Work Runs on Proximity
Arva AI operates with a flat structure where engineers, researchers, and commercial roles work in close contact. The team is based primarily in London, with one enterprise account executive role posted in New York. Eight salaried openings on Zero G Talent span research, engineering, product, data, and sales, a hiring mix roughly even across AI/ML, data, product, and sales that reflects a deliberate choice to keep every function senior enough to operate without supervision. Rhim Shah, who led anti-financial crime teams at Revolut before founding Arva in 2024, and Oli Wales set technical and commercial direction directly. No middle management layer exists.
| Role | Location | Salary Band |
|---|---|---|
| Head of Research | London | £130K–180K |
| Head of Product | London | £90K–130K |
| Senior Full-Stack Product Engineer | London | £70K–110K |
| AI Research Engineer | London | £80K–120K |
| Forward Deployed Engineer | London | £80K–110K |
| Data Associate (AI Labelling & Evaluation) | London | £30K–45K |
| Strategic Customer Success | London | £50K–75K |
| Enterprise Account Executive | New York | $150K–230K |
Arva's agents handle millions of compliance reviews each month, screening alerts, KYB processes, and transaction monitoring, automating up to 80% of manual review work. That volume creates a tight feedback cycle. The FairPlay partnership, announced September 2025, adds independent stress-testing of agentic systems against SR 11-7 and NIST AI RMF standards. SOC 2 Type II and ISO 42001 certifications mean every model change carries audit-trail requirements. Engineers write evaluation harnesses before features; they also write the regulator-ready evidence packs FairPlay's platform generates.
Decision-making authority sits with the people closest to execution. Shah said he learned at Revolut never to settle for less talent than you think you need. The strategic customer success role owns the post-sale relationship for financial institution clients, driving retention and feeding insight back to product. The enterprise account executive in New York closes deals requiring deep domain fluency. Both roles report to the founders.
Customers set the pace: banks and fintechs in the US, UK, and Canada that need agentic AI they can defend to examiners. Arva's Times Square billboard in September 2025 signaled external momentum. Internally the metric that matters is straight-through processing rate: 87% of KYC/KYB and 86% of transaction monitoring alerts resolved without human touch. Team off-sites and a four-weeks-remote-per-year policy exist alongside a default state of high-intensity collaboration in a room. Bi-annual salary reviews and yearly equity refreshes are the only formal people-ops cadence.
Three Commands, Not Aspirations
Three explicit operating principles appear on Arva AI's careers pages and in founder interviews, each phrased as a command: "Own With Urgency", action and rapid iteration beat polish; ship, learn, repeat. "Customer First, Team Second": create value for customers, and the team grows from it. "Transparency": open communication and visibility across every level. These are not values posters; they are decision rules used when choosing between speed and completeness, between a feature a prospect asked for and one the roadmap promised.
The principles trace to Shah's path. At Revolut Business he led the financial-crime product team and watched analysts in Warsaw open a document, check a name, cross-reference an ownership chart, flag a risk, close the file, and start again, tens of thousands of times daily. The work was "mundane and repetitive" in his description. He chose startup chaos over Google's corporate structure on purpose: weekly access to the CEO, real ownership, faster learning. That preference for ownership over hierarchy shows up in "Own With Urgency." The same instinct appeared earlier: at Oxford he launched Carbon Codes, a sustainable food marketplace, and sold it; earlier still he self-published university entrance guides after turning down £50 for the rights, by his telling earning 100 to 1,000 times more.
A fourth principle, never written on the board but repeated in Shah's interviews, is "Guardrails first." In a regulated field, auditability is the feature. Build the safety, then the speed. The company's thesis — "Fighting financial crime shouldn't be reactive or slow. It should be intelligent, contextual, and auditable" — puts auditability on equal footing with velocity. Most software trades rigor for velocity; in compliance a tool that is fast but sloppy manufactures false confidence at scale. Shah's wager: a well-built agent can be both quicker and more thorough than a tired human on hour seven of document review, precisely because it does not tire. That wager holds only if every decision an agent makes is explainable and traceable, because a regulator will eventually ask the bank to justify it. The audit trail is not a constraint bolted on at the end; it is the starting point.
The principle extends to model selection. Shah has said publicly that "the value isn't in AI itself, but how AI can be used", a framing that treats models as commodities and workflow integration as the product. When the team benchmarked Gemini against alternatives for their Screening AI agent, they chose it because it delivered five times faster response times, a 6% increase in alerts automatically closed with confidence, and clearer audit trails. The decision criterion was not raw benchmark scores; it was the combination of speed, confidence calibration, and traceability a compliance officer can defend.
"Customer First, Team Second" sounds harsh until you see the hiring mix. A bank does not have one compliance problem; it has a stack: Know Your Business, Know Your Customer, transaction monitoring, adverse-media screening, sanctions and PEP checks, ongoing monitoring. Arva frames itself as an AI workforce rather than a single tool because the customer's problem is a workforce problem. The principle means the roadmap is pulled from customer workflows, not pushed from research curiosity.
Transparency operates as a forcing function in a flat team. With no middle management layer, information asymmetry is a bug, not a feature. Shah has argued that UK investors need "a greater risk-taking culture" and that more startups should emerge from and receive backing within the UK, a view he shares openly. The same candor applies to the five-to-ten-year horizon he sets for a profitable, durable business aimed at what he calls a "$10 to $100 billion problem," said in a Bae HQ interview. The number is deliberately wide; the point is the timescale. Quick flips don't survive regulator scrutiny.
Together the principles form a coherent system: urgency without guardrails produces regulatory risk; customer focus without transparency produces feature bloat; ownership without a long horizon produces burnout. Early metrics — automating 80 to 92% of manual review work, cutting review times by a similar margin, onboarding customers ten times faster while strengthening controls — suggest the system holds. Whether it holds at scale, or when the first major model regression hits production, is the test the principles were built for.
What the Hiring Bar Reveals
The roles posted on Zero G Talent's board reveal a deliberate split: four technical positions spanning research, applied engineering, and forward-deployed work; two product roles; one data operations role; and two commercial roles in enterprise sales and strategic customer success. That balance mirrors the observation that the company selects for versatile, high-autonomy contributors who can operate across the full stack of a banking-focused AI product.
The technical bar is visible in the compensation bands. The Head of Research and AI Research Engineer roles signal that proprietary model development remains a core activity, not an outsourced capability. The Senior Full-Stack Product Engineer and Forward Deployed Engineer sit on either side of the product-customer boundary: one builds the platform, the other embeds with large banks to integrate AI agents into financial crime compliance and lending workflows. The forward-deployed role demands engineers who can translate research-grade output into production systems inside regulated environments.
On the commercial side, the Enterprise Account Executive and Strategic Customer Success roles point to a sales motion that is consultative and technical, not transactional. Selling AI agents to large financial institutions requires fluency in compliance frameworks, model risk management, and the procurement cycles of tier-one banks. The Head of Product sits at the intersection, translating regulatory constraints and customer feedback into a roadmap the research and engineering teams can execute against.
The Data Associate role rounds out the picture: even at this stage, Arva invests in dedicated evaluation infrastructure. That role exists because the product's value proposition, "more consistency, effectiveness, and control" for automated decisions, lives or dies on measurement rigor. Candidates who have built or operated eval pipelines for LLM-based systems, especially in regulated domains, will surface quickly.
The salary band ($58K–$232K, median $152K across eight salaried roles), Zero G Talent's board data shows, reflects this spread: research and forward-deployed engineering anchor the top, data operations the floor, with product and commercial roles clustered around the median. No junior generalist track exists. Every open role assumes the hire will own a meaningful surface area from day one, with minimal onboarding scaffolding. That is the hiring bar in practice: not a checklist of keywords, but a demonstrated ability to operate as a force multiplier in a team where the next hire may not arrive for months.
The Employee Voice Gap
Glassdoor lists two distinct employer pages: "Arva Intelligence" with five anonymous reviews and "ARVA" with four, both accessed in early 2026. The split suggests a possible rebrand or dual registration: Arva AI operated as Arva Intelligence before shortening the name, but the research provided does not include the text of those reviews, their star ratings, or their dates. Without the review content, any summary of specific praise or criticism would be fabrication.
What the research shows is a company barely two years old that has posted eight salaried roles on Zero G Talent since late 2024. The job postings act as a proxy for what the organization expects from employees, and by extension, what current employees likely experience. The only first-person account in the research comes from the founders. In the Ascendants profile (January 2026), CEO Rhim Shah and CTO Oli Wales frame the product as "giving people breathing room" and "letting analysts stop staring at spreadsheets." That rhetoric, reducing grunt work and elevating judgment, is the cultural pitch to both customers and candidates.
Whether the team lives it is what the missing Glassdoor reviews would answer. The research gap is real: no employee quotes, no attrition data, no tenure distribution. For a Y Combinator Summer 2024 company that raised a $3M seed led by Gradient Ventures in January 2025, the public record is unusually quiet. That silence could mean the team is too small to generate review volume, or that employees are bound by tight NDAs common in early-stage fintech AI, or simply that no one has bothered to write one yet. Until those reviews surface, the most grounded read is structural: the hiring mix, the salary spread, and the founder narrative all point to a high-autonomy, high-exposure environment where the gap between the 91% benchmark, FairPlay's press release found, and the remaining 9% is where the human work — and the culture — actually lives.
Who Thrives, Who Burns Out
Arva AI's team operates with a flat structure, direct access to founders Shah and Wales, and an explicit operating principle posted on the Head of Research careers page: "Deliver Value Fast — Speed starts with clarity. We first understand what value actually means, for the customer, the business, or the system, and then take the shortest credible path to delivering it." That principle is the lens through which every hiring decision gets filtered.
Who thrives
Founder-stage operators who have already chosen chaos. The team sits at the intersection of those functions, each represented in the current open roles. The spread — London and New York, technical and commercial, individual contributor and leadership — signals a team where everyone ships, talks to customers, and fills gaps without a ticket queue.
Engineers who treat compliance as a product problem, not a constraint. Building agents that autonomously resolve the vast majority of screening alerts, KYB/KYC processes, and transaction monitoring alerts requires ML rigor and product intuition. The Forward Deployed Engineer role explicitly sits at the customer interface. The AI Research Engineer is expected to "represent Arva's research externally where it serves the mission, through publications, talks and the wider AI and compliance community." People who need a spec handed to them will stall.
Commercial hires who speak the language of regulated buyers. Those roles target the same institutions, institutions that move slowly and audit thoroughly. The FairPlay Compliance partnership exists because trust is the product. Reps who have sold into compliance or risk functions, who understand SOC 2, model governance, and the difference between a pilot and a procurement cycle, will close. Those who haven't will spend quarters learning on the buyer's dime.
People who treat half a million in first-year revenue as a starting gun, not a finish line. The Ascendants profile (January 2026) notes roughly $500K ARR and millions of monthly reviews across multiple institutions. That is early traction in a market where sales cycles are long and the bar for deployment is regulatory approval. The Y Combinator badge (late 2024) and Gradient Ventures seed (January 2025) mean the next 18 months are about proving repeatability, not optimizing efficiency.
Who burns out
Specialists who expect a lane. With a small team covering research, engineering, product, data labeling, sales, and success, there is no "not my job." The Data Associate role exists because model quality depends on evaluation, and in a team this size, the person labeling data today may be designing the eval framework next quarter. Anyone who joined a "Head of" title expecting a department will find themselves writing SQL, debugging a deployment, or joining a customer call.
Candidates who need process to feel safe. The operating principle "shortest credible path" is a license to skip steps that don't add customer value. It is also a demand for judgment. In a regulated domain, the wrong shortcut creates liability. People who rely on approval chains, detailed requirements documents, or a QA gate will either slow the team down or ship something that fails a bank's model risk review.
Anyone optimizing for work-life balance at this stage. The Glassdoor footprint (nine reviews total across two pages) is too small to draw statistical conclusions. The Head of Research and Enterprise AE bands reflect market-rate cash for London and New York, but the equity upside is the real compensator, and equity only pays if the company scales past the compliance trust barrier.
People who want mentorship more than autonomy. Shah and Wales are operators (Shah from Revolut Business financial crime products, Wales from deep engineering), but they are the only two people who have built this specific company before. No VP of Engineering, no Sales Enablement, no People Ops exists. The mentorship available is "watch how we decide, then decide yourself." For a junior hire, that can be the best career accelerator on the market. For someone who needs structure to grow, it is a recipe for paralysis.
The litmus test
If you read the Head of Research posting, which leads with the "Deliver Value Fast" principle, and your first thought is "show me the roadmap," this is not the role. If your first thought is "I know exactly which three experiments to run this week," it might be.
Working in AI? Zero G Talent tracks the openings: see every open Arva AI role, browse AI jobs, the companies hiring, and the people building the field.