Scaling Finance Operations with Agentic AI in a High-Growth EV Manufacturer

lucid_motors · Automotive · 2025 · source Agentic: yes · promotional bias: high · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: finance operations in high-growth manufacturing

Lucid Motors needed finance to support rapid EV growth with faster forecasting, analytics, treasury, accounting, and audit workflows.

Constraints:

• The finance organization had to prepare for growth around the launch of a mid-size vehicle platform.
• Data lived across SAP, Redshift, Salesforce, and other company systems.
• Multi-agent workflows needed guardrails, policy adherence, and cross-agent coordination.

Architecture

Lucid partnered with PwC and AWS to build 14 finance AI proofs of concept in 10 weeks. The architecture uses PwC Agent OS on AWS/Bedrock as a common orchestration layer, connects finance agents to SAP, Redshift, Salesforce, and other data, and targets workflows across demand forecasting, investor analytics, treasury, accounting, finance, and internal audit.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  A["Finance use-case backlog"] --> B["Agent OS orchestration"]
  B --> C["AWS Bedrock"]
  B --> D["SAP finance data"]
  B --> E["Redshift data"]
  B --> F["Salesforce data"]
  B --> G["Finance decision workflows"]

Components

Data flow

Finance teams identify pain points, Agent OS orchestrates agents with guardrails and policy adherence, agents access data sources such as SAP, Redshift, Salesforce, and other platforms, and the workflows produce outputs for forecasting, investor analytics, treasury, accounting, and audit decisions.

Why this architecture

Lucid chose a shared orchestration layer because individual agents were less important than coordinated outcomes across finance workflows with data access, guardrails, and policy adherence.

Models used: AWS Bedrock models · PwC Agent OS agents

Scale

• Lucid started with 14 use cases.
  “I think there's 14 different use cases you're starting with out of the gate.”transcript, stated
• The project moved from initial work to discussion in roughly 10 weeks.
  “Like, this was 10 weeks, something like that, right?”transcript, stated
• The initial POCs covered treasury, accounting, finance, and internal audit.
  “we've worked on these cons, these POCs across treasury, accounting, finance, internal audit.”transcript, stated

Alternatives considered (roads not taken)

Start with one or two prototypes

Rejected because: Lucid's pain-point backlog produced 14 early use cases rather than a narrow pilot.

Single-agent workflows without shared orchestration

Rejected because: Multi-agent finance workflows needed common guardrails, policies, and agent-to-agent lanes.

SAP-only finance data

Rejected because: PwC helped uncover valuable finance data in Redshift, Salesforce, and other platforms.

Patterns

multi-agent-orchestration (agent-architecture)

Agent OS coordinates multiple finance agents with shared guardrails and agent-to-agent workflow lanes.

tool-use-orchestration (agent-architecture)

Finance agents draw on SAP, Redshift, Salesforce, and other enterprise systems to produce workflow outputs.

guardrails-io-filtering (governance)

The orchestration layer provides safeguards, guardrails, and policy adherence for finance agents.

Tradeoffs

Pilot scope

Chose 14 use cases from pain points over one or two narrow prototypes.

Rationale: Lucid wanted to support growth across the finance organization and identified many critical areas at once.

Operating rhythm

Chose sprint iteration model over multi-year transformation project.

Rationale: The finance leaders treated 10-week iterations as the new model for AI capability delivery.

Gotchas (what bit them)

Finance teams may underestimate where valuable data lives.

Root cause: Lucid knew SAP but discovered important data in Redshift and Salesforce through the engagement.

Resolution: Connect agents to multiple enterprise data platforms rather than a single finance system.

Agent programs can focus on agents instead of outcomes.

Root cause: The business value comes from finance outcomes, not the presence of agents.

Resolution: Use a use-case backlog and orchestration layer centered on outcomes.

Evaluation

The case evaluates progress through POC delivery speed, breadth of finance workflows, demand-forecasting usefulness, efficiency, accuracy, and decision speed.

Metrics: use cases delivered · weeks to initial delivery · forecast turnaround speed · finance workflow coverage · efficiency · accuracy

Outcomes

• Lucid built 14 early finance AI use cases in about 10 weeks.
• Use cases covered demand forecasting, investor analytics, treasury, accounting, finance, and internal audit.
• The program established Agent OS as the common Bedrock-native orchestration layer for future finance agents.

This is an early-stage transformation case; many benefits are framed as planned growth enablement rather than mature production metrics.