The AI Stack That Runs Our Agency: Full Transparency
We write extensively about AI systems for our clients. This post turns the lens inward — the specific tools, the specific workflows, and the specific results from the AI stack that runs SA Solutions. No vague claims. Exact tools, exact processes, exact numbers.
With Costs and Purposes
| Tool | What We Use It For | Monthly Cost | Team Users |
|---|---|---|---|
| Claude Pro (Anthropic) | All writing tasks, proposal drafts, client communication, research | $20/month | All team members |
| Make.com Core | Automation scenarios connecting all platforms | $9/month | 1 Make.com specialist |
| GoHighLevel | CRM, pipeline, lead scoring, follow-up automation | $97/month | Sales and account management |
| Bubble.io Growth | Client portals, internal tools, custom applications | $119/month | Development team |
| Otter.ai | Meeting transcription and summary | $16.99/month | All team members |
| Buffer | Social media scheduling | $15/month | Content team |
| Xero | Accounting with Make.com integration | $65/month | Finance |
| Apollo.io | Lead enrichment and prospecting data | $49/month | Sales |
| Claude API | Automation workflows, document processing, AI features in apps | ~$35/month | All automations |
| Total | ~$426/month |
Exactly How They Work
Lead scoring on every new enquiry
Every new enquiry submitted through our website form or any other channel triggers a Make.com scenario within 3 minutes. Apollo enriches the contact with company size, industry, and job title. Claude scores the lead against our ICP criteria and returns a score (0 to 100), a tier (A, B, C, or D), and a one-sentence qualification summary. GoHighLevel is updated with all three fields. A Tier A lead triggers an immediate Slack notification to the founder. A Tier B lead triggers a 24-hour follow-up sequence. This runs automatically for every lead, 24/7, with no manual involvement.
Same-day proposal generation
After every discovery call, the account manager completes a structured debrief in a Notion template (10 minutes of focused reflection). Make.com detects the new completed debrief. Claude generates a complete proposal draft — executive summary, situation analysis, proposed approach, deliverables, investment, and why us — in approximately 3 minutes. The draft appears in a Google Doc shared with the account manager. The account manager reviews, personalises (adds specific examples from the call, adjusts any sections that need context only they have), and sends via PandaDoc. Total time from call to sent proposal: typically 60 to 75 minutes. Win rate since implementing: up from 26% to 38%.
Weekly client status updates
Every Monday at 6am, Make.com runs a scenario for each active client project. It collects task completion data from our project management tool, milestone status from Bubble.io, and any flagged items from the previous week. Claude generates a client status update in our brand voice: what was accomplished, what is planned, any decisions needed. The update is posted to the client’s portal and emailed from the account manager’s address at 7:30am. Clients receive consistent, professional updates before their working week begins. Zero manual writing involved.
LinkedIn content batch production
Every Sunday, a 90-minute session produces the week’s LinkedIn content. The session uses our insight capture library (a running Notion page of observations and ideas), Claude for drafting, and our brand voice system prompt for consistency. Typically produces 5 to 7 posts: 2 longer educational pieces and 3 to 5 shorter observations or stories. Posts are scheduled in Buffer for the week ahead. The session also produces the weekly newsletter draft (reviewed and sent Tuesday mornings). Zero content gaps in 14 months since implementing.
Payment chasing automation
Xero tracks invoice status. Make.com checks daily for invoices overdue by 3, 10, or 21 days. For each overdue threshold: Claude generates a professionally worded reminder calibrated to the relationship and the overdue duration. The reminder is emailed from the account manager’s address — personalised, not template-looking, specific to the invoice. Average collection time dropped from 48 days to 29 days since implementing. No awkward manual chasing. Zero invoices falling through the cracks.
Can a smaller business replicate this stack at lower cost?
Yes — the core of this stack for a solo founder or 2-person business: Claude Pro ($20), Make.com Core ($9), and GoHighLevel ($97) gives you the AI, the automation infrastructure, and the CRM. Total: $126/month. The Bubble.io, Apollo, Otter, and Buffer components are additions that become valuable as the team grows. Start with the core three; add the others when the specific use case is clear and the cost is justified.
How long did it take to build all of this?
The full stack as described took approximately 6 months to build, with new components added one at a time. The first component (lead scoring) took 2 weeks to build and has been running unchanged for 11 months. The last component (payment chasing) took 3 days to build — the team’s familiarity with Make.com and Claude had made each subsequent build faster. Building a similar stack today, with a clear plan and the guides in this series, would take 3 to 4 months rather than 6.
Want SA Solutions to Build This Stack for Your Agency?
We build the same AI stack described in this post for other service businesses — customised for your specific workflows, your team size, and your client base.
