Occupation manual 019 / technical sales

Let AI prepare the evidence. Let humans make the technical promise.

A field manual for turning approved technical facts into a discovery and demo review packet—without letting a model decide what the product does, what a customer needs, or what anyone may promise.

Short answer: give AI only a bounded clerical job: organize approved, non-sensitive discovery material; tie each proposed statement to a current source/version/owner; mark unknowns; draft a neutral demo outline; and hand it to named humans. It must not determine capability, customer fit, architecture, security, configuration, price, terms, availability, implementation, or a customer commitment—and it must not send, submit, demonstrate, access, or change anything.

ForSales engineers, solutions consultants, technical founders, and product specialists.
OutputTechnical Discovery Evidence & Demo Review Packet.
Review date · published guide

What this manual is—and is not

O*NET describes sales engineers as a role that handles proposals, customer requirements, technical presentations, configurations, demonstrations, documentation, and customer communication. That makes source control practical: the role needs a clean line between a verified fact, an assumption, and a decision that belongs to a qualified human.

Educational boundary: this is not technical, security, privacy, legal, procurement, export, financial, contract, accessibility, or professional advice. Use actual product documentation, organization policy, current jurisdiction/industry requirements, and qualified reviewers.

Field map

Build a review packet before a technical promise

Bound the scenario

Record the exact question, approved purpose, minimum necessary information, named sales/technical owners, and excluded material. Stop when the request contains credentials, production data, a private architecture, private pricing, a contract, or a request to decide.

Output: scenario card

Bind facts to sources

For every needed capability or condition, capture the approved source, version, effective date, source owner, and exact support. An old slide, model memory, customer assertion, or inference is not a source of record.

Output: source ledger

Separate fact from unknown

Build a matrix for requirement, supported fact, condition, conflict, unknown, and human decision owner. Preserve missing evidence. Do not turn a gap into a promise.

Output: capability matrix

Draft a neutral demo outline

AI can organize an approved scenario and source-bound talking points. It must not claim availability, security, compatibility, performance, timeline, price, or customer outcome.

Output: demo draft

Hand off for approval

Name the technical, security/privacy, commercial, implementation, and customer-facing reviewers. Their decision—not the model's draft—controls any proposal, demo, message, or action.

Output: review receipt

Assign AI narrow roles

AI roleApproved inputPermitted outputHuman gate
Evidence clerkApproved public or organization-approved source excerptsFact/source/version/owner ledgerTechnical source owner confirms direct support
Assumption finderReviewed discovery notes and ledgerUnknowns, conflicts, and questionsQualified owner decides whether/how to answer
Demo-outline scribeReviewed scenario and supported factsNeutral agenda and talking-point draftTechnical and commercial owners approve delivery
Packet assemblerReviewed evidence and named rolesDraft handoff receiptAuthorized humans control every external action

Safe preparation vs. approval-required work

AI may prepare

  • source and version inventories
  • fact/unknown separation
  • missing-evidence questions
  • neutral demo outlines
  • reviewer and dependency checklists
  • draft-only handoff packets

Humans must decide or do

  • customer need, product fit, architecture, and configuration
  • security, privacy, data processing, accessibility, and compliance
  • price, discount, scope, dates, service level, terms, and contract
  • product capability, compatibility, benchmark, and outcome claims
  • proposals, RFPs, demos, messages, CRM records, and public content
  • account access, data sharing, deployment, spending, and commitments

Failure modes that should stop the packet

Model memory becomes product truthRequire an approved source/version/owner for every factual statement.
An unknown becomes a promiseUse the visible label unknown; route it to a human decision owner.
Private context enters the promptUse minimum necessary approved information; stop on credentials, production data, personal data, or private commercial terms.
A demo draft becomes an approved demoDrafting is not approval. A named human still controls scope, account, audience, and delivery.
Technical language masks a marketing claimClaims still need truthful, non-misleading, appropriately substantiated review.
One reviewer owns every riskName distinct technical, security/privacy, commercial, implementation, and customer-facing owners when applicable.

Runnable local artifact

Technical Discovery Evidence & Demo Review Desk

This browser-only form accepts fictional or approved non-sensitive placeholders and produces a needs_review receipt. It stores nothing, makes no network request, and cannot access systems or take an external action.

Required gates
Complete the fictional fields and gates to create a review-only receipt.

Primary sources

Reviewed July 25, 2026. Recheck source versions, product documentation, organization policy, and jurisdiction/industry requirements before real-world use or customer-facing use.

Where this fits next

Use the AI Agent Workflow Builder for least-privilege job cards, the Sales Pipeline Management guide for pipeline context, and the AI Marketing Operating System for claim-review context. None of these links authorizes outreach, publishing, spending, or account action.