How Alpha Scaling works

What Alpha Scaling is

Alpha Scaling runs AI-assisted business audits: a structured diagnostic that looks at where an established business is leaking margin, and quantifies the opportunity to recover it. The work blends a versioned analysis engine and a maintained knowledge base with AI assistance, and every result is reviewed and published by a human audit team.

Who it is for

Alpha Scaling is built for established service businesses — for example, plumbing and home-service companies with roughly $5M+ in annual revenue — where operational and financial complexity has outgrown spreadsheets and gut feel, and small percentage improvements translate into meaningful money.

How an engagement flows

Here is the path a business's information takes — from what it shares to the findings it receives — and where humans stay in control.

Structured data intake

An engagement begins with structured intake rather than a blank page. A client works through a guided checklist (the questionnaire) and uploads the supporting business documents it references. Two things happen as material arrives:

• Files enter over encrypted connections and are stored in private storage that is not publicly accessible; they are retrieved only through short-lived signed URLs.
• Each uploaded file passes a deterministic, server-side validation step that runs no AI model and records a verdict before the file is accepted for analysis.

The intake is designed so that the audit team is working from consistent, structured inputs — not an unstructured pile of attachments.

The information involved

An operational and financial diagnostic needs real operating information. Depending on the business, that typically includes the intake questionnaire, engagement details, and uploaded operational and financial documents. This material is treated as sensitive.

• Uploaded business and financial documents are handled as sensitive: private storage, signed-URL-only access, and no exposure to another organization.
• Document contents are never sent to the AI assistant. They are processed server-side by the analysis engine.
• Access is scoped to the client's own organization and to audit staff, enforced in the database with forced Row-Level Security.

A category-by-category breakdown is on the Trust Center and Data Governance pages.

The analysis engine & knowledge base

At the center of an audit is a versioned, tested analysis engine — a structured analysis system, not a generic AI wrapper. It works against a maintained knowledge base:

• The engine models findings, leak zones, and precedence, and evaluates structured inputs consistently across engagements.
• The knowledge base holds benchmarks, operational playbooks, market context, prior case patterns, and a precedence matrix.
• A scoring step evaluates each potential lever, so findings can be prioritized rather than dumped as a flat list.

Because the engine is versioned and tested, the same inputs are analyzed the same way — a property that matters when a person later has to stand behind a finding.

Where AI assists — and where it does not

AI is an analytical tool inside a defined workflow, not the decision-maker. Our AI-assisted features use Anthropic's Claude (Haiku 4.5), and their role is deliberately bounded.

• Data minimization — only metadata such as file names, statuses, and labels reaches the model; document contents and financial figures do not.
• Server-side and scoped — model access happens only on the server, and the assistant's context is derived from the caller's own organization, never from request parameters.
• Not persisted — prompts and outputs are not retained beyond a single minimal audit event (input size and success only).

Full detail is on AI Governance and the Security & AI Governance Overview.

Human auditors & the publish gate

This is enforced in the product, not just intended:

• Publish gate — a finding that has not been published is invisible to the client by design; access rules return only published findings, and only for the client's own organization.
• Staff-only, logged writes — findings are written only through role-checked server-side functions that log every change to an append-only event log.
• Findings only from published rows — any number a client sees traces to a published finding reviewed by the audit team.

In short: AI-assisted analysis, reviewed and published by the audit team. More on Audit Integrity.

What a client receives

The output of an engagement is a set of published findings on the client dashboard — leak zones with estimated dollar ranges and a confidence signal — together with any deliverables the audit team prepares. Reports and deliverables stay private until a person publishes them.

Every published number is one a human auditor reviewed and stood behind. The dashboard is the single source of client-facing figures; the in-panel assistant never is.

Security, privacy & governance

Because an audit involves sensitive operational and financial information, the controls around it are documented in detail: