PUBLISHED DATE: 2026-11-02 04:00:51

A Joint Perspective from Publicis Sapient & AWS

The Emergence of Agentic Banking

Across financial services, 73% of executives say AI is critical to their future, but only a third report meaningful ROI. Most cite fragmented data, operational complexity, and poor system integration as the primary barriers to scaling AI.1

AWS sees this pattern across its 1,000+ financial services customers globally: organizations that treat AI primarily as a model problem, rather than a systems problem, often stall at the pilot stage.

The challenge is not only model capability; it is the lack of integrated decision systems.

Deploying intelligent agents across a bank is not the same as deploying automation. In financial services, systems are complex, and restrictions are more stringent. This makes the consequences of failure far more significant.

Without the right safeguards, even advanced AI can be difficult to trust in high-stakes decision environments.

This raises a critical question: What kind of enterprise architecture allows agentic systems to operate safely at scale?

Developed jointly by Publicis Sapient and AWS, this playbook explores how banks can move beyond isolated AI pilots to enterprise-scale impact by combining agentic systems with a shared context layer. It also shows how banks can support more consistent, explainable, and governed decision-making across the organization.

Agentic AI Requires a Different Enterprise Architecture

In most banks, decision-making context lives outside core systems in emails, committee discussions, and human interpretation.

At the same time, more than half of firms report that their data is fragmented, inconsistent, or not AI-ready, and nearly 40% struggle with poor system integration.1 The result is predictable: AI systems can access data, but not always the conditions, constraints, and reasoning that shape real decisions.

Without this context, AI can optimize outputs but often struggles to navigate the complexity of enterprise decision-making.

To move from pilots to enterprise value, banks must rethink how AI is designed and deployed.

Leading firms are already moving in this direction. Over 80% have implemented formal AI governance frameworks, including auditability, testing, risk monitoring, and human-in-the-loop controls.1

What differentiates these organizations is not oversight alone, but system design. Governance can be supported through architecture, helping make decisions traceable, explainable, and aligned with compliance requirements by design. AWS refers to this as “governance by construction,” a principle embedded in the AWS Well-Architected Framework’s Security and Operational Excellence pillars. In practice, this means controls can be automated and supported at the infrastructure layer, reducing reliance on manual review processes.

Five Shifts Required to Scale Agentic AI

  1. From Use Cases to Systems
    Individual use cases can prove value, but enterprise scale requires reusable systems and shared governance.
  2. From Models to Context
    Competitive advantage comes from enterprise context and system design, while services like Amazon Bedrock preserve model flexibility.
  3. From Pilots to Proof of Value
    While pilots may succeed in controlled environments, real value comes from operating in complexity.
  4. From Automation to Reimagination
    Agentic AI requires redesigning how decisions are made. Automating existing workflows is not enough.
  5. From Manual Review to Embedded Governance
    Governance should be embedded directly into systems, not treated only as an after-the-fact review.

Context as a Decision Memory Layer

Many enterprise architectures are missing a key layer: context. The Enterprise Context Graph fills that gap.

The Enterprise Context Graph does not replace systems of record. It sits alongside them, connecting data with the reasoning, constraints, and relationships behind each decision. In effect, it becomes a decision memory layer for the enterprise, capturing how the business operates in practice.

Unlike traditional knowledge graphs, which capture entities and static relationships, the Enterprise Context Graph also models the full decision lifecycle: linking actions to constraints, rationale, and outcomes as they evolve over time. Without this layer, AI operates with outdated inputs. With it, the enterprise gains something closer to a GPS, continuously adapting decisions based on real conditions and past outcomes.

At its core, the Enterprise Context Graph enables something most systems cannot do. It captures not just what happened, but how it happened and why. This enables AI to improve reasoning about decisions, not just detect patterns.

In practice, this is delivered through five key capabilities.

Together, these capabilities create a shared memory for the enterprise, enabling AI to operate with real decision context.

The Enterprise Context Graph: 5 Key Capabilities

  1. Shared Ontology
    Creates a common language across systems so data, policies, and decisions are interpreted consistently.
  2. Decision Context Capture
    Records decisions with their triggers, constraints, rationale, and outcomes, making them more explainable and reusable.
  3. Exception Memory
    Preserves policy overrides and edge cases so future decisions can learn from how exceptions were handled.
  4. Cross-Systems Relationship
    Connects systems without moving data, using federated query patterns and event-driven integration on AWS to show how information and actions flow across the enterprise.
  5. Time and Causality
    Tracks how decisions unfold over time, using temporal graph models on Amazon Neptune to trace outcomes back to their root causes.

How Agentic AI Uses Context to Assist Banking Decisions

With an Enterprise Context Graph, agents can operate within a structured decision environment rather than relying on isolated data.

This allows agents to:
Crucially, agents operate within defined boundaries, providing recommendations while preserving human oversight. On AWS, Amazon Bedrock Guardrails can help support these boundaries by applying content filters, denied-topic policies, and PII redaction at the model invocation layer. This approach is particularly important in banking, where accountability should not be delegated to a system without oversight.

Ending Repeat Debates in Investment Committees

A bank’s investment committee meets to review a proposed portfolio change. Past trades, outcomes, and performance are all available. But the real question quickly comes up: why did we make similar decisions before? That context is often buried in old emails, meeting notes, or remembered differently by each person in the room.

With an Enterprise Context Graph, AI agents can review comparable past decisions and show the conditions behind them. It explains what risks were considered, what constraints were in place, and how those decisions played out.

Instead of revisiting past debates, the team can anchor decisions in documented precedent and move forward more decisively.

This is powered by Amazon Bedrock AgentCore (agent orchestration and reasoning), Amazon Neptune (decision graph), and Amazon OpenSearch (contextual retrieval).

Beyond investment decisions, the same pattern applies across the enterprise.
In each case, the agent understands the decision environment, not just the data inputs.

However, without orchestration and governance, context cannot be applied consistently at scale.

Operationalizing Agentic Systems at Enterprise Scale

Capturing enterprise context is only the starting point. The real challenge is turning that context into consistent, governed decision-making across the enterprise. What separates pilot success from enterprise scale is not better models. It’s building the right system.

To operationalize agentic AI, banks need four core capabilities:
  1. Shared Context
    A unified understanding of decisions, constraints, and relationships across the enterprise (enabled by Amazon Neptune + Bedrock Knowledge Bases).
  2. Orchestration
    The ability to coordinate agents and workflows across systems, not just within a single use case (enabled by Amazon Bedrock AgentCore and AWS Step Functions).
  3. Embedded Governance
    Policies, risk thresholds, and controls enforced directly in the system, not reviewed after the fact (enabled by Amazon Bedrock Guardrails and AWS Config rules).
  4. Full Traceability
    Every recommendation and action is visible, explainable, and auditable (enabled by AWS CloudTrail, Amazon CloudWatch, and Amazon S3 audit logs).
Together, these capabilities create an environment where AI can operate more safely at scale.

Platforms such as Publicis Sapient’s Bodhi bring these elements together into a single operating layer. Bodhi provides banking-specific ontologies, pre-built decision templates, and FSI governance policies on top of AWS primitives like Amazon Bedrock, Neptune, and Step Functions. This helps organizations focus on outcomes rather than stitching systems together.

Enabling the Architecture with AWS

Agentic enterprise systems require more than intelligent models. They require a secure, scalable platform that can support reasoning, orchestration, governance, and observability in production.

That foundation is especially important in banking, where AI systems may interact with sensitive customer data, financial decisions, policy constraints, and regulatory requirements. LLMs can plan, reason, and work toward specific objectives, but they do not become trusted enterprise systems on their own. They need architecture that makes context accessible, decisions traceable, and controls enforceable.

AWS provides the cloud foundation for this operating model. It enables banks to run AI workloads at scale, support model flexibility, connect agents to enterprise knowledge, orchestrate workflows across systems, and monitor agent behavior over time.

In practice, this means different AWS services can support different parts of the agentic architecture:
When combined with Publicis Sapient’s Bodhi platform, AWS provides the secure cloud environment needed to scale these capabilities. Bodhi brings the enterprise context, agent orchestration, and banking-specific decision frameworks, while AWS supports the infrastructure, model access, workflow execution, and governance controls required for production deployment.

The result is an architecture designed to help AI operate with the reliability, control, and governance required for real-world banking systems without requiring banks to build this integration from scratch.

SCENARIO

SME Loan Approval with a Policy Exception

A relationship manager submits a $2M loan request for a long-standing small business client. The credit score falls slightly below threshold, but the collateral is strong. The industry is volatile, but the customer has a reliable history.

Normally, this would trigger either a rejection or escalation.

With Bodhi’s agent orchestration, this isn’t treated as a one-off exception. It becomes a governed decision workflow.

Without Context

The committee sees the financials, risk scores, and policy thresholds.

What they don’t see is how similar decisions were made before, under what conditions exceptions were considered acceptable, or how those decisions actually performed.

The result is familiar:

With Bodhi’s Enterprise Context and Agentic Support

With Bodhi’s Enterprise Context Graph in place, the system surfaces comparable past decisions.

Bodhi maps the request against ontology patterns, highlighting cases where strong collateral and long-term relationships led to successful outcomes, as well as cases where similar industry conditions resulted in defaults.

For each case, Bodhi provides decision context: what triggered the exception, what constraints were applied, who approved it, and what the outcome was.

A Bodhi risk agent identifies a relevant risk pattern: prior defaults were tied to industry volatility and overestimated collateral value.

Rather than forcing a binary decision, the system presents structured options—such as approving with reduced exposure and additional safeguards—each grounded in precedent and expected outcomes.

Governance by Design

Before the decision advances, governance controls can be applied automatically.

The system validates whether the exception is allowed, routes the decision to the appropriate authority, and logs the full decision context, including rationale and supporting precedent. On AWS, this governance flow can be supported through automated controls:
Governance becomes less dependent on separate review processes and more deeply embedded into the system itself.

Outcome

The committee approves the loan with conditions.

Not because the system made the decision, but because it provided the context needed to make a better one.

The impact is clear:
Governance becomes less dependent on separate review processes and more deeply embedded into the system itself.

Preparing Banks for the Agentic Enterprise

The shift to agentic AI requires a fundamental rethinking of how banking systems operate.

Organizations that succeed will not be those that simply automate existing workflows. They will be the ones that reimagine their systems end-to-end.
In the agentic enterprise, AI does not replace human decision-making. It strengthens it.

It provides the memory, context, and insight needed to make faster, more consistent decisions with greater confidence.

In an industry defined by trust, that may be the most important transformation of all.

Contributors:

Pinak Kiran Vedalankar, CTO Agentic AI Solutions EMEA & APAC, Publicis Sapient

David Murphy, Financial Services Lead, EMEA & APAC, Publicis Sapient

Prashant Yadav, Global Technology Lead, Amazon Web Services

Kiran Randhi, Principal Solutions Architect, Amazon Web Services

Footnotes:

  1. Publicis Sapient & ThoughtLab, The AI-Powered Investment Firm: An AI playbook for wealth and asset management firms in the agentic era, 2025. Based on a survey of 500 global investment firms.
  2. AWS Well-Architected Framework — Machine Learning Lens, 2024.
  3. AWS, Financial Services Industry on AWS: Compliance and Security, 2025.
Publicis Sapient is a technology company that provides enterprise AI platforms and services. With over 30 years of digital business transformation experience, we enable enterprise clients to transform how they operate and serve their customers, unlocking new value and enabling them to thrive in an AI-driven world. Our platforms use AI built off this deep enterprise context to help them modernize, build agentic solutions, and sustain their competitive advantage. The combination of our AI platforms and the expertise of our people enables us to deliver faster and more effective outcomes through solutions that are specific to the unique needs of our clients’ businesses, their industries and their customers. Publicis Sapient is the technology hub of Publicis Groupe, uniting 20,000 people worldwide.

For more information, visit publicissapient.com.

Amazon Web Services (AWS) is guided by customer obsession, pace of innovation, commitment to operational excellence, and long-term thinking. By democratizing technology for nearly two decades and making cloud computing and generative AI accessible to organizations of every size and industry, AWS has built one of the fastest-growing enterprise technology businesses in history. Millions of customers trust AWS to accelerate innovation, transform their businesses, and shape the future. With the most comprehensive AI capabilities and global infrastructure footprint, AWS empowers builders to turn big ideas into reality.

Learn more at aws.amazon.com and follow @AWSNewsroom.