This appendix presents a metrics system covering the full chain of FDE work, organized across four layers: delivery, customer, business, and organization. Metrics value precision over quantity—for each customer project, keeping an eye on 3 to 5 core metrics beats a dashboard covering an entire wall.

0. Quick Reference for the Book's Frequent Jargon ​

One plain-English sentence per term. If you get stuck while reading the main text, come back here to look it up.

  • FDE (Forward Deployed Engineer): An engineer who embeds at the customer site and is accountable for outcomes.
  • Ontology: A semantic layer that turns an enterprise's data, logic, and actions into something a model can understand.
  • Bootcamp: The customer brings real data, and within one to five days a usable prototype is built, with executives making decisions on the spot.
  • Echo / Delta: Palantir's two-person pairing—the former understands the customer, the latter builds.
  • PoC graveyard: A pilot with no deadline, no metrics, and no referee—it dies a slow death as it drags on.
  • MVD (Minimum Viable Deployment): Using minimal engineering effort to verify, once, in a real environment that value genuinely occurs.
  • Shadowing method: Sitting next to a real user and watching them get through a real day.
  • Lighthouse customer: A customer that sends a signal to an entire industry—their endorsement is worth more than the contract amount.
  • Requirement locust: A customer with ample budget and strong demand, but who drains the team without generating compounding returns.
  • NRR (Net Revenue Retention): Whether the same group of existing customers pays more or less this year than last year; 100% is the passing grade.
  • RPO (Remaining Performance Obligations): The contract value already signed but not yet recognized as revenue.
  • ARR / ACV: Annual Recurring Revenue / Annual Contract Value.
  • Rule of 40: Revenue growth rate plus adjusted operating margin—a software industry health metric; 40 is the passing score.
  • Outcome-based pricing: The customer doesn't pay for software; they pay for "problems solved."
  • Per-person-day pricing: Charging by the number of engineers multiplied by days—the old outsourcing pricing model.
  • SLA (Service Level Agreement): Commitments on availability, latency, and error rates.
  • Health score: A customer checkup score synthesized from four categories of signals: usage, value, relationship, and business.
  • QBR (Quarterly Business Review): A regular meeting between vendor and customer each quarter to align on value.
  • RAG (Retrieval-Augmented Generation): Having the model look up information before answering.
  • Evals system: A system that builds quantifiable yardsticks for fuzzy business quality—and also serves as the foundation for billing.
  • MCP (Model Context Protocol): An open standard, led by Anthropic, that lets models connect to external tools.
  • Customization decay rate: The amount of customization for the Nth customer should be significantly less than for the first; if it doesn't decline, you're doing outsourcing.

I. Delivery-Layer Metrics (Is the Project Done Right) ​

TTV (Time to Value): The time from entering the site to the customer obtaining their first measurable value. The lifeline metric of the FDE model. Reference standards: minimum viable deployment validation should be measured in weeks (2 to 6 weeks), and full deployment in months (1 to 4 months). A steadily lengthening time to value is the first signal that something is wrong with the methodology or the platform foundation.

PoC (Proof of Concept) conversion rate: The proportion of validation projects that convert into paid deployments. Palantir's bootcamps raised this number from an early 5% to 10% to roughly 75% per the company's disclosed figures (some sources report higher for certain sessions)—it simultaneously measures both "customer screening quality" and "delivery quality." A conversion rate that is too low means the entry screening failed; too high (approaching 100%), and you should check whether you're only taking on projects with no challenges.

Eval pass rate: The proportion of AI outputs that pass the business eval set, and its time-series trend in production. The alarm for quality drift.

Discipline in defining resolution rate and deflection rate: Resolution rate (the proportion of sessions closed without human intervention) and deflection rate (the proportion of users intercepted by self-service who never enter the human channel) are the core value metrics for customer-service deployments—and also the metrics with the messiest definitions across vendors. Take ServiceNow's official definition as an example: a deflection counts only when two conditions are met simultaneously—no ticket is submitted within 24 hours after the interaction, and a positive engagement signal appears (e.g., the user gives positive feedback on the result). Before citing any vendor's resolution or deflection numbers, ask three things: what is the denominator, how long is the determination window, and who makes the determination—an 86% with unclear definitions is worse than a 51% with clear ones.

Deployment frequency and rollback rate: Hard metrics for iteration speed. A healthy deployment cadence should maintain high-frequency, small steps (weekly or even daily), with a low and stable rollback rate.

Defect escape rate: The proportion of defects discovered only after going live. It measures the completeness of the testing and evals system, not the engineers' skill.

II. Customer-Layer Metrics (How Is the Customer Doing) ​

Activation rate: The proportion of the target user group that forms stable usage habits. Note the denominator is the "target user group," not "number of system accounts." Warning: a deployment with a chronically low activation rate is nominally alive but actually dead.

Usage depth: What fraction of people use key features (how many scenarios are covered), the distribution of usage frequency (clock-in-style usage or workflow embedding), and the emergence of self-directed exploration (users starting to discover new uses on their own—the most precious signal).

Health score: Synthesized from four categories of signals: usage, value, relationship, and business (see Section 5.6 for details). Key discipline: full review every week, and automatically trigger an intervention process whenever the score falls below the warning line.

Champion coverage: The number and hierarchical distribution of active allies within the customer organization. A single point is high risk; three points make a network.

Using NPS with caution: NPS (Net Promoter Score, the metric asking "how likely are you to recommend us to others") has limited reference value in enterprise scenarios—small samples, politicized responses. A more reliable alternative is "early inquiry about renewal intent": two quarters before expiration, directly ask your champions, "If we renewed today, would you renew?"

III. Business-Layer Metrics (Is the Business Worth It) ​

NRR (Net Revenue Retention): The annual change in revenue from existing customers (including churn, downgrades, and expansion). The final judge of the FDE business model. Passing line: 100%; excellence line: 120%. Companies with net revenue retention above 120% have a growth engine that is already self-sustaining.

Delivery gross margin: (Revenue per customer minus direct delivery costs (labor, travel, cloud resources)) ÷ revenue per customer. The passing line for the FDE model fluctuates with the degree of platformization: during pure human delivery it may be only 20% to 40%, and once platform reuse ramps up it should rise above 60%. How fast gross margin rises says more about whether the model works than the absolute level—an FDE whose margin doesn't rise is a consulting firm.

Customization decay rate: McGruther's touchstone—the customization workload for the Nth customer should be significantly less than for the first. If customization volume doesn't decline across three consecutive customers, immediately examine the product feedback loop.

Sales cycle: The time from first contact to signing. Palantir used bootcamps to compress it from 9 to 12 months down to a few weeks. It is a composite reading of customer acquisition efficiency and the depth of trust assets.

CAC payback period: The time it takes to recover customer acquisition cost (CAC, Customer Acquisition Cost, including investment in free validation) through contract gross profit. It generally looks bad early on in the FDE model; the key is to see a trend of it shortening as case studies accumulate.

LTV/CAC: The ratio of customer lifetime value (LTV) to customer acquisition cost. For enterprise businesses, the healthy threshold is generally above 3; in the FDE model, since acquisition costs are front-loaded, it may fall below 2 early on, and it only becomes meaningful when read together with net revenue retention.

IV. Organizational-Level Metrics (Can the Team Go the Distance) ​

Field-to-Product Feedback Rate: The number of outputs per unit of time that are distilled from the field into components or platform capabilities (components added to the library, playbooks updated, platformization proposals submitted). This metric measures whether the "soul organ" of the FDE model is still beating.

Delivery Asset Reuse Rate: The proportion of existing components, templates, and checklists reused in new projects. The reuse rate is a composite reading of the three-tier replication leverage (Chapter 7), and the target should rise quarter over quarter.

Revenue per FDE: The annual revenue supported by each Forward Deployed Engineer. It is the master ledger of scaling: the ceiling of a pure headcount model is obvious, and after platformization it should keep rising.

Team Endurance Metrics: Travel intensity (business trip days per month), on-call load, attrition rate, and burnout warning signals. The biggest complaint among FDE practitioners on Reddit is travel and endurance—if the team burns out, all the preceding metrics are just fireworks.

Usage Recommendations ​

  • Lock in 3 to 5 "core metric combinations" for each customer project: typically one value metric + one usage metric + one relationship metric.
  • Collect baseline data on day one of the project—miss it and it's gone forever.
  • Build and agree on metrics jointly with the customer; otherwise they carry no weight at the renewal negotiation table.
  • Overhaul the metric system itself every six months: remove the ones no one looks at, add the ones that keep getting asked about.