover funds from parents in say 2032, by which time children may well have completed their education, families moved on, or split up?
5. Finally in this scenario, who is going to fund the interest due to HMRC should these sums be subject to VAT?

By Jared Goodrich, Senior Manager, Digital Transformation, Moore Kingston Smith

The field of Generative AI is advancing at an unprecedented pace. While progress once unfolded over months or even years, new techniques and models are now being released in a matter of days. Assumptions made about AI just one year ago are quickly becoming outdated.

Hundreds of billions are being invested in Generative AI, from governments such as the US and France to companies funding nuclear power stations and data centres. Leading players including OpenAI, Google, and Anthropic, are now joined by a growing opensource community. Notably, DeepSeek-R1 is one such model from the open-source community that has been making waves in the news, matching the performance of frontier models.

AI systems are now outperforming highly trained professionals, PhD students, and even doctors, at their own game. For instance, a recent medical diagnostics study, Large Language Model Influence on Diagnostic Reasoning: A Randomized Clinical Trial, found that an AI model achieved 92% accuracy in complex case evaluations, compared to 74% by a doctor. While this is one aspect of the job, the implications are clear. With the right framework, AI can handle data-intensive tasks better than human experts. In schools, AI could help bursars and administrators analyse large volumes of financial, attendance, and scheduling data to support decision-making.

Despite these impressive advancements, integrating AI into daily workflows has been slower than many anticipated. For most users, especially in school settings, AI functions primarily as a simple copilot, rather than a fully autonomous intelligent agent. Whilst there is largely a consensus that AI represents the future of work, a recent RAND report, The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed: Avoiding the Anti-Patterns of AI, shows that more than 80% of AI projects fail. This demonstrates the need for robust frameworks and realistic implementation strategies.

Reasons for limited impact

The most prominent limitation of Generative AI is hallucinations, this is when AI fabricates incorrect information which prevents us from relying on the output. Newer models that can perform multi-step reasoning tend to hallucinate less and are more reliable. As AI advances, this trend will continue. In addition, various techniques can further reduce hallucinations. For example, granting AI access to tools such as web searches can help ground responses in factual data.

A thought-provoking question emerges: do AI (Large Language Models (LLMs)) need to completely eliminate hallucinations, or is it sufficient for it to outperform even the most accurate humans? “It turns out that humans also get things wrong sometimes and that the best model, whilst not perfect, produced fewer instances of fabricated information than even human medical experts. Far from introducing inaccuracies, LLMs could end up reducing fabricated information”. Further information can be found from this study from Stanford University: AI can Outperform Humans in Writing Medical Summaries.

Another key consideration is the AI’s context window size, which determines how much information it can hold in working memory at once. In practice, everyday tasks require far more information than we might realise. Even replying to a single email can involve recalling numerous details, this might mean juggling budget lines, payment schedules, staff payroll records, and vendor contracts all at once. While techniques such as Retrieval-Augmented Generation (RAG) can help by selectively retrieving and passing only the relevant context to the query to the AI, they are not without their drawbacks. Schools must still develop a robust data strategy for effectively gathering, organising, and supplying the information AI requires.

Data privacy remains a major concern for schools and organisations alike. There is also the question of whether schools should grant AI access to all of their sensitive data, and if they do, is that data being used to train AI? There are solutions to this problem, one solution is open-source models, which allow you to host these models locally, ensuring your data stays on school premises and securely under your control.

Putting the right framework in place

These limitations can be addressed by having the right framework and tools in place. We have therefore developed our own Generative AI platform to harness the latest AI advancements, with data security and quality at the forefront. One such framework is OpenAI’s Deep Research, which uses multi-step reasoning and web search to produce comprehensive, in-depth reports. The following highlights an example of an output generated on the topic of AI detection in schools: AI detection in schools.

Looking ahead: The year of the Agentic AI

2025 is set to be the year of Agentic AI, intelligent systems that act with a degree of autonomy. While tools such as the OpenAI Operator are not yet fully autonomous or production-ready, they offer a glimpse into the future. Schools will be able to adjust learning materials to cater to individual student needs, manage the logistics of school events like field trips with minimal human oversight, and handle the administration such as policy generation. The use of AI is here to stay; it’s only a matter of when it will be adopted in schools.

The new Charity SORP

The next iteration of the Charities Statement of Recommended Practice (SORP) is expected to be implemented from 1 January 2026, so for most charities this will be applicable for the first time for the year ended 31 December 2026, 31 March 2027 and 31 August 2027. Feels like a long way away, but when a new SORP is applied for the first time there are “retrospective applications” resulting in restatements for the comparative period and adjustments to opening balances at the start of the prior year (so 1 January 2025 or 1 April 2025).

The Economic Crime and Corporate Transparency Act 2023 (ECCT Act)

Having received Royal Assent on 26 October 2023, the new rules under the Act came into effect (subject to secondary legislation) on 4 March 2024 and introduced a transformation of Companies House’s role from registrar to regulator. This will impact any charitable company or a charity that has a trading subsidiary or a charity that has a corporate director. An updated timeline of change events is as follows:

• Effective 27 January 2025: Companies House has been able to receive and assess applications from individuals seeking to have residential addresses suppressed from public disclosure where they’ve been used as a registered office address.
• From 25 February 2025: Companies House should be able to (a) expedite the striking off of companies where the registrar has concluded the company has been formed for a false basis and (b) carry out checks on Authorised Corporate Service Providers (ACSPs) to authorise them to carry out verification services – ACSPs will need to be registered in the UK and subject to the UK’s anti-money laundering regime.
• From 25 March 2025: Companies House will be able to allow individuals to voluntarily verify their identity.
• By Summer 2025: Companies House will be able to (a) allow access on request to certain trust information on the Register of Overseas Entities; and (b) receive and assess applications from individuals seeking to suppress their date of birth for documents registered before 10 March 2015, signature, business occupation, and residential address in most other instances.
• By Autumn 2025: Companies House will be able to (a) make identity verification a compulsory part of incorporation and new appointments for new directors and PSCs; and (b) begin the 12-month transition phase to require more than 7 million existing directors and PSCs to verify their identity – the identity verification will happen as part of the annual confirmation statement filing.
• By Spring 2026: Companies House will be able to (a) make identity verification of the presenters a compulsory part of filing any document; (b) require third party agents filing on behalf of companies to be registered as an ACSP and (c) reject documents delivered by disqualified directors as they will be prohibited from doing so, unless they are delivered by an ACSP for specified filings permitted by law.
• By the end of 2026: Companies House will be able to (a) require all limited partnerships to submit more information, providing greater transparency for users of the register; (b) complete the transition period for all individuals on the register requiring identity verification, and start compliance activity against those who have failed to verify their identity; and (c) facilitate greater cross-checking of information and data between Companies House and other public and private sector bodies.

Further information about the Act can be found at on the government website, Economic Crime and Corporate Transparency Act 2023.

Data Use and Access Bill 2025 (e-mail marketing rules)

For the first time since 2003, charities will be on a level playing field with for-profit entities with respect to their ability to contact potential donors by e-mail/SMS. The detail revolves around the ability for charities to use a soft opt-in approach rather than relying on consent as the only lawful reason to send direct marketing messages about their charitable purpose. The amended bill is currently sitting with the House of Commons. Once the bill receives Royal Assent, charities will be able to send direct marketing e-mails/text messages if they obtained the contact details when the individual was expressing an interest in supporting or donating to the charity. The charity also needs to provide a simple means of refusing direct marketing messages, both at their initial point of data capture and during all subsequent communications.