RESEARCH DOSSIER
International Insurance Regulatory Sandboxes Institutional Design
Abstract. Insurance regulatory sandboxes institutionalize supervised experimentation under bounded legal permissions. Their design determines who can enter, which activity can be tested, which obligations remain operative, and how temporary permission becomes market access or orderly cessation. This monograph compares insurance-specific and cross-sector arrangements in Brazil, Hong Kong, Italy, the United Kingdom and South Africa, supplemented by European and international supervisory evidence. It distinguishes insurer licensing, technical experiments, intermediation and cross-border coordination. The analysis integrates primary supervisory records, empirical fintech funding research, descriptive admission and conversion statistics, and a hypothetical actuarial exposure calculation. Brazil's March 2026 administrative snapshot establishes definitive authorization for six of twenty selected insurers, with seven conversion proceedings in progress. British cross-sector admissions and funding results concern different populations and cannot be pooled with those outcomes. The research finds that institutional performance depends on clearly bounded legal permissions, credible protection of insurance obligations beyond the testing interval, transparent entry and exit records, and the capacity to transform observations into general supervisory learning. Authorization and funding establish particular intermediate outcomes; durable insurance protection, consumer welfare and competitive effects require insurance-specific longitudinal evidence.
Keywords: Insurance regulation; regulatory sandboxes; experimental authorization; insurtech; insurance licensing; prudential supervision; policyholder protection; market access; comparative institutions; regulatory learning.
Introduction
Insurance is a promise to perform when a specified uncertain event occurs. Its institutional support consists of enforceable contractual rights, collectible assets, competent risk management, reliable information and supervisory arrangements that protect the promise through time. A technological change in distribution can leave the underlying contractual promise substantially unchanged, while a new arrangement for assuming risk can affect the entire institution. Regulatory experimentation must consequently distinguish the transformation of the service from the transformation of the risk bearer.
This study examines the institutional design of international insurance regulatory sandboxes: the allocation of permission, discretion, accountability, monitoring and market access during supervised experiments. Three questions organize the inquiry. What establishes admission and its legal consequences? How is an experimental insurance activity supervised while contractual obligations arise? What can be inferred from conversion, launch, investment, cessation and regulatory change? These questions connect insurance markets with the legal architecture of public supervision and the technological transformation of insurance operations.
The unit of institutional comparison is a scheme in a given jurisdiction and period. Each legal entity, proposition and authorized experiment is a separate observational unit. A firm participating in several tests remains one firm but supplies multiple test records. An insurer conducting several product trials retains an underwriting balance sheet shared across its activities. Evaluating each experiment individually must therefore be complemented by scrutiny of aggregate obligations at the legal-entity level.
The cutoff is November 19, 2026. The research is a purposive comparison of documented institutional arrangements, not a worldwide census or a comprehensive legal opinion. The available evidence allows direct observation of administrative decisions and specific empirical outcomes. It does not supply a harmonized, policy-level international dataset. The study accordingly distinguishes observed facts, analytical interpretation, proposed measurement methods and hypothetical arithmetic.
Conceptual and Historical Foundations
Insurance promises and institutional experimentation
An insurance regulatory sandbox is a supervisory arrangement for evaluating a specified innovation under agreed controls and a legally bounded permission structure. A live experiment involves actual market activity, policyholders or insurance processes; its scope can be constrained by duration, customer population, product, geography, transaction value or the responsibilities of a licensed partner. Program enrollment, supervisory guidance, modification of an administrative rule, conditional licensing and statutory exemption have distinct legal effects.
The IAIS (2024) Insurance Core Principles connect protection of policyholders and financial stability with supervision of the risks associated with insurance activities. The framework supplies internationally shared supervisory concerns; the enforceable arrangements in a particular market depend on implementation and competent legal authority. The fact that an international standard supports proportionate supervision does not supply a domestic supervisor with an independent power to waive national law.
The economics of experimentation begins with uncertainty about technology and regulatory treatment. An incumbent may face high fixed compliance costs to implement an unfamiliar channel. An entrant may need to resolve questions about classification before it can apply for an appropriate authorization. Supervisors may lack practical evidence about the operation of a new product or about adverse customer outcomes. A bounded experiment can generate information at a smaller scale than unrestricted deployment. The benefit is the reduction in the cost of resolving uncertainty, including avoided compliance redesign. The resulting benefit should be assessed alongside the costs imposed on customers, firms and supervisory resources.
The IAIS (2018, pp. 24–25) digital inclusive insurance paper situates new digital activities within test-and-learn approaches and proportionate licensing conditions. The implications are particularly important when conventional distribution is expensive or consumers have limited experience of formal insurance. Technology can reduce costs while creating barriers involving literacy, identification, connectivity or inaccessible complaints mechanisms. Enrollment alone cannot establish effective inclusion. An appropriate test observes persistence of coverage, comprehension, service reliability and actual protection.
Historical differentiation
The FCA opened applications for its live sandbox in June 2016. Its 2017 lessons-learned summary locates the scheme within its competition mandate. Digital financial services comprise multiple sectors, so the history of broad fintech experimentation cannot be equated with a history of experimental underwriting.
In Hong Kong, the Insurance Authority's September 2017 launch announcement established two distinct mechanisms: controlled testing by authorized insurers and a dedicated authorization queue for new insurers using fully digital distribution. Accelerated ordinary authorization and supervised experimentation respond to different barriers. One relates to the administrative process of obtaining permission and the other to testing a proposition under bounded conditions. Treating both as identical would distort the meaning of graduation and new market entry.
Italy developed an arrangement involving the Ministry of Economy and Finance and multiple competent supervisory authorities. The IVASS program overview identifies Ministerial Decree 100 of April 30, 2021 as the basis for this experimental environment. South Africa also illustrates interagency organization: the SARB fintech overview records the IFWG's establishment in 2016 and its 2020 Innovation Hub, which differentiates guidance, regulatory experimentation and exploratory collaboration. These examples demonstrate that a sandbox's institutional organization depends on the distribution of supervisory responsibilities within the financial system.
The proliferation of these mechanisms reflects digitally mediated distribution, data-intensive risk classification, embedded insurance, automation of claims and specialized suppliers operating within insurers' value chains. The relevant form of novelty follows the function changed. A digital interface may support a customary insurance contract; a simple-looking product may create substantial new uncertainty about the identity of the risk carrier, the source of financing or the allocation of payment obligations.
Institutional alternatives
A sandbox is one available instrument for learning and regulation. Guidance clarifies existing requirements; an innovation hub facilitates access to the regulator; a prototype environment examines software without issuing live coverage; ordinary authorization accommodates established activities; generally applicable rulemaking addresses shared legal uncertainties. The practical sandbox design guide (CGAP 2020, introductory sections) emphasizes determining whether actual live testing is necessary and considering alternative facilitators.
For instance, a claims model can initially be evaluated against archived, de-identified damage photographs. It may be possible to estimate technical error rates without changing policyholders' rights. A system that automatically rejects real claims requires evidence about the consequences of such errors in actual service. A parametric product may require information about the comprehension of the trigger and the variability of payments relative to actual losses. These questions concern real institutional effects. A live test can address them if suitable controls and fallback procedures exist. Selection of the mechanism should follow the question requiring evidence rather than the attractiveness of the sandbox label.
Primary Institutional Documents and Comparative Research Design
Evidence architecture
Official announcements establish the institutional actions reported by the authority. They do not independently establish the social value of those actions. Registers establish recorded legal status, subject to differences between effective and posting dates. Summaries and inspected excerpts support only propositions contained in them. The journal's published abstract is considered together with its accessible working-paper version, which belongs to the same research program, not a separate replication.
The comparative dimensions are scheme and version, supervising authority, date, role of the participant, relevant activity, admission mechanism, permission, experimental constraints, exit process and recorded outcome. The full consolidated text of every applicable law has not been reviewed; legal findings describe the arrangements documented in the primary agency material. The study does not advise individuals on authorization or assert that a particular innovation qualifies in any jurisdiction.
EIOPA's European innovation facilitator inventory records 14 regulatory sandboxes distributed across 12 EEA countries as of October 2023. This dated regional count concerns facilities, not insurers, tests, permissions or customers. A country can contain multiple schemes with separate sectoral mandates, and facilities with cross-sector eligibility may have carried out no insurance experiment. A count of nominal program coverage is not evidence of actual use.
Comparative observations
| Jurisdiction and documentary period | Supervisory structure | Mechanism | Evidence boundary |
|---|---|---|---|
| United Kingdom, 2016 onward | FCA cross-sector service | Live testing, tailored authorization application, possible rule modification | Cross-sector admission and launch data |
| Hong Kong, 2017 launch | Insurance Authority | Authorized insurer pilots, separate digital insurer Fast Track | Historic design, not measured graduation rate |
| Brazil, 2020 framework and March 2026 snapshot | CNSP/SUSEP insurance arrangement | Temporary insurance authorization and conversion | Recorded insurer selections and definitive approvals |
| Italy, 2021 framework | Ministry, IVASS, Bank of Italy, CONSOB | Transitional supervised testing and limited derogation | Documented admissions design, not a matched outcome cohort |
| South Africa, 2020 Innovation Hub | IFWG multi-regulator cooperation | Guidance, sandbox and accelerator | Institutional structure, not insurance conversion data |
The FCA's live sandbox program accepts applications from regulated and prospective regulated firms as well as technology businesses across financial services. Relevant permissions remain required to perform regulated activities unless a lawful exemption applies. The authority identifies limits to its waiver powers. The purpose of comparison is thus to distinguish an application accepted for assistance from a legal authorization to perform a regulated activity.
Brazil's second-edition scheme index identifies CNSP Resolution 381 of March 4, 2020 and SUSEP Circular 598 of March 19, 2020 with their 2021 amendments, and lists the reporting resources for experimental insurers. These sources identify the historical legal lineage rather than establish the complete current consolidated law. The third-edition scheme page identifies the submission procedure and underlying instruments of that edition.
IVASS's admission instructions specify material novelty, benefits to users or market efficiency, operational readiness and economic sustainability; requests for derogations remain subject to primary legislation and European law. The Bank of Italy FAQ describes circumstances in which certain foreign and unregulated providers may seek admission. It would be erroneous to infer unrestricted cross-border authorization merely from the possibility of applying.
Institutional variation is intentionally selected here, rather than sampled randomly. The sample can support a comparison of mechanisms and identified events. It cannot rank the global prevalence of particular mechanisms or establish average treatment effects. Such estimates require a defined scheme population, a coding protocol that accounts for changed program versions, stable denominators, independent outcome measurement and explicit missing-data rules.
A source archive for future work should retain the effective date, original document, amendments, language and the translation methodology where needed. A program web page can be revised while the legal provisions affecting earlier experiments remain different. A dated institutional observation should therefore be retained as historical evidence even when the website subsequently changes.
Sandbox Entry
Identifying the activity and the applicant
The identity of the entity performing a regulated activity determines its obligations. A software provider may supply a claims system without assuming contractual payment risk. A broker may arrange insurance on another entity's behalf. A managing agent may exercise delegated underwriting powers subject to an insurer's authority. The legal risk carrier assumes payment obligations under the insurance contract. A business group can contain all four roles, making the financial relationships among its entities important to a test's design.
The applicant should establish which entity receives premiums, sets terms, collects information, makes underwriting decisions, holds reserves and is responsible for claims and complaints. A technology supplier's solvency and operational resilience matter differently from the capital adequacy of the insurer bearing risk. The supervisor's first task is therefore functional classification. Only after this step can the existing legal perimeter and any needed exceptional permission be identified.
Brazil provides specific documentary evidence of the distinction between selection and permission. The second-edition FAQ discusses corporate formation, management, a designated responsible director and provenance of invested resources before issuance of the temporary authorization. It states that activities reserved for participants cannot commence before the authorization. Selection for an experimental scheme is thus separate from effective legal permission. An outcome study should preserve each date and the intervening attrition.
A test application should specify the proposition, the exact regulatory question and what can be learned. For a novel distribution arrangement, the concern may involve clarity of disclosure, responsibility for customer money or suitability. For actual underwriting, the proposal should account for pricing, contractual limits, diversification, claims development and available financial resources. Innovation in a technical process should be tested against an appropriate baseline rather than presumed beneficial.
Administrative allocation and selection
Selection allocates scarce supervisory attention. Readiness screening can prevent unproductive tests and reduce customer risk. It can also select firms with greater managerial resources, better regulatory advisers or more established commercial partners. High post-test success may partially reflect this selection. A comparison between participants and all nonparticipants therefore does not automatically measure the effect of participation.
Consider two applicants, a well-capitalized incumbent testing automated underwriting and a small specialist working with an existing insurer to provide embedded protection. The first may be ready to begin with substantial data and governance resources. The second may reach new customers but require additional integration. Criteria emphasizing readiness may select the first. Criteria emphasizing prospective inclusion may favor the second. Transparent criteria permit an assessment of which objective actually determines access.
Cohort windows allow authorities to allocate specialist staff in advance and compare applicants within an intake period. Rolling windows reduce artificial waiting but can make admission rates sensitive to staff availability and pending applications. The SUSEP (2026) third-edition policy update reports the shift toward continuous proposals; the Italian admission instructions document announced application windows. These are distinct administrative choices, not proof that one system produces better market access. A study of administrative barriers would need time-to-decision data, application costs, withdrawal dates and applicant characteristics.
Permission and proportionality
A well-specified test permission identifies the applicable entity, activity, customer population, period, maximum exposure, conditions for changes, reporting and exit obligations. It identifies which requirements are modified and which remain applicable. A waiver of a supervisory rule has a different legal effect from a temporary insurance permission or exemption under primary legislation.
Proportionality in insurance must address the risk of contractual performance, not merely the scale of the technological change. A familiar digital interface can sell catastrophe coverage with substantial correlated loss potential. An unfamiliar claims interface can serve a simple low-limit product. Risks flow through the insurance activity, the financed exposures, contractual responsibilities and operational dependencies.
Eligibility should include a feasible testing plan and a means of distinguishing the effect of the innovation from the population in which it is introduced. A digital customer cohort may differ from an incumbent's existing book by geography, age or risk class. If the pilot cannot identify these differences, its findings are unlikely to establish a reliable counterfactual. At admission, the authority and participant should record the objective, baseline, observation horizon and limitations on extrapolation.
Supervised Pilots
The time structure of contractual risk
Insurance separates the time of contracting from loss, reporting, settlement and dispute. An experiment can expire while the obligation to process claims and pay covered losses continues. Test permission and underlying contractual liabilities cannot be treated as coextensive. An exit plan must account for coverage, claims administration, outstanding financial obligations, communication and legally effective portfolio arrangements.
The relevant ledger distinguishes written premiums, earned exposure, paid claims, estimated outstanding claims, unreported losses, expenses, reinsurance recoverables and the funds available for payments. Policyholders' claims may persist after a participant stops accepting new customers. Reinsurance can mitigate specified loss components but introduces contractual limitations, counterparty exposure and timing differences. A test should identify who is responsible for each obligation after an interruption.
A synthetic example makes the importance of exposure measurement explicit. Suppose a pilot covers 2,000 policies with one possible covered incident per policy and a maximum payment of USD 1,000 per incident. Assume expected claim frequency of 4%, average conditional severity of USD 400 and USD 30 gross written premium per policy. Aggregate expected claims equal 2,000 × 0.04 × 400 = USD 32,000; gross written premium equals USD 60,000.
| Hypothetical scenario | Frequency | Mean severity (USD) | Expected claims (USD) | Claims / written premium |
|---|---|---|---|---|
| Baseline | 4% | 400 | 32,000 | 53.3% |
| Frequency stress | 8% | 400 | 64,000 | 106.7% |
| Severity stress | 4% | 700 | 56,000 | 93.3% |
| Joint stress | 8% | 700 | 112,000 | 186.7% |
The ratio in the table is a synthetic quotient, not a standardized incurred loss ratio. The example omits expenses, uncertainty about parameters, taxes, timing and reserve development. It is not a capital estimate. Under the stated one-incident-per-policy condition, the theoretical covered payment cap is USD 2 million before other obligations. A common event affecting 200 policies at the full limit would require USD 200,000 of gross claim payments. This scenario illustrates potential accumulation; it supplies no estimate of event probability.
Even an experiment with a small premium limit can create obligations beyond the expected amount of claims. The appropriate risk controls depend on the contractual maximum, correlation, funding liquidity, reinsurance terms and uncertainty about actual experience. Separate monitoring of expected and stress outcomes is particularly important for novel products whose experience lacks mature data.
Consumer protection and operational resilience
Policyholders require intelligible information about coverage, exclusions, the identity of the insurer, claims procedures, and the consequences of an experimental arrangement. Consent to experimentation cannot be interpreted as blanket relinquishment of mandatory rights. A consumer may agree to a novel sales interface while reasonably expecting ordinary contractual performance. An experiment should include an accessible fallback route and effective means of making complaints.
Automated insurance decisions can generate errors through inappropriate variables, faulty integrations, drift or unusual input data. A system tested primarily on common claims can perform poorly on rare but valid ones. Aggregate accuracy can conceal unequal error rates in different customer populations. The appropriate evidence may include disaggregated measures, calibration, manual review, correction times, documentation of changes and routes to challenge decisions.
For parametric coverage, basis risk arises when the payment determined by an indicator differs from an individual's actual loss. An uneventful observation season cannot establish the quality of coverage under rare catastrophe conditions. Evaluation should therefore connect historical data, stress scenarios, customer understanding of the trigger and actual payment outcomes when available. Testing the reliability of the trigger is a different task from demonstrating that the resulting insurance contract is suitable for the intended population.
Technology providers create operational dependencies, including cloud services, identity verification and third-party data. An insurer should be able to recover relevant records and continue contractual performance if a supplier fails. Responsibility for reporting cyber incidents, restoring service and correcting policy or claim records must be established before actual customers become exposed.
Supervisory reporting and stopping
The Brazilian second-edition index includes reporting examples for premiums, claims, complaints, accounting and pending claims. These categories establish the importance of observing insurance operations as well as product distribution. A meaningful record must specify the denominator, event dates, scope, corrections, and completeness of data. Zero reported complaints may reflect inaccessible reporting rather than an absence of harm.
A test ledger should retain the identity and role of each participant, test version, policy inception and expiry, exposure, claim status, adverse operational events and amendments. Data minimization and confidentiality remain material considerations. Where aggregated reporting is sufficient, the authority may not require identifiable records; where an investigation demands transaction-level traceability, access should follow applicable law.
A stopping procedure should distinguish operational errors, deterioration in financial resources and serious customer harm. Quantitative thresholds need appropriate denominators, and qualitative triggers remain necessary for severe events that may occur only once. When an experiment changes in response to observed risks, the authority should record the new conditions and the reason. Otherwise subsequent observations may reflect changes to the original intervention rather than the original innovation.
Stopping new business does not eliminate the responsibility for existing coverage. A credible plan identifies the funding of run-off, administration of pending claims, outstanding complaints, communications and the retention of records. The decision to continue, modify, temporarily suspend or cease a test is a supervisory action with real consequences for parties beyond the participant.
Market Access Outcomes
The outcome sequence and Brazilian conversion
Graduation can mean completion of testing, recommendation for further authorization, application for ordinary permission, definitive authorization, continued partner-based access or commercial launch. These outcomes must be coded independently. Actual insurance market access in a legal sense concerns authorization of an entity to undertake a specified activity. Commercial viability and customer welfare are subsequent questions.
In February 2024, SUSEP's Darwin Seguros authorization announcement reported definitive nationwide authorization for damage and personal insurance in the S3 segment, identifying the entity as the first definitive authorization of the second edition. This establishes a completed licensing event. It does not determine additional protection, profitability or the effect of participation relative to ordinary licensing.
SUSEP's Circular 700 explanation (2024) describes a simplified procedural route for conversion from temporary to definitive authorization. The authority's March 2026 program snapshot reports twenty selected insurers across three editions, six with definitive authorizations and seven with conversions underway.
| Reported status, March 2026 | Count | Share of twenty selected |
|---|---|---|
| Definitive authorization | 6 | 30% |
| Conversion in progress | 7 | 35% |
| Outside the two reported categories | 7 | 35% |
The remaining category is calculated, not labeled by the authority. It cannot be classified as seven failures. If all seven pending cases were subsequently approved and the remaining seven never obtained permission, the eventual share would become 65%. This is a conditional arithmetic scenario, not an empirical forecast or a maximum possible share. Different entry times and pending decisions introduce right-censoring, and the denominator combines multiple generations of the scheme.
The participant status register documents separate temporary authorization dates, cancellations and extensions. Administrative cancellation is not itself a finding of insolvency, consumer loss or economic failure. A complete case-level study should link each status event to its official order, applicable effective date, and the treatment of liabilities. Neither a list of selections nor a separate status register can be added to a later administrative snapshot without matching entity identities and observation dates.
UK admission data and denominator discipline
The FCA's May 2026 FOI table permits calculation of descriptive acceptance-to-application ratios. Its population spans financial services and is not restricted to insurance.
| Period reported | Applications | Acceptances | Acceptance / application |
|---|---|---|---|
| July 2016 cohort | 69 | 24 | 34.8% |
| January 2017 cohort | 77 | 31 | 40.3% |
| Inception through December 2022 | 578 | 168 | 29.1% |
| FY 2024/25 | 52 | 16 | 30.8% |
| FY 2025/26 | 142 | 14 | 9.9% |
Ratios are calculated from the official table and rounded to one decimal place. The FCA notes balancing figures and pending decisions under rolling admission. It publishes the list of accepted firms once they begin testing, a stage distinct from the acceptance decision. Reporting periods can overlap, and the figures should not be added without reconstructing nonoverlapping observations. The apparent lower recent ratio cannot be interpreted as an estimate of increasingly strict selection without considering processing and applicant mix.
The first-year summary reported 50 supported firms from 146 applications for the first two cohorts. The later FOI table records 55 accepted applications in the two corresponding cohorts. These observations use different reporting dates and labels; the five-case difference cannot be assigned to a specific mechanism on the available evidence. A comparative dataset should preserve both instead of harmonizing them by assumption.
The FCA (2017) first-year outcomes announcement reported that approximately nine in ten first-cohort test completers progressed toward wider launch. At least two in five first-cohort testers received investment during or after the trial. Both are conditional, cross-sector outcomes. Neither is a definitive insurance authorization rate.
Comparing the UK application acceptance ratio directly with Brazil's selected-insurer conversion share confuses different stages and populations. More informative comparisons require comparable cohorts, dates and states, including applicants who never start testing, participants who withdraw and firms whose licensing decisions are pending.
Funding and causation
Cornelli et al. (2024), in Regulatory Sandboxes and Fintech Funding, report a 15% increase in capital raised following sandbox entry and a relative 50% increase in the probability of raising funds. The published abstract identifies staggered UK entry as the basis of the investigation. The accessible BIS working-paper version describes observations from 2014Q1 to 2019Q2 and comparisons using later-entering and matched firms. It explicitly recognizes limitations on evaluating consumer surplus and competition. The two versions are one research program. The estimate is a proportional increase, not a fifty-percentage-point change and not a treatment effect estimated specifically for insurance.
Participation may reduce regulatory uncertainty, convey information about applicant quality and generate evidence for prospective partners. In insurance, capital raised by an intermediary or software business is distinct from capital available to pay the insurer's contractual liabilities. Venture investment can improve market entry without directly improving risk-bearing capacity. Conversely, additional insurer capital can fund prudential resources without establishing customer benefit.
An insurance-specific evaluation should distinguish outcomes at the firm, product, policyholder and market levels. A staggered-entry design requires appropriately defined treatment dates, pre-entry observations, adequate comparators and checks for anticipatory changes. Rejected applicants might be informative, but they are selected through a process that can correlate with expected success. Differences in business maturity, access to capital, management expertise and partnerships require attention. A simple before-after comparison cannot isolate participation effects where market conditions and products change simultaneously.
Consumer welfare requires evidence about actual protection. Increased sales can reflect genuine inclusion or inappropriate distribution; lower premiums can reflect efficiency or inadequate risk pricing; rapid payouts can reflect better process or incomplete assessment of claims. Researchers should observe coverage quality, payment timing, complaints, continuation, price relative to comparable protection, and distributional effects. The outcomes cannot be represented adequately by startup survival alone.
Integrated Scientific Synthesis and Future Research
The allocation of supervisory discretion
Institutional comparison establishes that an insurance sandbox allocates regulatory attention and legal discretion under uncertainty. Entry allocates limited administrative resources; test design allocates obligations, constraints and observation responsibilities; exit determines whether permission becomes durable access or an orderly termination. This synthesis follows from the documented differences between institutional arrangements and does not claim a newly discovered theory.
The evidence gathered during a test should be linked to the scope of the resulting supervisory decision. Observing robust virtual onboarding may establish a proposition about distribution operations. It does not establish the solvency of a risk carrier operating a different underwriting book. A test of one policy cohort cannot establish customer comprehension in every market segment. The range of the finding follows the actual activity, participants and observation period.
A program can create perceptions of privileged access and official endorsement. Public criteria, documented reasons, availability of ordinary authorization and careful communications can mitigate those risks. Participation should be described with the actual legally authorized activity and its limitations. An entity may satisfy the criteria for testing without satisfying every criterion for unrestricted operation. Credibility depends on preserving these distinctions in both supervisory records and public communication.
Supervisory capacity has an opportunity cost. A program may consume specialist legal, actuarial and technical resources that could otherwise be used for ordinary supervision. A cost-effectiveness assessment should compare the work required for live testing with less expensive mechanisms such as guidance and technical prototypes. The benefit of an experiment also extends beyond participants when it generates publicly available knowledge, a clarification of ambiguous rules or a reduction in repeated administrative uncertainty across the market.
Cross-border coordination
A digital insurance proposition can operate through a shared technical architecture while relying on different local insurers, intermediaries and contractual forms. A cross-border test should identify customers' locations, legal risk carriers, allowed activities, data locations and responsible authorities. A coordinated experiment requires clarity about the parties responsible for claims and complaints in each relevant legal environment.
The GFIN cross-border testing program describes cooperation supporting sequential or simultaneous tests in multiple jurisdictions. Such cooperation can reduce duplicated inquiry and improve exchange of evidence. It does not itself constitute a common insurance license. The local ground for lawful insurance activity remains a necessary part of each test. The Italian FAQ's conditions for certain foreign applicants illustrate why the legal status of the participant and nature of the proposed service must be identified even in a shared regulatory environment.
International comparability can improve through common outcome definitions and transparent documentation without harmonizing every substantive law. A standardized record of selection, authorization, initiation, progression and cessation can support meaningful empirical comparison while preserving differences in national permissions.
Future evidence and research design
A future insurance-specific dataset should preserve the scheme version, legal entity, experimental activity, dates, permission conditions, observed exposure and outcomes. A stable identifier for the scheme, entity and test prevents double counting of related firms and repeated applications. Separate linked policy and claim identifiers should be protected according to applicable privacy law. This is a proposed measurement specification, not an asserted new conceptual system.
Admission data should record received applications, eligibility decisions, selections, effective permissions, test starts and withdrawals. Exit data should identify completion, ordinary authorization, partner-based access, cancellation, cessation and pending outcomes. Effective legal dates and publication dates should be distinct. A longitudinal conversion analysis may use competing-risk methods for authorization, withdrawal and cancellation, subject to explicit assumptions about censoring and loss to follow-up.
For consumer-level analysis the dataset should contain previously uninsured participation where lawfully measurable, retention, coverage quality, affordability relative to comparable risks, claim payment timeliness, unresolved disputes and accessible service. Exposure and claims development must be reconciled before any actuarial comparison. The analysis should consider adverse selection and shifts in applicant composition, economic conditions, catastrophe experience and changes in distribution that occur independently of sandbox participation.
A study of regulatory learning should trace each observed problem to an actual rule clarification, modification or administrative change. Temporal sequence alone does not establish causation because a regulator may have planned an amendment before the experiment began. Process tracing, dated decisions, consultation records and interviews can help identify the contribution of tested evidence.
The evidence agenda has four priorities: acquire a stable insurance-only population with full legal instrument versions and follow-up periods; link actual permissions to contractual exposures and service outcomes; compare participants with relevant nonparticipants; and quantify the supervisory costs and benefits relative to alternative instruments. This research would strengthen evaluation without imposing a false uniformity on the diverse institutional arrangements already observed.
Conclusion
Insurance regulatory sandboxes organize the entry and supervision of bounded innovations through a jurisdiction's allocation of legal authority. The comparative evidence identifies controlled testing by already authorized insurers, temporary permission for new insurance entities, coordinated supervisory programs and separate accelerated ordinary authorization. These mechanisms determine what admission means and which obligations persist throughout and beyond a test.
Insurance experimentation must be designed around the persistence of the contractual promise. Policyholder safeguards, measurement of exposure, financing of obligations, claims administration, operational dependencies and exit arrangements carry consequences outside the testing interval. The hypothetical stress calculation illustrates how frequency, severity and accumulation can affect potential claims even within a limited cohort. Actual prudential assessment requires real portfolio evidence and applicable capital and liability measures.
Brazil's March 2026 snapshot establishes actual definitive authorizations and pending conversion proceedings among selected insurers. UK cross-sector figures establish different administrative and market outcomes. Fintech funding research provides useful evidence about investment following participation and motivates further insurance-specific tests of regulatory and informational mechanisms. It does not establish a general causal impact on insurance protection, solvency or consumer welfare.
Institutionally credible experimentation connects a specified lawful permission, safeguards proportionate to the insurance activity, a transparent exit decision and an accountable record of what has been learned. Its lasting value rests in the quality of insurance protection enabled, the evidence generated for supervision, and the equitable treatment of market participants beyond the pilot.
References
IAIS. 2024. Insurance Core Principles and ComFrame. Updated December 2024. Basel: International Association of Insurance Supervisors. Source.
IAIS. 2018. Application Paper on the Use of Digital Technology in Inclusive Insurance. Approved November 7, 2018. Basel: IAIS. Source.
Financial Conduct Authority. Undated. Regulatory Sandbox. London: FCA. Accessed November 19, 2026. Source.
Financial Conduct Authority. 2017. Regulatory Sandbox Lessons Learned Report. October 20. London: FCA. Report summary. Source.
Financial Conduct Authority. 2017. Financial Conduct Authority Outlines Lessons Learned in Year One of Its Regulatory Sandbox. October 20. London: FCA. Source.
Financial Conduct Authority. 2026. Information on Number of Sandbox Applications Received from 2021–2025. May. London: FCA. Source.
Insurance Authority. 2017. Insurance Authority Introduces New Initiatives to Facilitate Insurtech in Hong Kong. September 29. Hong Kong: IA. Source.
SUSEP. 2022. Perguntas e Respostas sobre o Sandbox Regulatório – 2ª Edição. Rio de Janeiro: Superintendência de Seguros Privados. Source.
SUSEP. Undated. Sandbox Regulatório – 2ª Edição. Rio de Janeiro: SUSEP. Accessed November 19, 2026. Source.
SUSEP. 2024. Susep Concede Autorização Definitiva para Empresa do Sandbox Operar como Seguradora. February 8. Rio de Janeiro: SUSEP. Source.
SUSEP. 2024. Susep Publica Circular que Consolida e Simplifica Regras de Autorização para o Mercado Supervisionado. April. Rio de Janeiro: SUSEP. Source.
SUSEP. 2026. Susep Concede Autorização Definitiva para Mais uma Empresa do Programa Sandbox. March. Rio de Janeiro: SUSEP. Source.
SUSEP. 2026 access. Seguradoras Participantes do Sandbox. Rio de Janeiro: SUSEP. Accessed November 19, 2026. Source.
SUSEP. 2024. Sandbox Regulatório – 3ª Edição. Published July 25, 2024; modified October 28, 2025. Rio de Janeiro: SUSEP. Source.
IVASS. Undated. Regulatory Sandbox. Rome: Istituto per la Vigilanza sulle Assicurazioni. Accessed November 19, 2026. Source.
IVASS. Undated. Admission to the Regulatory Sandbox. Rome: IVASS. Accessed November 19, 2026. Source.
Bank of Italy. Undated. FAQ: Regulatory Sandbox. Rome: Banca d’Italia. Accessed November 19, 2026. Source.
European Insurance and Occupational Pensions Authority. Undated. Innovation Facilitators in the EU. Frankfurt: EIOPA. Accessed November 19, 2026. Source.
South African Reserve Bank. Undated. Fintech. Pretoria: SARB. Accessed November 19, 2026. Source.
Global Financial Innovation Network. Undated. Cross-Border Testing. Accessed November 19, 2026. Source.
Cornelli, Giulio, Sebastian Doerr, Leonardo Gambacorta, and Ouarda Merrouche. 2024. Regulatory Sandboxes and Fintech Funding: Evidence from the UK. Review of Finance 28 (1): 203–233. DOI: 10.1093/rof/rfad017. Source.
Cornelli, Giulio, Sebastian Doerr, Leonardo Gambacorta, and Ouarda Merrouche. BIS Working Papers No. 901. Regulatory Sandboxes and Fintech Funding: Evidence from the UK. Basel: BIS. Working-paper version accessed November 19, 2026. Source.
CGAP. 2020. How to Build a Regulatory Sandbox: A Practical Guide for Policy Makers. Washington, DC: Consultative Group to Assist the Poor. Source.