System data and documentation are insufficient for understanding a business. The most crucial information—the exceptions, workarounds, and undocumented knowledge—lives in employees' heads. Conducting deep interviews to capture this "tribal knowledge" is essential before any automation begins.
Companies often believe their processes are linear and simple, like a 7-step plan. However, process mining reveals the messy reality: a complex, 20-step workflow with numerous exception loops that handle the majority of cases. Exposing this gap is the first step to optimization.
When overhauling a workflow, don't just think "automate." Categorize each step into one of four buckets: delete it entirely, handle with simple deterministic code, use a judgment-based AI agent, or reserve for a human decision-maker (especially for high-risk approvals or negotiations).
To get executive buy-in, visually contrast the client's current, convoluted process with a simplified "agentic future" diagram. Back this up with hard data showing dramatic improvements in KPIs like a reduction in cycle time from 24 days to 6, or cost per invoice from $31 to $6.
To master the Forward Deployed Engineer craft, start with yourself. Map your personal apps and inboxes, document a routine process like paying a bill, categorize each step (delete, code, agent, human), and design a more efficient, automated workflow for your own life before tackling a business.
Instead of building a new platform for AI agents, integrate them directly into the systems employees already use daily, like Salesforce or Slack. This avoids retraining, overcomes massive resistance to new software, and provides the path of least resistance to adoption and utilization.
An elite Forward Deployed Engineer (FDE) is a rare hybrid. They must understand business operations like a consultant, ship production-grade code like a senior engineer, and navigate the AI landscape (models, evals, security) like an ML expert, while also possessing strong communication skills.
Successful AI implementation isn't about applying it superficially over existing operations. It demands deep process mapping and rebuilding workflows from the ground up. Simply automating flawed processes only makes the business execute its mistakes faster.
The biggest AI labs promote their frontier models, but these are often unnecessary for real-world enterprise agentic workflows. More practical and cost-effective solutions can be achieved using smaller proprietary models (like Anthropic's Sonnet) or even open-source alternatives like Muse.
For large-scale AI transformation across a portfolio, creating bespoke solutions for each company is inefficient. Instead, group companies by their core software (e.g., all NetSuite users, all Salesforce users) to develop repeatable playbooks and streamline deployment.