Verve Answers ESOMAR's 20 Questions for AI-Based Service Buyers
Verve is pleased to present its comprehensive responses.
A. COMPANY PROFILE
What experience and know-how does your company have in providing AI-based solutions for research?
Verve brings deep experience in delivering AI-based research solutions through Verve Vero, our proprietary AI platform that transforms human truth into superhuman advantage; our award-winning Verve Intelligent Personas & Simulations, recognized with AURA’s 2024 “Innovation of the Year”; and our 2025 ESOMAR Awards for “Excellence in AI & Automation” with Mars and “Outstanding Contribution to Consumer Insights” with Samsung.
Founded in 2008 by Andrew Cooper, who helped change the face of research as a co-founder of Research Now, the world’s largest online research panel at that time, Verve has consistently stood at the forefront of dynamic change in the research industry.
As pioneers of online communities at the advent of Web 2.0 and, for the last three years, as leaders in the development of powerful AI simulations of insight, our philosophy has always been simple: use technology to deepen empathy and human understanding, not replace it.
This core ethos manifests in Verve Vero, Verve’s AI platform. Vero unites high-quality proprietary datasets and real people with best-in-class generative AI and skilled consultants to deliver better, faster insights.
The Vero team’s Chief AI Architect, Jon Gumbrell, one of the original founders of Verve, brings over 20 years of ResTech expertise and invaluable insights from his extensive research industry experience, delivering cutting-edge software and experience platforms for market research agencies and clients.
The Vero team brings exceptional expertise in applying advanced research tools and methodologies to deliver secure, high-impact AI-empowered programs across multiple sectors. Our multidisciplinary team includes AI-native product engineers, senior product managers, and experts in computational and corpus linguistics, working alongside Verve’s Information Security Officer, who brings more than 24 years of IT and data experience and our AI-first insight consultants ensuring our solutions are both innovative and robust in today’s AI-driven research landscape.
Verve have shared expertise at events and workshops over the last three years, including practical training and events for the MRS and AURA on topics such as ‘How to Use AI for Better Insight’ and contributing to the MRS Delphi Report on AI, synthetic data, and artificial participants, Verve is helping clients navigate change in the new AI-powered world.
This expertise is further demonstrated through our partnership with CACI, one of the UK’s leading data, customer segmentation and profiling businesses, where we developed Fresco Live. Fresco Live is a real-time, interactive extension of CACI’s industry-standard Fresco segmentation, transforming static segments into dynamic Personas that enable organisations to explore behaviours, attitudes and scenarios instantly.
Where do you think AI-based services can have a positive impact for research? What features and benefits does AI bring, and what problems does it address?
AI is a game-changing technology that has already created new “rules of engagement” for how research is conducted. It is transforming the research industry in a similar way that the internet and online research did 20 years ago, but with a much faster rate of adoption and a more varied and dynamic range of applications.
Initially, for many people in the industry, AI was all about greater efficiency. Since then, the direction of travel has become unmistakable: new models are emerging that don’t just streamline research, but deliver genuinely better insight.
These new approaches are attracting new business start-ups. And those new entrants are, in turn, attracting private equity money, backing AI-led models and accelerating the rate of change across the industry. This influx of new methods, new players and new investment will undoubtedly challenge the established order, but it can also be argued to be a breath of fresh air: a new, disruptive, energy for our industry.
As AI moves beyond efficiency, ‘synthetic’ approaches have become far more visible across the insights landscape. We’re seeing AI moderated qual open up new ways to gather depth at scale. Alongside this, adoption of synthetic panels, synthetic datasets, digital twins and AI personas is accelerating – useful when traditional fieldwork is too slow, impractical or places too much burden on respondents.
But quality varies wildly. While these methods expand the ways we can model people and behaviour, none of them are magic. Many are difficult to audit and often amplify the weaknesses of the data they rely on.
This is where our industry remains essential: by bringing the cultural understanding, nuance and judgement needed to ensure these new systems reflect people rather than approximate them. Indeed. AI hasn’t removed the need to understand people; it has made that need sharper and more urgent, raising the bar for what ‘good’ now looks like.
At Verve, we are being bold and going all in on Intelligent Personas and Simulations. We see them becoming as integral to the new AI-powered insights model as surveys were to the old.
When built well – as we do! - simulations support a wide range of insight needs, from audience understanding to innovation workflows. Enterprise-grade simulations cannot be bought off the shelf because the devil is very much in the detail: that is in the quality of the data and the specialist craft required to build, validate and continually refine them.
When applied correctly, AI enables clients to reserve human participants for the most valuable work. Ultimately, this supports a more respectful and sustainable approach to insight; one that’s critical for addressing the growing data quality challenges visible across the industry.
3. What practical problems and issues have you encountered in the use and deployment of AI? What has worked well and how, and what has worked less well and why?
The ability for AI to generate seemingly compelling responses, as if it were representative of a customer segment or group, is both exciting and potentially dangerous for research.
There are therefore very real challenges in using ‘synthetic data’ alone or off-the-shelf general LLMs in a research context: ‘hallucinations’, seemingly plausible responses, unseen biases, over-simplification and black box syndrome. Used in the wrong way AI can be a very dangerous tool for decision making.
However, used in the right way, with sufficient human anchoring, AI is mightily powerful.
For this reason, at Verve we don’t see AI as something to use solely on its own - our insight is about the interplay of AI with human intelligence and cultural insight.
There are crucial steps we take to navigate the practical issues of AI which we discuss in some detail in this document: implementing guardrails, codifying best practice processes and basing our solutions on known, high-quality datasets, but none more so than the benefit of overlaying human and cultural expertise whenever we’re using AI for research purposes.
This focus on delivering an augmented approach, building transparent simulations using real, high-quality data from real people, helps to ensure our insight is credible, actionable, and reflective of the real world.
Ultimately, everything we do is anchored in human truth and can be verified back to humans, a critical element in building confidence when using AI-empowered insight for decision-making.
B. IS THE AI CAPABILITY/EXPLAINABLE AND FIT FOR PURPOSE?
Can you explain the role of AI in your service offer in simple, non-technical terms in a way that can be easily understood by researchers and stakeholders? What are the key functionalities?
First and foremost, we are research specialists. Our starting point is always to create relevant, robust and compelling insight that helps our clients make better decisions.
In developing solutions that combine human inputs with broader cultural forces, we use the power of AI to unlock vast data sets into powerful, accessible and better intelligence.
The power of AI supercharges the work we do, bringing quality, speed, cost and other advantages, but its application is always grounded in best practice processes and research principles that have informed Verve’s work since day one.
In essence, AI enables us to theme, summarize and make data accessible to clients in new, dynamic, intuitive and exciting ways.
At the heart of our offer is Verve Intelligent Personas & Simulations (also known as VIPS)
They’re intelligent dynamic simulations of different audiences, categories, markets, or behaviours trained on trusted, validated data. They combine real data, human insight and cultural context to explore scenarios, predict outcomes, and test ideas instantly.
They're always-on and accessible, whether you're using them in a live session, building a strategy, or plugging them into a broader workflow.
They support a wide range of insight needs, from audience understanding to innovation workflows
Every output is traceable, auditable, and designed for decision-making, delivering reliable, actionable answers that traditional research or static AI models cannot provide.
They’ve already been adopted by Global businesses and our work has been industry recognised with an ESOMAR award for Excellence in AI, as well as being named Innovation of the Year by AURA the UK’s leading client-side insight network.
What is the AI model used? Are your company’s AI solutions primarily developed internally, or do they integrate an existing AI system and/or involve a third party and if so, which?
We build custom, private simulations, combining high-quality proprietary client data with the generative capabilities of large language models and Verve’s own customized solutions and processes.
This enables us to be platform agnostic when it comes to the foundational technology and will use a variety of different LLMs and 3rd party solutions- selecting the right one for the specific project requirements and migrating as appropriate, as newer models or additional functionality becomes available.
Within the Vero platform, sits Verve Maven our machine-learning-powered engine. It intelligently analyses large volumes of diverse, unstructured data, social content, open ends, reviews, clustering and mapping it using proprietary analysis frameworks to generate insights and foster more informed decision-making processes. This analysis can often form the basis of our client simulations and solutions.
How do the algorithms deployed deliver the desired results? Can you summarise the underlying data and the way in which it interacts with the model to train your AI service?
As highlighted elsewhere in this document, our absolute focus is on building transparent AI simulations that use real, high-quality data from real people to ensure clients can be confident every insight is credible, actionable, and reflective of the real world.
We don’t just throw AI at the internet and hope for the best. We train AI on proprietary, high-quality datasets, ensuring the insights are grounded in real behaviour, not statistical guesswork.
The underlying data and the way it interacts with the AI to ensure it is verifiable, is therefore fundamental to our proposition:
Wherever possible, we utilize our clients’ proprietary high quality data sources, to provide proprietary competitive advantage. Not just access to the same insight as everyone else
We integrate carefully curated contextual datasets, layering in social trends and cultural artifacts reflect real-world trends.
We believe clients should know what goes into their insights. We don’t hide behind black-box solutions—we’re open about our methods and assumptions.
And at every step, we overlay human expertise, ensuring quality through using proven, unique frameworks of real-world testing, validation and human empathy to extract meaning.
C. IS THE AI CAPABILITY/SERVICE TRUSTWORTHY, ETHICAL and TRANSPARENT?
What are the processes to verify and validate the output for accuracy, and are they documented? How do you measure and assess validity? Is there a process to identify and handle cases where the system yields unreliable, skewed or biased results? Do you use any specific techniques to fine-tune the output? How do you ensure that the results generated are ‘fit for purpose’?
For us, everything starts and ends with data, combined with an expert human overlay.
VIPS are built to stay grounded in trusted data. Unlike generic AI tools, VIPS responses are generated within a structured environment that tightly controls extrapolation from the underlying dataset. The outputs are fully auditable and traceable back to the underlying insight set so we can check reasoning, limitations and share confidently.
As part of the build process, we undergo rigorous human-led validation process designed to test the performance of the simulation across three tenets:
Recall: how well Persona simulations retrieve facts and subtle details from real data
Prediction: how accurately they can infer and simulate insights from the data set they’re based on, whether attitudes, behaviours or reactions to concepts.
Creativity: how well they capture nuance and tone. Or how well they can expand your thinking
Prediction accuracy tests are conducted using hold-out data and blind tests. Typically, we utilise a range of complementary accuracy metrics to test quant predictivity. This gives a balanced view for everyday accuracy, spotting major deviations, and confirming that the underlying relationships in the data are being represented well. We set benchmarks to a degree of accuracy, appropriate to the project and the hold-out dataset
Results and future data are fed back into the system to improve accuracy and reduce risk of drift over time.
What are the limitations of your AI models and how do you mitigate them?
We fully understand that the generation of inaccurate or biased outputs by AI models can bring significant challenges and consequences for clients.
With VIPS, these risks are significantly lower. The simulations are built on specific proprietary data they are extrapolating from known knowns, and are not used outside of specific research use cases. This approach enhances the accuracy and reliability by design:
All content is rooted in real data from the input dataset, either directly or through logical extrapolation from known facts.
Extrapolation is controlled: Personas and Simulations can give realistic answers to related questions, but only within the boundaries we set.
No additional external knowledge is introduced: They don’t pull from the open web or wider AI model training.
Guardrails are in place: If the underlying corpus doesn’t support the query, the simulation will say so.
Human oversight ensures reliability: Responses are tested and reviewed to ensure realism and coherence.
During onboarding, we set custom instructions to prioritise specific data sources or weight inputs according to client preferences. That ensures the simulation reflects the hierarchy of evidence you trust most.
What considerations, if any, have you taken into account, to design your service with a duty of care to humans in mind?
Naturally, we have in place a robust Internal AI Usage and Guidelines Policy that the Verve team follows in the creation and implementation of all AI-empowered services.
The principles of this are built on ISO / IEC 42001 and it provides a framework that outlines guidelines for the ethical and responsible use of artificial intelligence (AI), with the primary objectives to mitigate potential risks and boost transparency and accountability.
Its scope extends to all Verve Team members, contractors, vendors and partner agencies who may utilize AI where there is a risk to Verve Partners and its clients.
The policy encompasses safeguarding privacy and personal data, using the technology for lawful and beneficial purposes, abstaining from deceptive or harmful applications, and fostering transparency and comprehension regarding the technology’s capabilities and constraints while also considering legal and regulatory requirements.
There is also clear duty of care for the human-generated data used within the models themselves:
The data used in the simulation does not include any Personally Identifiable Information, VIPS hold aggregated summarized information only
For any data provided to use for the purposes of building the VIP simulations we track back to the lawful basis of processing the data from a GDPR perspective.
VIPS environments are isolated, with cross-tenant training disabled and usage restricted to your deployment. Data isn’t used to train unrelated models or shared outside approved parties.
As a final, and very important, point, we believe our use of AI simulations also helps to take care of the real people (the humans!) who participate in research and are the lifeblood of our industry.
The research industry is facing a ‘respondent crisis’, with a reduction in the supply of good quality participants. This problem is unarguably augmented by respondent fatigue: we are all getting surveyed more, arguably less professionally, due to the advent of self-serve tools and customer experience platforms.
And many businesses still need to do research which is either dull or onerous for us as human being “respondents” to take part in. We aren’t going to change this easily and yet we need to find a way to ensure the business model is sustainable for participants.
AI provides one way to alleviate this challenge, through leveraging the power of existing data sources or, via simulations, that are happy to do the dull/onerous stuff all day and all night.
This enables us to save real people for more interesting, more engaging work where we need rich human response. This is particularly true in the B2B space where the respondents are, by definition, a small research universe.
D. HOW DO YOU PROVIDE HUMAN OVERSIGHT OF YOUR AI SYSTEM?
Transparency: How do you ensure that it is clear when AI technologies are being used in any part of the service?
For clients, the use of AI technologies, what we’re deploying and for what purpose, is clearly detailed within our project proposals and Statements of Work. The use of technologies and platforms are built into the Master Service Agreements we hold with our clients, as requested.
As noted above, we don’t hide behind black-box solutions – we’re open about our methods and assumptions. We believe clients should know what goes into their insights.
Any outputs from work that utilize AI, specifically the VIPS outputs, are always clearly declared as such. We also work closely with our clients to provide best-practice guides, prompt templates and training, to ensure they can get the most out of the solutions we deliver.
Clear messaging is used throughout any participant experiences where AI is being deployed, such as the use of conversational AI interviewing techniques in quantitative surveys.
Do you have ethical principles explicitly defined for your AI-driven solution, and how in practice does that help to determine the AI’s behavior? How do you ensure that human defined ethical principles are the governing force behind AI-driven solutions?
As noted, upfront in this document, ensuring that human principles act as a governing force in AI solutions is central to our entire ethos. The Vero platform is designed to anchor AI in human truth to enable better insight and decision-making.
This human oversight, built into every stage of our projects and set out in our AI Usage and Guidelines Policy, ensures that ethical principles are embedded into our solutions.
It is further formalized through Verve’s Internal AI Usage and Guidelines Policy – the principles of which are built on ISO / IEC 42001. The policy:
Establishes ethical and responsible AI use guidelines to reduce risk and promote transparency and accountability.
Applies to all Verve team members, contractors, vendors, and partners involved in AI use that could impact Verve or its clients.
Covers data privacy, lawful and beneficial use, avoidance of harm or deception, and clear communication of AI’s capabilities and limitations, aligned with legal standards.
Ensures ethical deployment of AI with consideration for team members, clients, and research participants; regularly reviewed to stay current with industry and regulatory changes.
Our human-driven approach creates a ‘Gold Standard’ for simulations and helps to ensure that our clients continue to be the experts and lead in an AI empowered world.
Responsible Innovation: How does your AI solution integrate human oversight to ensure ethical compliance?
Verve is built around the principle that great insight is found at the intersection of Artificial, Cultural and Human Intelligence. This mindset is embedded into all processes for the architecting and delivery of AI solutions: human oversight is built into every stage of our work.
We run VIPS under Verve’s Internal AI Usage & Guidelines Policy (ISO/IEC 42001–based) and our ISO 27001 ISMS
Teams complete governance steps such as lawful-basis review and DPIAs where appropriate, and document decisions for auditability
Senior researchers curate and approve inputs (removing PII, checking relevance/ representativeness/ bias), then validate outputs against known truths and client “truth sets,” with contested items escalated for review, correction and versioned release notes.
Guardrails (policy prompts, input filters, abstention rules and runtime safety checks) plus logging of login/API/model-config events provide traceability and enable targeted bias checks aligned to MRS and ICO guidance on fairness, explainability and non-discrimination.
E. WHAT ARE THE DATA GOVERNANCE PROTOCOLS?
Data quality: How do you assess if the training data used for AI models is accurate, complete, and relevant to the research objectives in the interests of reliable results and as required by some data privacy laws?
Once we receive the data, we follow a structured process to prepare and deploy it within Verve Intelligent Personas & Simulations (VIPS):
1. Define scope and governance: We clarify the intended use, review lawful basis, recording the lawful basis and purpose in Verve’s Records of Processing, and completing a Data Protection Impact Assessment (DPIA) where high-risk processing is likely, (including GDPR), and assess representativeness and provenance. The data is categorised into training, validation and test sets as appropriate.
2. Pre-processing and privacy protection: Inputs are cleaned, aggregated, and anonymised before use. We apply data minimisation principles and ensure no PII is included. Where needed, we re-balance or relabel to mitigate bias and flag any risks or anomalies.
3. Summarization and transformation: Raw data is summarised and structured into simulation-ready inputs. This includes aggregation, synthesis, and annotation, often supported by Maven, our machine-learning-powered engine. We then tokenise the content for compatibility with language models.
4. Validation and acceptance: We test the exactness of retrieval and assess generative performance against agreed KPIs. All key processing decisions, splits, data quality, bias mitigation, are documented to ensure transparency and traceability.
Data lineage: Do you document the origin and processing of training or input data, and are these sources made available?
Yes. For VIPS, we maintain full data lineage: we document each approved source, its provenance and lawful basis, the curation steps (e.g., minimisation/aggregation), and any transformations applied, with versioned updates and release notes.
Clients see and approve a transparent source list during build, and our traceable RAG pipeline links every output back to the specific passages used so users can audit “where it came from” and why.
We also record assumptions and decision logic in our build artefacts/validation packs and can provide the documented source list and processing notes on request. This approach aligns with MRS/ICO guidance to track provenance and development of datasets.
Please provide the link to your privacy notice (sometimes referred to as a privacy policy). If your company uses different privacy notices for different products or services, please provide an example relevant to the products or services covered in your response to this question.
You can access our Privacy and Acceptable Data Policies here.
What steps do you take to comply with data protection laws and implement measures to protect the privacy of research participants? Have you evaluated any risks to the individual as required by privacy legislation and ensured you have obtained consent for data processing where necessary or have another legal basis?
Verve designs VIPS to protect research participants’ privacy by default and to comply with data protection laws:
We build and operate VIPs on aggregated, anonymised research artefacts (no PII by default); where any Verve project ever includes personal data, we establish and document the lawful basis (typically consent or legitimate interests for research), carry out DPIAs where appropriate, and record purposes/decisions for auditability.
We track each approved source back to its lawful basis and apply data-minimisation and anonymisation before inclusion; clients review and approve the source list.
We operate under our ISO 27001 ISMS and Internal AI Usage & Guidelines Policy (ISO/IEC 42001–based), with segregated environments, least-privilege access and encryption.
Our approach aligns with ICO guidance on documenting lawful basis, transparency and consent requirements; participants’ consent is used where appropriate and must be specific, informed and withdrawable.
See our Commitment statement - GDPR Statement
What steps do you follow to ensure AI systems are resilient to adversarial attacks, noise and other potential disruptions? Which information security frameworks and standards do you use?
Our approach to VIPS:
Resilience & hardening: Private, per-client environments with network segmentation, WAF/firewalls, VPN/IP allow-listing/geo-blocking, least-privilege RBAC with MFA, and encryption in transit (TLS 1.2+) and at rest (AES-256). Continuous logging/alerting, tested backups/recovery, monthly vulnerability scans and annual penetration tests strengthen defence-in-depth.
Adversarial robustness (model layer): We constrain the simulations with policy prompts, input filters, abstention rules, and red-team tests; outputs are validated with ongoing monitoring and versioned fixes. These controls reduce attack surface (prompt injection/data exfiltration) and noise-driven errors.
Privacy attack posture: Because VIPs aren’t trained on personal data and use API-only inference, risks like model inversion/membership inference against PII are inherently minimised; we still follow ICO guidance on such attacks and apply data minimisation and access controls.
Operational safeguards: Event logs cover login success/failure, API calls and model-config changes; incidents are managed under our ISO 27001 ISMS. Client terms also prohibit introducing harmful code and PII into VIPs.
Frameworks & standards: ISO/IEC 27001 (certified since 2016) governs our ISMS and security controls; our Internal AI Usage & Guidelines Policy is aligned to ISO/IEC 42001; practices align with MRS/ICO guidance on fairness, explainability and security.
At Verve, we strive to deliver an incredible customer experience, whilst continuing to make additional required operational changes resulting from new legislation, and keeping our clients, partners and regulatory authorities informed.
Data ownership: Do you clearly define and communicate the ownership of data, including intellectual property rights and usage permissions?
Yes, all contracts and agreements clearly stipulate the ownership of the training data used in the Verve models and any derived IP. Naturally, this includes employing robust data security and staff training measures to protect proprietary data.
In short, clients naturally retain all rights in their Client Data, and all Output Data is theirs. Verve owns the platform IP and provides a licence to use Verve Deliverables for your internal business purposes. On expiry or termination, we will, on request, transfer all copies of all Client Data and Output Data we hold.
All this is done with the purpose of safeguarding our clients’ interests and ensuring they retain a competitive advantage and stronger market positioning.
Data sovereignty: Do you restrict what can be done with the data?
Yes, as noted in question 18, individual contracts and agreements are put in place with each client to restrict and protect the usage of all clients’ proprietary data only for the purposes agreed with each client.
We also provide contractual protection that client owned data (input and output) is not used to train the AI platforms during the creation and usage of the VIP models.
Each simulation is hosted on a private secure instance.
These are private instances which store all data. The data remains private, it is isolated from other instances.
Each instance has a unique set of application files and database and is not used by any other client.
All data transfers to and from the instance are secured using TLS encryption.
The simulations are built on a dedicated instance, so data can be removed completely and can be wiped out anytime if it is no longer required.
Each party may use the other’s confidential information only for the permitted purpose and not disclose it except as allowed.
Access is restricted through strict access controls which include strong password policies, Geo-blocking, Web Application Firewalls and network security groups. TLS encryption is enforced to ensure protection of any data transferred to or from the platform. Instances are regularly backed up to ensure protection against data loss.
Ownership: Are you clear about who owns the output?
Yes, all client data provided as input to a Verve built model is owned by the client, as is the output generated by the models.
The outputs from models built by Verve and licensed to the client for the duration of their use are intended for internal use solely by our clients, unless otherwise stated and agreed by the client.
We are committed to protecting the output for clients in the building of Verve models such as VIPs, through measures aiming to safeguard their interests and ensure they retain competitive advantage and control over the AI's outputs.
In Sum...
AI is undoubtedly a revolutionary force that is changing the research industry.
Our response to ESOMAR’s 20 questions underscores not only Verve’s ongoing commitment to stand as industry leaders as this space develops but also demonstrates our dedication to maintaining transparency and open dialogue as the industry continues to learn and develop.
AI will only keep getting better. Or, at the very least, become increasingly interesting. Make sure you’re on the journey.
Get in touch with Verve today to understand how we can help you navigate this new world and start to create better AI-driven insight for your business so that you have the power to bring your customers into every decision.