The AI Stack for a $10M Professional Services Firm

By May 2025, I've advised enough professional services firms to see what actually works. The firms winning with AI aren't the ones with the most tools—they're the ones with the right stack, implemented with discipline.

Here's the foundation I recommend for a $10M firm. This isn't theoretical. These are tools I've tested, firms are using, and that directly drive revenue and efficiency.

The Three Layers

A professional services AI stack has three layers: foundation models, workflow tools, and integration infrastructure.

Layer 1: Foundation Models

You need depth, not breadth. For professional services work, I recommend:

• Claude 4 Sonnet (Anthropic) — Strategic analysis, client-facing advisory, complex reasoning. This is your primary tool. Fast enough for daily use, intelligent enough for nuanced work. Cost: ~$20/month for individual, or API per-token for at-scale use.
• GPT-4 (OpenAI) — Complementary for specific workflows. Slightly different reasoning style, excellent for content generation and certain research tasks. Cost: Similar to Claude.
• Gemini 2.5 Pro (Google) — Keep it accessible for brainstorming and secondary analysis. Good multimodal capabilities. Cost: Included in Workspace if you're a Google Shop.

Don't over-build here. Three models, used well, beats ten models used poorly.

Layer 2: Workflow and Integration Tools

Where the real work happens:

• Document and Research: Perplexity AI for real-time research, Claude for deep analysis, native PDF tools. Most firms use a combination of Claude Web, Claude API, and their own document stores.
• Spreadsheet and Data: Claude with spreadsheet integration (Excel/Sheets via API), plus native BI tools. Avoid "AI-powered spreadsheet" toys—native tools with AI assistance beat standalone AI products.
• Workflow Automation: Zapier or Make for connecting tools. By May 2025, both support Claude and other LLMs directly. Critical for routing data between systems without manual handoffs.
• Knowledge Management: A searchable knowledge base (Notion, Confluence) with AI retrieval. This is non-negotiable for scaling advisory work.

Layer 3: Infrastructure and Governance

The plumbing that makes it all work:

• API Access and Limits: Set up API accounts with rate limits and cost controls. By May 2025, both Anthropic and OpenAI offer usage-based billing with clear spending caps.
• Data Governance: You need clear policies on which data can use which tools. Implement logging and audit trails.
• Security and Compliance: Enterprise agreements with vendors you use at scale. Most firms at this scale need SOC 2 compliance, which the major vendors offer.

What This Costs (May 2025)

For a 20-person firm using this stack:

• Claude Pro subscriptions (10 team members): ~$2,000/year
• ChatGPT Team or individual subscriptions (10 team members): ~$2,000/year
• API costs (models for workflow automation): ~$500–$1,500/month depending on use
• Workflow automation (Zapier): ~$1,000–$2,000/year
• Knowledge base tool: ~$2,000–$5,000/year (depends on choice)
• Total: ~$8,000–$12,000 per year, or $40–$60 per team member per month

For a $10M firm with 20–30 people, this is almost nothing. The ROI is 10x in the first six months if implemented correctly.

Implementation Sequence

Roll this out in four phases:

Month 1: Foundation

Get Claude and ChatGPT accounts for your leadership team. Start with strategic analysis and client research. Build muscle memory.

Month 2: Workflows

Layer in workflow tools. Connect Claude to your spreadsheets and CRM. Start automating repetitive analysis tasks.

Month 3: Team Rollout

Train your full team on approved tools. Establish data classification and governance policies (from the governance framework post). Create templates for common workflows.

Month 4+: Optimization

Monitor usage. Identify which workflows drive the most value. Double down on those. Refine governance based on what you learn.

What NOT to Do

I've seen firms stumble by:

• Building custom models. You don't have 2025 yet. Use foundation models. Custom fine-tuning makes sense at scale ($50M+), not at $10M.
• Treating AI as a replacement for people. It's not. It's a multiplier. The question is: what can your team do that's 10x more valuable with AI?
• Over-automating. Some workflows need the human in the loop. Client work especially. Use AI to augment, not replace.
• Neglecting governance. You don't need a 100-page policy, but you need documented rules. Governance is risk management.

The Real Advantage

By May 2025, most firms have access to similar tools. The advantage isn't having AI—it's using it better than your competitors.

Firms that win are the ones that:

1. Invest in training and adoption (not just buying licenses)
2. Design workflows around AI, not bolt-on tools
3. Measure and iterate on what works
4. Maintain strict data governance while moving fast

This stack gives you the foundation. What you build on it is up to you.