SentiLink Pays Up to $350k for Fraud‑Analytics Talent in Six New Verticals

Why the Screen Exists

The fraud platform that verifies more than three million identities a day, as SentiLink's About page reports, is now adding people faster than it adds features, and the two problems are the same problem. Volume doubled in the first two months of 2025, Forbes reported in February, after the company added 50 customers in 2024. That surge sits on a base that already covers 13 of the 15 largest U.S. banks, six of the top 10 credit unions, and two of the three biggest telecom carriers. The client count exceeds 350 institutions and government agencies, with the company pushing into tenant screening, cryptocurrency compliance, healthcare eligibility, and state-and-local government contracts, verticals that each speak a different risk language.

Zero G Talent's board lists 31 open roles as of this week, three posted in the last seven days. The newest listings — Head of Crypto GTM, Enterprise Account Executives for insurance, crypto, and healthcare, plus paired SLED account executives for East and West coasts — map directly to the verticals the company named in its expansion remarks.

Naftali Harris and Max Blumenfeld founded SentiLink in 2017 after meeting as University of Chicago undergrads and working together at Affirm. They have steered the company from a synthetic-fraud niche to a broader identity-risk platform. But the capital event that matters now is operational: 30,000 fraud attempts blocked daily, a scoring engine that runs zero to 1,000 on both identity theft and synthetic fraud, and a 2024 product launch targeting assumed-identity abuse, fraudsters exploiting real identities of people who entered the U.S. on temporary visas and left.

Each new vertical brings a distinct compliance regime, a distinct buyer, and a distinct fraud vector. The platform's core scores travel across all of them, but the go-to-market motion does not. That is why the newest roles are titled by vertical, not by function.

The strategic intent is visible in the sequence: product expansion creates volume, volume demands vertical specialization, specialization demands headcount. The screen that follows — the one every candidate now hits — was built to filter for the people who can operate at that intersection. SentiLink has tightened its initial screen around two competencies: fraud-analytics fluency and collaborative problem-solving. Every hire, regardless of role, must clear that bar.

What Every Hire Must Know

Despite a spread of 31 open roles across go-to-market leadership, enterprise sales, and a salary band running from $90,000 to $350,000, the company screens every candidate against those same two competencies. The requirement shows up in the product itself. The platform verifies that volume daily, blocks roughly 30,000 fraud attempts, and scores each identity on a zero-to-1,000 scale for both identity-theft risk and synthetic-fraud risk. Every role — whether it's a Head of Crypto GTM at the top of the band or an account executive selling into state and local government — touches that engine or the customer conversations around it.

Fraud-analytics fluency means more than knowing what synthetic identity fraud is. The company was founded after its founders encountered twelve credit applications with the same name and date of birth but twelve different Social Security numbers, all returning 700-plus credit scores. That origin story still shapes the bar. Candidates must understand how synthetic fraud differs from traditional identity theft, why a simple yes/no flag fails, and how SentiLink came to label 24 distinct synthetic-identity subtypes. They need to grasp the scoring model well enough to explain score differentials to a risk officer at a top-15 bank. The research team, led by Head of Fraud Insights David Maimon, operates on dark-net markets, Telegram channels, and shell-address networks to map criminal infrastructure that moves across SNAP, Medicare, Medicaid, Federal Student Aid, tax refunds, and SBA loans. That intelligence feeds the models. A sales hire who cannot trace that line from darknet signal to API score will not clear the screen.

Collaborative problem-solving is not a soft-skill slogan here. It is operationalized every Thursday when the entire company spends an hour reviewing live cases together. The practice dates to the Affirm playbook the founders brought with them: a Risk Operations team manually reviews a daily subset of applications, hunting new fraud vectors and validating model performance. The weekly case review extends that discipline to engineering, product, sales, and marketing. A candidate who treats collaboration as a meeting cadence rather than a forensic method will not fit. The company's 2024 launch of assumed-identity-abuse detection — catching fraudsters who exploit those identities — came from that cross-functional loop. Analysts spotted the pattern, engineers built the detector, and partner-success teams led by Allison Lemaster translated it for customers before the tactic scaled.

Technical literacy rounds out the baseline. The platform leans on licensed and proprietary data fused with machine-learning models that update continuously. Non-technical hires do not need to train models, but they must speak the language of feature drift, label quality, and the eCBSV (electronic Consent Based SSN Verification) pipeline that SentiLink was first to operationalize. The board's concentration of enterprise and SLED account executives — six of the newest listings — signals that commercial hires now carry the same expectation. They sell to the aforementioned client base. Those buyers ask about model governance, false-positive rates, and how the 24 synthetic subtypes map to their loss columns. An account executive who cannot hold that conversation stalls the deal.

Adaptability closes the set. Fraud tactics shift weekly. Fourth of July 2024 data showed total card applications down 17 percent while high-risk applications jumped 92 percent, a holiday anomaly that reversed the usual pattern. The team caught it because the Thursday review habit forces constant re-calibration. Candidates who demonstrate a history of retraining their own mental models when the data turns — not just when the roadmap says so — match the culture. The screen filters for that signal before the first interview.

How the Screen Works

SentiLink's screening process reflects the same rigor the company applies to synthetic-fraud detection. The firm evaluates roughly three million identity checks daily across 350 clients, and that operational tempo shapes how it filters candidates. Recruiters and hiring managers look first for evidence that an applicant can operate inside a system where every signal matters and false positives carry real cost.

The initial resume scan weighs three dimensions. First, fraud-analytics fluency: experience with identity-risk scoring, synthetic-fraud typologies, or the eCBSV verification pathway SentiLink pioneered. The company's risk analysts label 24 distinct synthetic-identity categories rather than a binary yes/no, and the team that built that taxonomy expects candidates to speak its language. Second, cross-functional collaboration: the Thursday case-review hour — where the entire company spends sixty minutes dissecting live fraud patterns — means engineers, analysts, and product managers must translate technical findings into actionable intelligence for each other. Third, scale awareness: processing volume doubled in that period alone, so resumes that show comfort with high-throughput, low-latency data pipelines rise faster.

Candidates who clear the resume review face a structured assessment. Technical roles receive a take-home exercise modeled on SentiLink's Intercept investigation platform: a synthetic dataset containing mixed fraud signals (assumed-identity abuse, first-party fraud, PII anomalies) and a prompt to design a scoring approach that balances detection rate against false-positive cost. The evaluation rubric mirrors the company's production scores (Synthetic Fraud Score, ID Theft Score, First Party Fraud Score, PII Risk Scores), each calibrated 1–999. Non-technical roles complete a written case analysis drawn from the Thursday review format: a redacted fraud cluster, a set of enriched attributes, and a request to recommend rule changes for a specific client segment, say, a top-10 credit union onboarding flow versus a SLED agency verifying benefit applicants.

Hiring managers then conduct a 45-minute "signal call", not a culture-fit chat but a working session. The interviewer shares a recent fraud trend (for example, the 92% surge in high-risk card applications while total applications fell 17%) and asks the candidate to walk through hypothesis generation, data pulls, and stakeholder communication. The goal is to see whether the applicant thinks in the same loop the risk-analyst team uses: observe, label, model, deploy, monitor, repeat.

Reference checks focus on a single question: "Describe a time this person changed a fraud model or process based on a false-positive review." The answer reveals whether the candidate treats errors as noise or as the primary training signal, a distinction that separates SentiLink's approach from vendors who still sell binary risk scores.

The screen is narrow by design. With 31 open roles spanning enterprise sales, crypto GTM, and SLED account management, the company cannot afford to onboard people who need months to internalize its fraud ontology. The Thursday case review is not a ritual; it is the operating system. Candidates who cannot contribute to it on day one rarely make it past the signal call.

Preparing for the Test

SentiLink's screen is built around the work its platform actually does: stopping synthetic identity fraud at the application stage for 350 financial-institution clients. The company's public positioning, "combines technology and expertise," signals that the initial evaluation weighs applied modeling judgment as heavily as raw coding speed. Candidates who clear the resume scan typically show they have wrestled with the same structural problems that define identity-risk decisioning: extreme class imbalance, adversarial feature drift, and the cost asymmetry of false negatives versus false positives.

A modeling workflow captured in public research digests mirrors SentiLink's domain. Fraud is a rare event; the positive-to-negative ratio is "very very low," so a naïve model that predicts "not fraud" everywhere achieves high accuracy while missing every attack. The speaker's first move is deliberate oversampling of fraudulent transactions, a practical step SentiLink engineers repeat every retraining cycle. Applicants should be ready to explain why they would oversample, how they would validate that the oversampled distribution still generalizes, and what alternative strategies (weighted loss, focal loss, anomaly-detection framing) they would consider when the fraud rate shifts after a new synthetic-identity ring emerges.

Feature engineering is the next discriminator. The transcript breaks factors into seller-based, listing-based, and transaction-based buckets. Translate that to identity risk: applicant-supplied PII (name, SSN, DOB, address) maps to "seller-based" signals; the internal consistency of those fields (does the SSN issuance year match the stated age? does the address history align with credit-header data?) maps to "listing-based" consistency checks (the speaker's binary feature for title-description word overlap is a direct analog); velocity, device fingerprint, and linkage across applications map to "transaction-based" factors. Candidates who can articulate a feature set spanning all three layers, and who can describe encoding categorical PII (the transcript notes decision trees cannot ingest raw strings) without leaking future information, demonstrate the fluency the screen targets.

Model selection and validation rigor separate senior contributors from junior ones. The transcript walks through ROC and precision-recall curves, AUC comparison, and, critically, historical backtesting on held-out data to set decision thresholds. SentiLink's production environment demands the same discipline: a model that looks strong on a static test set but degrades when a new synthetic-identity pattern appears is a liability. Applicants should cite concrete examples where they used time-aware cross-validation, monitored PSI (population stability index) on key features, or built a champion-challenger framework that retires models before drift causes losses.

Network-level signals are a SentiLink specialty. The transcript highlights device IDs and IP addresses as a graph: "if a particular email address has been flagged but it's connected to an IP address then maybe you could draw an inference that all other email addresses connected to this IP address should be suspected." That is exactly the linkage analysis SentiLink's platform automates at scale. Candidates who have built or queried graph-based fraud features, such as shared device clusters, SSN-to-address multiplicity, and phone-carrier risk tiers, should surface those projects. Even a well-scoped side project that ingests public breach data to map identity-element reuse carries weight.

Cost-aware decision making closes the loop. The transcript's future-looking note, "if we have some a priori quantification of what each fraud transaction costs in the long term... we may be able to use that to determine at what point we wish to determine the transaction," reflects the business reality SentiLink's clients face: every false positive is a rejected legitimate customer; every false negative is a loss that compounds through money-movement channels. Applicants who frame threshold selection in terms of expected cost (customer lifetime value, operational review cost, regulatory exposure) rather than F1 score alone speak the language of the product team they would join.

Underpinning all of this is the critical-thinking foundation a Wired piece emphasizes: "not accepting the immediately visible logic," checking whether a valid argument is also sound (premises true in the current fraud environment), and upgrading judgment through continual self-assessment. In practice, that means a candidate can walk an interviewer through a past model failure, isolate the flawed premise (e.g., "I assumed SSN randomization eliminated geographic signal"), and describe the concrete retraining or feature change that fixed it.

The board data shows SentiLink adding roles across crypto GTM, insurance, healthcare, and SLED verticals, all segments where synthetic identity risk manifests differently. A candidate who prepares vertical-specific talking points (crypto on-ramp KYC failure modes, medical-identity theft vectors, state-benefit fraud rings) and ties each to the modeling toolkit above will clear the initial screen faster than one who treats fraud detection as a generic classification task. The screen is tight because the problem is specific; preparation that mirrors that specificity wins.

Ripples in the Talent Market

SentiLink's hiring velocity, 31 open roles on Zero G Talent's board as of this week, with three added in the past seven days, is pulling specialized fraud talent off a market that was already tight. The company's salary band, the previously mentioned salary band, sits at the upper end for risk-analytics positions and signals what it takes to attract candidates who can operate across synthetic-fraud detection, identity-theft scoring, and the emerging assumed-identity-abuse vector SentiLink began tracking in 2024. Those six new roles, Head of Crypto GTM, two SLED account executives, and enterprise reps for insurance, healthcare, and crypto, all carry $275,000–$350,000 ranges, a clear indicator that go-to-market roles now demand the same fraud-fluency the engineering side requires.

The talent drain is most visible in the Bay Area, where SentiLink is headquartered, but the remote-friendly posture on several listings extends the reach nationally. Candidates with direct experience at the aforementioned institutions that already use SentiLink's platform have a demonstrable edge; they understand the three-million-identities-a-day throughput and the 30,000 daily blocks the system generates. That operational familiarity cuts onboarding time for a company that doubled its daily check volume in early 2025 and added 50 customers in 2024. For smaller fraud-prevention vendors, losing mid-level analysts to SentiLink's "deep understanding" culture, where the entire company reviews live cases every Thursday, means backfilling roles with less experienced hires or contracting out to consulting shops that bill at a premium.

Competitors in the identity-verification stack, companies selling document-centric onboarding, device fingerprinting, or credit-bureau overlays, are responding by emphasizing their own data moats and by raising compensation bands to match. Several have introduced "fraud-analyst-in-residence" programs that mirror SentiLink's risk-operations model, though few replicate the weekly all-hands case review that CEO Naftali Harris and COO Maxwell Blumenfeld institutionalized from their Affirm days. The market is also seeing more acqui-hire activity: a handful of early-stage fraud-signal startups have been absorbed in the past 18 months primarily for their analyst teams, not their IP. That pattern suggests the scarce resource isn't modeling technique, gradient-boosted trees and graph embeddings are commoditized, but the institutional memory of how synthetic rings actually operate across SNAP, Medicare, SBA loans, and the telecom port-out scams Dr. David Maimon testified about before Congress in July 2026.

Conversely, analysts who have worked eCBSV queries, investigated assumed-identity-abuse patterns tied to temporary-visa holders, or mapped the credential-theft supply chains behind airline-booking scams, the exact vectors SentiLink's Head of Fraud Insights has been public about, are fielding multiple offers with signing bonuses that weren't common in this niche two years ago. The talent market isn't just tightening; it's re-pricing around a very specific kind of operational fluency that SentiLink's growth has made the new benchmark.

Where the Hiring Leads

SentiLink's hiring trajectory maps directly to a product roadmap that has moved well beyond its synthetic-fraud origins. The company now processes more than three million identity checks a day, a figure that doubled in early 2025, and blocks about that many fraud attempts daily across 350 clients that include the aforementioned clients. That volume creates a feedback loop: every verified identity sharpens the models, and every blocked attempt surfaces a new vector the risk-operations team can codify. The Thursday case-review ritual, as previously described, only scales if the headcount grows in step with the data firehose.

The roadmap's next lanes are already visible in the roles SentiLink is filling today. These are not generic sales hires; they are vertical-specific go-to-market leads for sectors the company has explicitly named as expansion targets. Forbes reported in February 2025 that SentiLink is "expanding further into sectors like telecommunications and tenant screening," and the SLED roles confirm a parallel push into government programs where the Head of Fraud Insights, Dr. David Maimon, testified before Congress in July 2026 that "fraud against government programs isn't a series of isolated schemes, it's a durable the previously described criminal infrastructure."

Product development is keeping pace. In 2024 the company shipped a detector for assumed-identity abuse, a vector where fraudsters exploit those identities, and it continues to extend Intercept, the investigative workspace that bundles identity intelligence, fraud signals, and analyst workflows, and Facets, the signal-bundle API for custom model integration. Each new score or module requires risk analysts who can label edge cases, data scientists who can retrain models without degrading latency, and partner-success managers who can translate the output into a bank's decisioning logic. The board's 31 open roles span all three disciplines.

Revenue implications are concrete. At a $430 million valuation (PitchBook, July 2021) on $85 million raised, as SentiLink's LinkedIn page shows, SentiLink is operating with the capital to fund a multi-year hiring plan, but the unit economics depend on converting the 50 new customers it added in 2024 into recurring volume. The doubling of daily checks in early 2025 suggests that conversion is happening. The next inflection point will be whether the vertical sales leads, crypto, healthcare, insurance, SLED, can replicate the financial-services playbook in markets where the fraud patterns differ (mechanic's-lien fraud in auto, travel-booking scams in airlines, financial-aid fraud in higher education) and the buying cycles are longer. The company's culture of weekly case review, which worked at 50 people, will be tested at 200; the hiring bar for collaborative problem-solving, described in earlier sections, is effectively a scaling mechanism for that ritual.

If the vertical leads hit their targets, the 350-client base could approach 500 within 18 months, with a higher share of non-bank revenue. That would diversify the concentration risk inherent in a client list dominated by large banks and give the product team more varied fraud signals to train on, completing the loop between hiring, product velocity, and revenue growth. The screen that filters for fraud fluency and forensic collaboration today is the same mechanism that will decide whether the loop holds at scale.

Working in frontier tech? Zero G Talent tracks the openings: see every open SentiLink role, browse frontier tech jobs, the companies hiring, and the people building the field.