How do you evolve your fraud strategy in the AI age? (Featuring Coast)

Coast, a modern expense management platform for trade businesses, is always rethinking how they underwrite credit and detect fraud as both have become more complex. 

I recently had the chance to sit down with Chief Risk Officer Anurag Puranik, a long-time Inscribe partner and one of the sharpest voices in credit, fraud, and risk today. He was to share his thoughts on what AI fraud tactics he's seeing and how his team is adapating.

Document fraud in particular has changed quickly. Across our customer base, we’ve seen template-based and AI-generated documents more than double in 2025. Coast is seeing the same pattern. This conversation focused on what they’re observing, how they’ve adapted, and what’s next as we head into 2026.

How Coast approaches risk management 

Coast serves a unique segment: trade businesses with employees in the field who rely on corporate cards for fuel, tolls, parking, and materials. That creates real operational complexity, and fraud gets expensive fast. Because of this, Anurag oversees both risk and finance, a pairing he considers essential: one month of poor credit or fraud performance can transform an otherwise healthy P&L.

Coast’s philosophy is straightforward

• Catch fraud at onboarding.
  
• Distinguish early between first-party and third-party fraud
  
• Don’t rely on post-onboarding interventions unless absolutely necessary

If a fraudster enters as a “customer,” the risk shifts dramatically. False positives become more painful, customer experience is at stake, and regulators pay closer attention. Coast’s goal is simple: don’t let bad actors in at all.

How their fraud strategy has evolved in 2025

Historically, Coast used application intent data to sanity-check claims (for example, asking applicants how much they spend on fuel each month). Fraudsters struggled to answer specifics about industry-specific behaviors.

That signal has now evaporated.

Anurag shared a striking example: when Coast calls suspicious applicants, the conversations now sound polished and credible. His team tested their theory, and confirmed that fraudsters can generate full call scripts using ChatGPT in seconds. The same applies to website legitimacy: placeholder GoDaddy sites have been replaced by AI-generated sites with real images, professional copy, and convincing structure.

As Anurag put it: “It’s so easy to appear legitimate now. The old signals aren’t reliable anymore.”

Fraudsters are also creating templatized bank statements with line items designed to mimic the spending patterns Coast expects, down to SaaS tools like Samsara.

This has forced Coast to shift from surface-level checks to deeper metadata analysis, such as domain registration age, technology spend, and document provenance. Fraudsters may be clever, but they’re also cost-sensitive. If you raise the cost of committing fraud, they usually move on.

Catching fake documents before Inscribe

Before implementing Inscribe in 2022, Coast’s document review process was entirely manual:

• One to 1.5 hours per application
  
• Analysts comparing IDs to state-level template changes
  
• Bank statements parsed manually in Excel
  
• Scanned, out-of-order PDFs
  
• Long investigation loops

Document review was the slowest and most painful part of underwriting and it caused major friction in the customer journey. Because reviews were asynchronous, the team had to reconnect with the applicant afterward, often losing them in the process.

Why Coast choose Inscribe for document fraud detection

A few capabilities immediately changed Coast’s workflow:

1. Automated Transaction Parsing

Inscribe’s transaction categorization became essential. Coast can now accurately distinguish income from transfers, which is critical when assessing true revenue. A customer showing $1M in inflows but only $50K in true income is a completely different credit story. Automation here removed a huge manual burden.

2. Fraud Signals in Document Metadata

The ability to detect tampering, templates, and underlying document structure validated whether Coast could trust a bank statement at all. Fraud and credit analysis became unified.

3. Synchronous Review

What once took an hour now returns in under a minute. Coast can approve good customers instantly, while routing flagged cases to analysts with full context.

Together, these capabilities unlocked a fully automated path for 80–90% of applications.

How Coast uses Inscribe today

Coast has built a best-in-class orchestration flow using Alloy and Inscribe:

• Applicants receive instant prequalification.
  
• High-risk signals trigger document verification.
  
• Inscribe returns fraud, compliance, and credit insights within seconds.
  
• Good customers get approved automatically.
  
• Analysts review flagged cases using Inscribe’s detailed portal, which compresses a 1–2 hour investigation into 5–10 minutes.

The team still reviews edge cases manually, but the volume has dramatically decreased and operations stay unblocked even during spikes.

The gains are meaningful:

• 0–5 minutes to review flagged cases (down from 60–90 minutes)
  
• 80–90% of applications fully automated
  
• Strong protection from large-ticket fraud (“Just a handful of misses can cost six figures”)
  
• Same-day SLA for 90% of applications
  
• Retention of good customers through faster decisions
  

As Anurag highlighted, the cost of being wrong is enormous—and Inscribe helps them avoid those mistakes while preserving customer experience.

Planning for 2026: Adopting agentic AI with governance

As Coast plans for 2026, Anurag and his team are focused on how AI can strengthen fraud detection and credit decisioning while maintaining the oversight required in a regulated industry. To him, the opportunity is not to replace human judgment, but to use AI to complete the heavy analytical work so analysts can evaluate risk faster and more consistently.

This perspective comes from experience on both sides of the industry. Before joining Coast, Anurag spent more than a decade leading credit and fraud analytics at major banks, including KeyBank, PNC, and Discover. He has seen firsthand how regulatory expectations, legacy systems, and siloed data make it difficult for banks to adopt new technology quickly. He has also seen how fintechs can move faster, provided they maintain tight control and clear governance. His advice blends both worlds.

Coast is prioritizing four areas as they adopt agentic AI:

1. Expanding access to credit by lowering underwriting costs: Many small businesses are simply too expensive to underwrite manually. If AI can perform the initial diligence automatically, Coast can safely approve more of these customers without raising operational costs. Anurag views this as a major growth unlock, and a path to broader access to credit for underserved segments.
   
   
2. Building 360-degree fraud and credit agents: Coast is exploring agents that bring together signals across:
   
   1. website legitimacy
   2. document intelligence
   3. application intent data
   4. traffic and acquisition patterns
   5. bank transaction behavior

When combined, these inputs create a more complete risk picture than any single signal alone. Anurag believes this type of holistic evaluation will become the standard as fraud becomes harder to detect visually.

3. Automating transaction monitoring and SOP-driven reviews: Many downstream workflows follow predictable patterns. Agents can replicate standard operating procedures for disputes, transaction reviews, and fleet-level analysis. This reduces manual burden while giving analysts a clearer starting point. Human oversight remains, but the time spent per case drops significantly.
   
   
4. Human-in-the-loop for regulated or adverse decisions: This is where Anurag’s bank experience is most visible. Any decline, adverse action, or compliance-sensitive decision must involve a human reviewer. AI can gather evidence, run the analysis, and surface inconsistencies, but final judgment stays with an analyst. Coast’s goal for 2026 is to perfect three fully agentic use cases built around this model of responsible oversight.

AI adoption advice for banks and credit unions

During our conversation, he shared several insights for regulated lenders considering AI adoption:

• Start small, prove value, and scale from there.
• Keep humans in the loop for any decision that affects a customer negatively.
• Use AI to reduce friction and paperwork for good customers, not to automate declines.
• Lean on agents to unify signals from outdated systems, rather than trying to rebuild those systems all at once.
• Focus on use cases where AI can replicate an analyst’s workflow, not replace the analyst.

He believes the institutions that take this approach will be able to modernize safely while preserving trust and regulatory confidence.

Where risk teams go from here

Coast is a strong example of a company that pairs operational discipline with forward-looking technology. Anurag’s experience across both fintechs and large banks gives him a unique view into what the next phase of fraud and credit decisioning will require. Fraudsters are moving faster, AI is lowering the cost of looking legitimate, and traditional signals are losing reliability.
But the answer is not to remove humans from the process. It is to give them stronger tools.

The real opportunity, as Anurag put it, is in using AI to handle the analytical heavy lifting so analysts can focus on judgment, context, and compliance. Agentic systems can surface risk faster, unify fragmented signals, and reduce the cost of underwriting. Humans remain responsible for the decisions that matter most.