Become a $1M/yr FDE (Full Course)

An FDE’s potential compensation is tied to value delivered, not to the headline figures used to describe that value: the discussion frames deployments creating $5 million, $10 million, or $25 million in company value. The role is to re-engineer a process around AI, rather than bolt an agent onto an unchanged workflow. FDEs map the people, systems of record, decisions, handoffs, and exceptions, then redesign and deploy the process. Voss said Thrive had 35 engineers working across 70 firms at the time of his research, and reported tax returns completed 30% faster with 98% accuracy.

Effective mapping combines system evidence with what the speakers call the “human API”: interviewing employees to learn why steps exist, who decides, and how exceptions are handled. Salesforce activity can reveal repeated loops, but system data alone misses context, while internal documentation may be incomplete or outdated. At a public software company with $5 billion in revenue and more than 150 products, a quote process described as simple proved to have 20 steps and seven loops. Sixty-one percent of requests looped back, legal returned requests 12% of the time, and 30% of quotes repeated approval. The speakers recommend deciding step by step what to delete, handle with deterministic code, assign to an agent, or reserve for a person. Agents should generally fit into familiar channels and existing systems of record, rather than require a CRM or ERP replacement; one cited ERP migration cost about $10 million.

Shared software does not imply shared workflows. Five NetSuite portfolio companies had processes of 12, 9, 15, 18, and 13 steps, so private-equity firms can reuse process maps and playbooks, but must still map each company’s actual work. In an accounts-payable example, mapping a 17-step workflow took two to three weeks; redesign cut it to seven steps, reduced cycle time from 24 days to six, and cut exception loops from six to one. Straight-through processing rose from 18% to 87%, and cost per invoice fell from $31 to $6, an 80% reduction. A separate 60-person accounting firm with $12 million in revenue and 400 clients found that its stated six-step process actually had 14 steps. Collections had to be re-requested 70% of the time, and a partner later sent work back 35% of the time. The speakers stress measuring waiting time and exception loops: speeding individual steps may not shorten a process if work sits between handoffs.

FDEs should establish a baseline, estimate agent costs, track accuracy and operating KPIs, and be accountable for projected results after deployment. They describe value in cost savings, revenue uplift, and risk mitigation; automation may move employees to higher-leverage work rather than trigger mass layoffs. A faster, more accurate close can be valuable to a CFO even without reducing costs: the speakers contrast 22-day, four-week, or six-week closes with a potential four- or eight-day close. The business case should fit the buyer: CFOs may prioritize financial outcomes, technology leaders productivity and efficiency, and CHROs hiring and training. Their playbook is to identify an owned process, measure it, sort its steps, build against a baseline, deploy broadly, and keep measuring. Voss described a four-week CFO audit to understand systems and build proofs of concept, followed by four weeks of agent development and checks at three and six months.

Many deployments do not need a frontier model; the speakers describe combining frontier, lower-thinking, and open-source models and benchmarking workflows across options. They distinguish chat-based sidekicks from background agents that work without prompts and notify users when needed. Background agents can be configured with personal-agent tools, but enterprise governance remains insufficient, so deployment still needs significant engineering and human oversight. FDEs must understand processes and systems of record, ship production code with auditability, security, and governance, evaluate and manage models, and handle incorrect actions and rollbacks. They also need to communicate with senior leaders. Voss said his team routes through Azure Foundry, AWS Bedrock, and Vertex, which he said offer no training on customer data and zero data retention. The discussion raises on-premise deployment for sensitive data but does not resolve it.

A computer science degree is not required, but the speakers say FDEs need to learn the craft by deploying solutions and gaining experience across ecosystems, including Microsoft, Salesforce, open-source, and closed-source tools. One speaker offers 10% of delivered value as an illustrative compensation arrangement: $10 million in value could support a $1 million payment. Another example is a PE firm giving an FDE 0.5% of the increase in value between a hypothetical $1 billion purchase and $5 billion sale. For a first project, the advice is to inventory apps and communication channels, list 20 tasks from the prior week, document one workflow, and classify its steps. They suggest starting with a simple SMB workflow and, if needed, doing the first deployment free to build experience.