AI Operations

The AI-Run Business Stack: Website, CRM, Content, Reporting

A four-piece stack for a small services business that wants AI in the loop without inheriting a SaaS farm.

Published 2026-03-25 · By Claire Miller

A small business that wants AI in the loop without inheriting a SaaS farm in 2026 needs four pieces, no more, no fewer. The four pieces are:

• A website a small operator can publish to without waiting on engineering.
• A CRM that holds the human side of the customer relationship.
• A content system that lets the business speak publicly about its work.
• A reporting layer that says what is happening in the business without a manual report.

The four pieces can each be bought, built, or assembled. The trick is to choose across all four pieces together, so the seams do not add up to a second full-time job.

The website

For most small services businesses, the website is a static site, deployed to an edge CDN, generated from Markdown or similar content. The publishing loop is: write a Markdown file, commit it, the site rebuilds and deploys in under three minutes. The editing surface is either the GitHub editor or a file-based CMS layer (Decap, Tina, Sveltia) that front-ends a Git repo.

What the website needs to do:

• Render service pages as HTML readable by crawlers.
• Emit schema.org JSON-LD for the business, the services, and any FAQ.
• Serve a sitemap and an RSS feed.
• Load in under two seconds on a phone.
• Allow an operator to publish a new page or update an existing one without filing a ticket.

What the website does not need to do:

• Render dynamically per visitor.
• Host authenticated content (use a separate app if needed).
• Run a CMS that the team uses directly (the Git workflow is fine).

The CRM

For most small services businesses, the CRM is the place where the human relationship lives. Customers and leads and proposals live here. Email is partially here (sent from here, received into here). Tasks assigned to humans live here.

The interesting question in 2026 is which CRM. The answer depends on what the business already runs. If the business has been running HubSpot, Attio, Pipedrive, or Folk, the answer is probably "the one you already use." Adding a third CRM is rarely worth it.

What the CRM needs to do:

• Hold every customer, lead, deal, and conversation.
• Allow a worker (an agent) to read and write specific objects.
• Generate activity and pipeline reports with one click.
• Surface "what needs human attention today" cleanly.

What the CRM does not need:

• A built-in AI assistant that the business uses via the CRM UI.
• Cross-channel orchestration that pretends to replace the marketing stack.
• A complex automation engine that only one person understands.

The content system

The content system is the publishing pipeline. It is what produces the website updates, the social posts, the newsletter, and the LinkedIn presence. It is also where the AI workers earn their keep: most small businesses under-publish by an order of magnitude, and the content system is what closes that gap.

The content system is, typically:

• A versioned folder of source Markdown (the same Git repo as the website, or a sibling).
• A scheduled worker that turns source Markdown into platform-shaped drafts.
• A review surface (Slack channel, Notion database, Telegram thread) where the human approves or rejects.
• A publishing worker that takes approval and pushes the draft to the destination.

What it does:

• Produces draft posts for the website weekly.
• Produces draft social posts for the platforms the business uses.
• Tracks which drafts were approved, rejected, or escalated.
• Surfaces which posts are landing, via simple per-post metrics pulled weekly.

What it does not do:

• Run an AI-generated infinite content calendar.
• Replace the operator's editorial judgment on hot takes.
• Auto-publish anything. (Draft-first. Always.)

The reporting layer

The reporting layer answers one question: what happened in the business this week, what is happening now, and what should we do about it?

For most small businesses, the reporting layer is:

• A daily snapshot of CRM state (new leads today, deals advanced, deals stalled, follow-ups due).
• A weekly summary from the content system (posts drafted, drafts approved, posts published, post-level metrics).
• A weekly summary of any other system (banking, ad spend, support tickets).
• A weekly meeting (or a written memo) where a human reads the dashboard.

The AI workers in this layer are largely mechanical: SQL queries against the CRM database, weekly aggregation scripts against the content system's logs, schema-validated summaries. The agent surface is small here. The benefit is consistency: instead of "did anyone update the spreadsheet," the dashboard is the spreadsheet.

The seams between the four pieces

The four pieces do not need to be tightly integrated. The integration points are:

• CRM data into the website (testimonials, service proofs): a script that runs nightly.
• Content system outputs into the CRM (publishing events that update lead interactions): a webhook.
• Reporting layer pulling from all three: a scheduled job.

That is it. No tightly-coupled platform, no shared APIs, no custom middleware. The seams are scripts and the scripts are tiny.

What to actually pick

There is no single stack that wins for everyone; the right answer is the one that fits what the business already has. A small services business that already runs HubSpot, ships via a Next.js site, and publishes to LinkedIn is best served by extending those rather than switching.

For a business starting from zero in 2026, the working minimum is:

• Astro or Hugo or Eleventy for the site.
• Attio or HubSpot or Pipedrive or Folk for the CRM.
• A custom content system built around a Git repo, a cron, and a review surface.
• A scheduled reporting job that aggregates from the other three.

Total vendor cost: low hundreds of dollars a month. Total complexity: the cost of one careful operator who owns the stack. Total leverage: the ability to run a content-driven small business at 5x the manual rate.