Home / Customers / Tier-1 Central American Bank
Case Study  /  Banking  /  Central America
IN PRODUCTION · ANONYMIZED AT CUSTOMER'S REQUEST

A Tier-1 financial group in Central America.

IDENTITY WITHHELD BY AGREEMENT

Three Saptiva Studio applications in production: KYC, conversational customer agents, AI-powered credit origination. Replacing eighteen months of stalled hyperscaler pilots.

Sector
Tier-1 Banking
Region
Central America
Apps in production
3 · and growing
Status
Active
001 / The Eighteen Months Before

The technology was ready. The governance was not.

Before Saptiva AI, this bank's AI initiatives followed a pattern familiar to anyone who has worked inside a regulated financial institution in the region: promising capability, experienced technology partner, clear business case — and a deployment that never cleared the risk committee.

The bank spent approximately eighteen months evaluating and attempting to operationalize AI with global hyperscaler platforms. Two pilots reached late-stage review. Both were held back by the same obstacle: the governance framework around the model — residency, audit trail, jurisdictional exposure — did not satisfy the bank's compliance committee.

The bank was not unusual in this. Across the region, the reason enterprise AI stalls is almost never the model. It is the layer around the model that regulators and risk committees require before authorizing production use.

002 / What's Running

Three Saptiva Studio applications, live.

Inside the first production deployment window, three applications moved into active use. Each is a Saptiva Studio application orchestrated through FrIdA under a single bank-authored compliance policy.

01
KYC document processing.
STUDIO · KYC · CENTRAL AMERICA
Identity document readers handling the onboarding document flow. Extraction, validation, and compliance checks running inside the bank's approved environment. Exception cases route to human reviewers with structured model reasoning attached. Replaced a manual intake workflow that had been a bottleneck for customer acquisition.
02
Conversational customer agents.
STUDIO · OPS · MULTI-CHANNEL
First-tier customer inquiries resolved automatically across voice and chat. Complex cases escalated to human agents with full conversation context. Compliance and disclosure requirements enforced at the policy layer. Every conversation logged and auditable to the bank's retention standard.
03
AI-powered credit origination.
STUDIO · CREDIT · HUMAN-IN-LOOP
Credit officer copilot that reads the full application, assembles supporting context, and produces a structured recommendation with reasoning attached. The credit decision stays with the human officer. The assembly work — not the judgment — is what the AI removes.
003 / The Shift

What changed between the eighteen months and the first production deployment.

The models the bank ultimately deployed are, by most measures, similar in capability to what the hyperscaler pilots had been running. The shift was not in the AI. It was in the layer around the AI.

Before · Hyperscaler pilots
  • Residency review never closed
  • Audit trail inadequate for regulator
  • Model hosted in foreign jurisdiction
  • Policy expressed in vendor-specific terms
  • Procurement negotiation dragged for quarters
  • Pilot → production handoff had no path
After · Saptiva AI
  • Residency enforced by architecture
  • Signed, immutable audit stream
  • Runs in customer-approved environment
  • Policy authored by the bank, applied by FrIdA
  • Forward Deployed Engineer on day one
  • Production by default, not as an afterthought

Each of these is a sentence. Assembled together, they are the reason enterprise AI projects either live or die in a regulated bank. The bank did not need a better model. It needed a layer around the model that could clear the committee.

004 / Outcomes

What production actually looks like.

01
Three applications passed risk committee review and are operating in production against live customer traffic.
02
KYC intake bottleneck is no longer the constraint on customer acquisition velocity.
03
First-tier customer operations are increasingly automated, with a tightening escalation curve to human agents.
04
The bank is evaluating full-tier call center replacement as the next scope expansion.
005 / In Their Words

From the bank.

Pending clearance from the bank's communications team
Statement from the bank's Chief Risk Officer and Head of Digital — to be added following internal clearance for external attribution.
TIER-1 CENTRAL AMERICAN BANK · 2026
006 / What's Next

The roadmap the bank is already on.

The deployments in production today are the beginning of the engagement, not the end of it. The next scope expansion being scoped includes broader customer operations coverage (full-tier call center replacement), additional document-processing workflows across lending lines, and internal knowledge copilots for the bank's product and compliance teams.

The pattern is consistent: each new application ships against the same platform, under the same policy, into the same audit surface. The bank is not managing a portfolio of AI tools. It is operating a single AI layer, with new capabilities added as Saptiva Studio applications governed by FrIdA.

007 / What This Page Does Not Say

What's the customer's to disclose, stays the customer's to disclose.

Confidentiality principle
Detail that belongs to the bank and its customers is not on this page — and will not be.
  • Bank's identity — named only in customer-approved communications
  • Transaction volumes, customer counts, deployment footprint metrics
  • Jurisdiction-specific compliance artifacts or regulator correspondence
  • Credit, KYC, or operational outcome metrics — those are the bank's to publish, at a time of the bank's choosing

If you are evaluating Saptiva AI and want to speak with a reference customer in this market, we can arrange an introduction under an appropriate NDA. The willingness to stay quiet in public is part of why named customers trust us to write about them at all.

008 / Get In Touch

Bring the pilot that stalled.

Eighteen months of hyperscaler pilot fatigue. Risk committee cycles that never close. KYC and credit origination projects that keep bouncing. If that's the shape of your current AI roadmap, a Forward Deployed Engineer will respond within 48 hours.

Request a demo