The canonical definitions.

01 · THE CATEGORY

Knowledge as a Service

The category that converts the organic brain knowledge of a trained, distributed workforce into a continuously refreshed intelligence layer distributed to three endpoints: an internal dashboard, a scoped brand-partner API, and an AI-agent retrieval feed. The unit of input is a structured AI-guided interview. The unit of output is infrastructure rather than a report.

KAAS

The acronym for Knowledge as a Service. Pronounced kahss. Stacks alongside SaaS, IaaS, and PaaS in the same way: the suffix signals continuous distribution and API-native delivery.

Living corpus

The full set of enriched conversations across a workforce, indexed on the ten dimensions, timestamped with bi-temporal validity, and scoped through a segment vector. The corpus refreshes weekly and decays per-dimension according to calibrated half-lives.

02 · THE THREE STAGES

Gather

The first stage. Our AI conversation agent runs a fifteen to forty-five minute structured interview with each worker who holds knowledge worth capturing. Cost: three cents per enriched conversation. Output: a transcript, a structured field set, and twenty to forty enrichment-ready chunks per interview.

Codify

The second stage. Each conversation is decomposed into ten parallel structured dimensions and persisted into a fact store with bi-temporal validity and segment scoping. The output is a queryable layer with a stable schema. Same fact, three consumption modes (dashboard, API, agent feed).

Monetize

The third stage. The codified layer is distributed simultaneously to the internal dashboard, the scoped brand-partner API, and the AI-agent retrieval feed. Pricing is infrastructure-grade, with a percentage-of-downstream logic and an enterprise floor.

03 · PEOPLE AND ROLES

AI conversation agent

The interview-conducting agent. Built with a behavior tree, protocol enforcement rules, knowledge domains, and per-agent guardrails defined in a versioned specification. The agent runs every interview against the same protocol on conversation number one and on conversation number ten thousand.

Intelligence Layer

The product surface composed of the dashboard, the brand-partner API, and the AI-agent feed. The single deliverable a KAAS client purchases. The intelligence layer is what makes the deal a business unit rather than a research project.

Intelligence matrix

The internal name for the set of MCP tools and prompts that read against the codified corpus on behalf of agents, dashboards, and API consumers. Every operation in the matrix runs under a Zero Data Policy. The matrix is the runtime; the corpus is the data.

04 · METHODOLOGY

AI-guided structured interview

A fifteen to forty-five minute conversation between our AI conversation agent and a worker, run against a versioned protocol. Anchor-gated probing, neutrality enforcement, time-boxed depth, structured wrap. The instrument that holds both low cost and protocol fidelity, which neither human interviewers nor surveys can do at scale.

Anchor-gated probing

The first move in every theme is an opening anchor on a concrete recent event in the worker's domain (for example, the last time a customer asked for a product the worker did not stock). The agent does not advance to the next theme until the anchor is locked. Abstract questions produce abstract answers; the anchor surfaces specifics.

Neutrality protocol

The agent checks its next utterance against a neutrality classifier before delivering it. No leading phrasing, no value loading, no social-desirability primes, no acquiescence pulls. The protocol prevents the agent from biasing the response in a direction the analysis would later have to discount.

Wrap protocol

Conversations end on a structured close, never on a fade. The last two minutes of an interview produce the highest-fidelity content because the respondent has warmed up and trusts the instrument. The wrap captures that signal.

Ten dimensions

The ten parallel structured fields that every raw conversation is decomposed into during the Codify stage: pain points, opportunity scoring, brand-perception map, AI-readiness scoring, retention drivers, market signal, regulatory and compliance risk, training gap, customer-segment behavior, competitor mention map. The schema is vertical-specific in surface form and stable in shape.

Pre-enrichment gate

The verification step that runs before any enriched record is persisted. A deterministic filter confirms every quote that leaves the system is a literal substring of a real USER turn from a real interview. When the filter detects an invented quote, the quote is deleted, never rewritten.

Verbatim provenance

The contract that every claim in every report links back to the transcript that produced it through a chunk identifier of the form transcript_id / turn_id / sentence_id. The link is always present, on every conversation, with no sampling. Cleaned, shortened, and translated display layers all derive from the original verbatim, which is never translated and stays in its original language as the provenance record.

05 · TECHNICAL PRIMITIVES

Bi-temporal model

The data model in which every fact carries three independent timestamps: learned_at, valid_at, and invalid_at. The model lets a brand partner replay last May's state of the corpus against today's and get the correct answer for each. That property is what makes the layer underwritable in a contract.

learned_at

The timestamp at which the system observed a given fact (the interview date or the enrichment date). Distinct from when the fact became true in the world.

valid_at

The timestamp at which a fact became true in the world. Often inferred from the conversation itself (a worker says ‘I started noticing this last fall,’ so valid_at is approximately October of that year). Distinct from when the system learned it.

invalid_at

The timestamp at which a fact stops being true. Set explicitly when a contradicting conversation lands, or by half-life decay when no refresh confirms the fact within the calibrated window.

Segment vector

A sparse projection over the segmentation axes that matter for the vertical (tier, geography, store format, customer segment, brand partner, demographic). Every fact carries one. The query language becomes WHERE valid_at ⊃ today AND segment_vector ⊃ {region, format, partner}. Per-segment validity is what lets a single corpus serve every brand partner on the shelf without leakage.

Per-segment validity

The property that the same surface claim can hold for one segment slice today and not for another. ‘Shade X gets rejected most’ can be true for Latin America mass-market stores and false for EMEA boutiques. Per-segment validity is institutional memory in a structurally different shape from per-user agent memory.

Half-life

The per-dimension decay rate calibrated for each vertical. Brand-perception facts decay in weeks. Training-gap facts only decay when the training catalog changes. Tuning half-lives per dimension and per vertical is methodology, and it separates a credible refresh cadence from rerunning a survey.

Refresh cadence

The schedule on which a slice of the corpus is re-interviewed to confirm or invalidate facts. The cadence is wired into the fact lifecycle. The dashboard surfaces a refresh trigger when a meaningful fraction of facts have aged out of validity. The Refresh tier sells against that trigger.

MCP Server

The agent-facing contract surface. An MCP-compliant server (consumed by Claude, Copilot, and other MCP-compliant clients) that exposes a small tool surface: query_facts, traverse_relations, pivot_dimension, replay_as_of, cite. The session ticket carries the security scope. The scope cascades into which tools the LLM can see, which graph nodes it can traverse, which fact rows it can read, and which verbatim chunks resolve when it cites.

Brand-Partner API

The second monetization endpoint. Downstream third parties (for example, the brand partners who pay a retailer for shelf space) pay the client for scoped access into the corpus. Each partner's key projects onto a specific brand_partner value and surfaces only the facts whose segment vector intersects that projection. The data path is the access control.

AI-Agent Feed

The third monetization endpoint. The retrieval surface that LLM clients consume. RAG-ready, with stable schema and source citation. Used by the client's own AI assistants (the advisor copilot, the buyer's planning agent) and by brand partners' agents querying via the API. Once an agent is wired to consume the layer, switching cost compounds with every new agent task.

Internal Dashboard

The first monetization endpoint. The client's own strategy, merchandising, or operations team consumes the layer to make decisions. The lowest-controversy endpoint and the entry wedge for every deal.

Cloud Harness

The deployment configuration that runs the entire KAAS stack inside the client's own substrate (Cloudflare account, AWS, Azure, or on-premise). The capture pipeline, the cleaned transcripts, the bi-temporal layer, the fact store, the dashboards, and the AI agents all live on infrastructure the client controls.