Response to HMRC's Consultation on The Tax Administration Framework Review: Creating Innovative Change Through New Legislative Pilots

2023-07

I responded to HMRC’s open consultation on The Tax Administration Framework Review: Creating Innovative Change Through New Legislative Pilots" from HMRC published on 27 April 2023. With my background as a tax professional, having gained experience in both practice and corporate settings, I have been privileged to observe the transformation of the tax landscape firsthand. The sweeping changes brought on by digitalisation and innovation in the way businesses operate have affirmed that modernising our tax systems is not merely desirable—it is imperative. As tax systems become more complex, we must embrace innovative solutions and a modern, flexible approach to make tax administration more efficient and user-friendly.

I felt compelled to respond to this consultation because I believe that my hands-on experience with innovative tax solutions could provide valuable insights to HMRC. I’ve actively worked with innovative solutions in taxation, witnessing their potential to streamline processes, enhance accuracy, and reduce the burden on taxpayers and administrators. My experience navigating tax complexities also gives me a practical perspective on the challenges taxpayers face and the potential impacts of these changes.

By participating in this consultation, I hope to contribute to the broader discussion on the future of tax administration. Below is my original response to HMRC:

Dear Sir/Madam,

I’m writing to express my personal views, independent of my employer, regarding the the open consultation “The Tax Administration Framework Review: Creating Innovative Change Through New Legislative Pilots” from HMRC published on 27 April 2023. I appreciate the opportunity to participate in the open consultation. I’m also interested in participating in future discussions organised by HMRC that pertain to the concepts delineated in the consultation paper and this response.

Overarching Principles

The proposed sandbox legislative piloting model by HMRC (’the piloting model”) exhibits innovation and a forward-thinking stance. It’s reflective of similar methodologies, employed and tested in various sectors such as clinical trials in medicine and canary testing in software development.

The piloting model envisaged by HMRC will enable testing of new tax policies or processes within a controlled group of taxpayers. Similar to clinical trials and canary testing, it allows for discovery of hitherto unknown requirements, and early identification and mitigation of potential risks, thereby reducing the chance of unanticipated negative outcomes. This model shares several principles with the aforementioned methodologies, including:

Measurable and Transparent: The significance of clear communication with participants and the establishment of well-defined metrics for evaluation prior to launching the piloting program cannot be overstated. The metrics provide a clear framework for assessing the pilot’s outcomes and effectiveness, and insight into where improvements can be made.
Ethical and Fair: Drawing from the ethical guidelines and fairness principles observed in clinical trials and canary tests, the piloting model should ensure fair treatment of taxpayers and protection of their rights during the testing phase.
Robust and Resilient: The piloting model should be designed to be robust and resilient to deliver reliable results, even in the face of unexpected challenges, thereby ensuring its effectiveness.

With careful selection of pilot sample and the establishment of appropriate safeguards, the proposed piloting model aligns well with the 10-year strategy, published in July 2020, to drive productivity and innovation. By adopting an incremental and consultative approach, the piloting model will not only minimises disruption but also maximises the opportunity for feedback and continuous improvement in a fast changing political and economic environment.

Benefits and Challenges

Section 4.3 of the consultation paper aptly outlines several benefits and challenges of the piloting model. While I concur with HMRC’s perspective, I would like to add a few points:

Benefits:
1. Bridging Implementation and Policy Gaps: The piloting model can provide crucial insights into not only what works in principle, but also what’s practical and achievable in implementation.
2. Enhanced Outcome Measurement: The piloting model allows for a more precise estimation of policy impact timelines. This insight can aid in strategic planning and resource allocation, ensuring that HMRC is well-prepared to manage the impact of new policies and processes.
3. Risk Mitigation: By identifying and addressing potential risks early, the piloting model reduces the chance of negative outcomes after full implementation.
4. Cost-Effective: The piloting model can be more cost-effective than implementing a new policy across the entire taxpayer population at once by identifying and addressing roadblocks before significant resources are invested in a full rollout.
5. Stakeholder Engagement and Public Perception: By involving taxpayers and intermediaries in the pilot process, HMRC can build stronger relationships and foster a sense of collaboration and mutual understanding. Successfully executed pilots can enhance the public perception and confidence in HMRC as a trusted, innovative and forward-thinking organisation.
Challenges and risks:
1. Expectation Management: Managing the expectations of taxpayers and intermediaries participating in pilots can pose certain complexities. Clear communication is crucial, emphasising that every pilot represents a trial phase, and any alterations to policies or processes will be a result of these trial outcomes. Failure to convey this effectively could cause misunderstanding among taxpayers, especially those who are not directly involved in the pilot. To avert this, it’s essential to issue transparent public announcements to allay any perception of inequitable treatment towards the group involved in the testing phase.
2. Data Privacy and Security: Pilots involving new policies or processes may require the handling of sensitive taxpayer data. Ensuring the privacy and security of confidential information including Personally Identifiable Information (PII) is a critical challenge that must be addressed.
3. Evaluation Complexity: Assessing the success of a pilot can be complex, particularly if multiple pilots are run concurrently. Robust data analysis capabilities and development of new evaluation methodologies may be required by both HMRC and taxpayers.

Safeguards

Section 5 already provided a few examples of potential safeguards such as setting aside penalties for participants. Given the overarching principles defined in the aforementioned section, here are some general safeguards that HMRC could potentially consider when applying the piloting model approach. For more detailed examples on how safeguards should be tailored to individual pilots, see Examples section below.

Participant Consent: Participants of the piloting model should be fully informed about its purpose and the data being obtained. This includes information about the purpose and usage of collected data and metrics defined for measurement and assessment. Participants’ consent should be obtained before their involvement in the process.
Data Protection: Robust data protection measures should be in place to ensure the privacy and security of taxpayer data. This includes secure data storage and transmission, and strict access controls.
Fairness and Equality: Measures should be in place to ensure fairness and equality during the piloting process. This includes a process for addressing any potential bias in the selection of pilot participants and setting the correct expectations for the outcome of the piloting process. This could include providing additional support or resources (e.g. intermediaries) to participants if necessary.
Robust Evaluation: A robust evaluation framework should be in place to assess the pilot’s outcomes against the defined metrics. This should include independent oversight to ensure the evaluation is objective and unbiased, such as an independent third party advisor or agent.
Feedback Mechanism: A mechanism should be in place for participants and stakeholders to provide feedback on the pilot. This feedback should be used to refine and improve the pilot process. Feedback mechanisms are particularly important in policy areas where artificial intelligence is utilised such as large language models or traditional machine learning models.
Exit Route: An exit route should be in place along with early stopping criteria in case the pilot needs to be terminated early. This should include measures to minimise disruption for participants and to ensure that any data collected during the pilot is handled appropriately.

Further Examples

In this section, I list two innovative processes that HMRC could potentially implement to demonstrate further safeguards that might be warranted for individual pilots.

1️⃣ **An HMRC Guidance Explorer**

An HMRC Guidance Explorer backed by a Large Language Model (LLM) could provide better access to relevant guidance for taxpayers. The Explorer could be designed to handle a variety of queries from taxpayers and to feedback to HMRC on areas where additional example/case studies might be needed. Overall, this product should provide a more user-friendly experience for taxpayers and help to promote a more collaborative relationship between taxpayers and HMRC. I raise this as a specific example to emphasise the importance of safeguarding around sample selection. This proposed LLM-backed HMRC guidance explorer bears a relatively low impact if the tool gets something wrong. For example, the LLM may suggest a case study that could be added to HMRC’s guidance when it is not really needed. The key is to cover a diverse enough sample size so that a wide variety of potential circumstances are accounted for. In addition to the generic safeguards mentioned above, additional safeguards to consider in this case could include:

Representative Sampling: To ensure that the explorer tool can handle a variety of queries, and accommodate a range of different user profiles, a carefully crafted sample selection method should be put in place from a variety of different tax domains. Quality Control and Updates: Regular human reviews of the tool should be carried out to ensure its effectiveness. This should also include a feedback loop for users to provide comments and suggestions so the AI system can learn and improve over time. Accessibility: The tool is aimed for general purpose usage to provide a helpful and improved customer experience. It should therefore be accessible, inclusive and easy to use following best practice UX and accessibility design guidelines.

2️⃣ **Expense Classification Model**

Another example is an expense classification model using machine learning techniques to identify and categorise expenses for tax deductibility. This tool could equip taxpayers with a better understanding of their potential tax liabilities and thereby improve their ability to manage their finances. The model also includes a feedback loop, allowing HMRC to refine the accuracy of the model over time based on user feedback.

This example highlights instances where technology presents substantial opportunities to enhance HMRC’s routine task efficiency, while also necessitating an elevated level of governance. For this case, additional safeguards may include:

Transparency: The operation of the model, including its data usage and decision-making processes, should be clear to the participants. This clarity can uphold trust and encourage more effective feedback. Governance Mechanism: A robust governance mechanism should be established to oversee the machine learning model’s application. This includes defining roles and responsibilities for decision-making, maintaining data quality, ensuring model accuracy, and rectifying discrepancies or errors. Model Auditability: To uphold accountability, there should be a mechanism in place for auditing the machine learning model. This can ensure the model is operating as intended and adheres to the defined standards. Error Rate Tolerance: It’s essential to establish and communicate an expected range of error rates for the model’s predictions. Acknowledging that no model is flawless, this tolerance threshold allows for a degree of inaccuracies while also activating corrective measures when error rates surpass the pre-established acceptable limits.

Alternatives to Showcase Space

Interactive Digital Platform: Development of an interactive digital platform could provide stakeholders with real-time updates on ongoing pilots, facilitate feedback, and encourage discussions. This platform could include features such as forums, surveys, and live Q&A sessions. This would complement the ‘showcase space’ and make it easier for stakeholders to engage at their convenience.
Collaborative Design Workshops: HMRC could organise collaborative design workshops, inviting stakeholders to contribute their unique ideas and feedback to the design of new policies and processes. This could be executed in a ‘hackathon’ style event, with teams collaboratively developing and presenting their ideas. The most effective solution(s) can then be implemented in the pilot model for further showcasing.
Stakeholder Committees or Advisory Board: Forming committees with representatives from diverse stakeholder groups can ensure varied perspectives are taken into account, similar to the Financial Conduct Authority’s Innovation Advisory Groups. These committees can provide regular consultation, acting as a sounding board for innovative ideas from HMRC or provide checks and balances on some specific areas. It also provides an opportunity to engage stakeholders from sectors beyond tax policies, such as technology.
Open Data Initiatives: HMRC could share anonymised or synthetic data from the pilots with stakeholders, enabling them to conduct their own analyses and provide feedback. This would require robust data protection measures to protect the privacy of taxpayers.

Beneficial HMRC Activities

Generally, a pilot model employing an agile delivery approach is well suited to activities involving complex or evolving requirements, a need for flexibility, and a drive for continuous improvement. The pilot model, in particular, is effective where testing and validation of changes are required before a wider rollout. Activities in HMRC might benefit from this type of approach in addition to legislative and tax policy changes may include but not limited to:

Modernising the tax system to make HMRC central to UK’s national resilience and crisis response, as outlined in the 10-year strategy. The piloting model would allow HMRC to rapidly develop, test and iterate new systems with subsets of users, to gather feedback and refine the systems before full implementation whilst remaining resilient to sudden changes.
Enhancements to customer service including new contact channels, self-service options, or query resolution systems, could also benefit from this approach. Piloting allows for real-world testing and feedback before wider rollout across different taxpayer groups.
Creating data analytics tooling using cutting-edge AI technology for areas like fraud and anomaly detection, auditing, and tax return processing can be more effectively developed within a pilot framework. This approach would allow HMRC to evaluate and adapt policies and processes as necessary based on real-time feedback. It ensures that HMRC remains cognisant of technology limitations faced by taxpayers and intermediaries in order to formulate appropriate governance responses.