Oil and Gas Company: Converge AI - taming complex joint venture workflows
Multi-party approvals, long-running state transitions, and millions in AFE budget decisions, built reliably with Temporal.io and an event-sourced audit trail.
The problem: joint ventures don't fit standard workflow tools
This oil and gas company manages joint venture projects that require Authorization for Expenditure (AFE) approval across multiple stakeholder groups. These workflows can span days or weeks, involve staged approvals from different parties, and carry significant financial weight.
Off-the-shelf tools couldn't handle the complexity. Approvals were tracked in spreadsheets, email chains held decisions, and there was no reliable audit trail of what changed, when, and by whom. State could get out of sync. Budgets drifted. Accountability was unclear.
The solution: Converge AI - a purpose-built workflow platform
G3 Digital built Converge AI as a joint venture between G3 and the client, a custom web application purpose-built for managing AFE approval cycles, budget tracking, and stakeholder coordination in the resources sector.
The core insight was that workflow state needed to be durable and recoverable. Standard database rows break under the pressure of long-running processes that can pause for human input, time out, retry, or branch based on decisions made days earlier. We chose Temporal.io as the workflow engine specifically because it treats state transitions as first-class, persistent objects that survive server restarts, network failures, and extended pauses.
Technical architecture
Temporal.io workflow engine
Every AFE approval is modelled as a Temporal workflow, a durable, resumable state machine. The workflow persists across server restarts, waits indefinitely for human approval signals, handles timeouts and retries automatically, and maintains a complete execution history. No cron jobs. No polling. No lost state.
Event sourcing with CQRS
All state changes are recorded as immutable events in an append-only event store (stream_id, stream_type, version, payload). The system never updates records. It projects current state from the event stream, giving a complete, tamper-evident audit trail of who approved what, when, and what the system was told.
AFE budget cycle management
The platform manages the full AFE lifecycle across 10 cost category groups, including Personnel, Plant & Equipment, Well Services, Subsurface & Technical, External Consultants, Land & Regulatory, and more. Each item is tracked at category and line-item level against estimate vs. actual spend.
Multi-stakeholder approval routing
Approval routing is configurable per workflow type. Signals from different approvers are accepted in any order, with the Temporal workflow holding open until all required parties have responded. Escalation paths, delegation, and timeout handling are built into the workflow definition.
Full-stack TypeScript
Server and client are both TypeScript, sharing type definitions. The backend exposes a REST API consumed by a React frontend. Flyway handles database migrations, ensuring a consistent schema across environments. Docker Compose orchestrates local development.
Built to scale
Temporal's architecture separates workflow logic from execution infrastructure. As the number of concurrent workflows grows, worker capacity scales independently. The event store schema uses partitioned indexes optimised for high-volume stream reads by stream type and position.
What Temporal.io makes possible that nothing else does
The key architectural decision, using Temporal over a traditional queue or state-machine library, proved correct in practice. AFE workflows that pause for multi-week approval windows simply work. If a server goes down mid-approval, the workflow resumes exactly where it left off when it comes back up. There is no reconciliation logic, no orphaned records, no "did that approval actually register?" uncertainty.
For a sector where a single mismanaged AFE can mean millions in untracked expenditure, that reliability is the product.
Stack at a glance
Let's find the highest-value automation opportunity in your business.
A short conversation is the fastest way to see what's possible.