How Ashton uses AI in the work.

// 01 Position

Human judgment is the deliverable.

When you hire Ashton Advisors, you're hiring Brandon Ashton Stein. AI assists the work — it doesn't replace the work. Every recommendation, framework, document, and decision in a client engagement is reviewed, edited, and owned by the principal. The bar is the same as if no AI were involved.

// 02 Operating principles

Seven commitments.

• Human-in-the-loop on every output. No AI-generated artifact leaves the practice without principal review.
• Transparency with clients. When AI materially contributed to an analysis, deliverable, or recommendation, we say so — in scope documents and where it matters in the work itself.
• Confidentiality first. Client-confidential data, PII, financial details, M&A targets, and anything covered by NDA do not go into consumer-grade AI tools. See § 04 for the controlled list.
• Explainability. Conclusions in client work are sourced and traceable. If AI-assisted analysis can't be explained on its own merits, it doesn't ship.
• Bias awareness. AI models reflect their training. We test for bias in outputs that affect customer-segment recommendations, hiring frames, or evaluation criteria, and we flag it when we see it.
• Lawfulness. Use of AI tools complies with applicable law (GDPR, CCPA/CPRA, sector-specific regulations relevant to client engagements) and the AI provider's terms.
• Continuous review. Tool list, controls, and this page are reviewed at least quarterly.

// 03 Where AI is used in the practice

Synthesis, drafting, research — not decision-making.

• Synthesis: Compressing long source material into structured frames for client review.
• First-pass drafting: Outlines for deliverables, decks, memos, and engagement proposals — always edited materially before client delivery.
• Research: Background scanning on industries, vendors, and frameworks. Outputs are verified against primary sources before they influence recommendations.
• Code & data work: Light analysis, spreadsheet logic, prototype scripts — reviewed line by line.
• Newsletter drafts: Some Insights content is written start-to-finish by the principal; some is AI-assisted in early drafts and then rewritten. The substantive point of view is always Brandon's.
• Communication: Editing and proofreading drafts written by the principal.

AI is not used to make recommendations, set strategy, or take positions on behalf of clients without principal authorship and review.

// 04 Tools we use

Enterprise-grade, with controls.

Where client-confidential or sensitive material is involved, we use AI tools on paid or enterprise tiers with data-protection commitments that prevent training on customer input:

• Anthropic Claude (including Claude for Work / Cowork — paid tier with no-training defaults) — primary AI assistant for synthesis, drafting, research, and code review.
• Microsoft 365 Copilot — AI features integrated within Word, PowerPoint, and Excel, used for productivity assistance on practice-internal work.
• Google Workspace Gemini — AI features within Workspace boundaries.
• Aragon AI — used only for AI-generated professional portrait imagery of the principal. No client material is processed through Aragon.
• Canva (including AI-assisted features such as Magic Write and image generation) — used for visual design and brand assets, not for client-confidential analysis.

Tools without paid-tier or enterprise data-handling commitments are not used with client-confidential material. The list above is reviewed quarterly and updated here when it changes.

// 05 What clients can expect

Disclosure, control, and a clean separation.

• Disclosure in scope. Statement-of-work documents identify whether AI tools will be used in the engagement and how.
• Opt-out is honored. If a client prefers no AI assistance in their engagement, we work without it. The cost of the engagement reflects that scope.
• No client data trained. Client-confidential inputs are never used to train public models. We use enterprise tiers specifically to ensure this.
• Provenance on outputs. Where AI materially shaped a deliverable, we note it in the document or scope summary, so the client knows what was generated, what was synthesized, and what was reasoned through directly.

// 06 What we won't do

Lines we don't cross.

• Submit client-confidential data to a consumer (non-enterprise) AI tool.
• Pass AI-generated output as principal-authored work without review.
• Use AI to fabricate quotes, citations, statistics, or supporting evidence.
• Use AI to evaluate individual people (candidates, employees) in ways that would not be defensible if disclosed.
• Deploy AI in any way that contradicts a client's stated AI policy or industry regulation.

// 07 Accountability

The principal is responsible.

Brandon Ashton Stein, as principal, is accountable for the responsible use of AI in the practice. Concerns, questions, or incidents should be raised directly to brandon@ashtonstrategies.com.