Beyond ChatGPT: building AI into the processes your business already runs

Six months into AI adoption, most small businesses hit the same plateau. ChatGPT is open in a tab. Useful prompts are saved somewhere. The owner uses it daily. One or two staff use it occasionally. Time savings are real but unmeasured. Nothing has fundamentally changed in how the business runs.

This is not failure - it is a normal stage. The next step is moving AI from a tab someone visits to part of the actual workflow. This article covers the practical patterns.

The ChatGPT-tab plateau

The plateau looks like this. The owner has saved prompts. The team knows AI is approved. People use it when they remember. But:

• Half the team forgets it exists most days.
• The owner is the only person consistently using it well.
• When someone leaves, their prompt knowledge leaves with them.
• Time savings are uneven - some weeks AI saves five hours, some weeks zero.
• The same questions get answered manually because no one builds the AI into the process.

The fix is to embed AI into the work, not as a separate tool. Three practical patterns where this lands well for SMEs.

Pattern 1: Email workflow integration

Most SMEs spend significant time on email. The natural place to embed AI is the inbox itself.

Microsoft Copilot for Microsoft 365 sits inside Outlook. It can summarise threads, draft replies based on context, and answer questions about your inbox content. Built on RAG so it knows your actual emails.

Gemini in Google Workspace does the same for Gmail. Click a button to draft a reply; it sees the thread context.

ChatGPT and Claude do not have inbox integration as cleanly, but you can paste threads in to get summaries and drafts.

For most SMEs already on Microsoft 365 or Google Workspace, enabling the built-in AI is the simplest meaningful step beyond "tab open". The cost is bundled into existing licensing.

Pattern 2: Document and CRM integration

If your business runs on a system - Xero, QuickBooks, HubSpot, Salesforce, a property management system, a booking system - check what AI integration is built in.

Most major SaaS products in 2026 have AI features integrated. They are uneven in quality but rapidly improving. Examples worth a look:

• Xero has AI for transaction categorisation and bank reconciliation.
• HubSpot has AI for email drafting and content generation, integrated with your CRM data.
• Microsoft Dynamics, Salesforce and similar have built-in AI features.
• Most modern booking platforms (Cloudbeds, Mews, Sirvoy) have AI for guest comms.

These integrations are powerful because they see your actual data. The drafts are not generic - they reference your specific clients, your specific bookings, your specific accounts.

Pattern 3: Custom automations

For specific workflows that are central to your business, a custom automation can be the right answer. Two examples from real businesses I have worked with.

A North Wales B&B wired up an automation that takes incoming guest enquiries from email, drafts a reply using their saved knowledge base, and surfaces the draft in a queue for the owner to review and send. Time per enquiry: down from 4 minutes to 30 seconds. Hours saved per week: 6 to 8.

A small accountancy practice built an automation that takes raw bookkeeping data, runs an AI first-pass review against their internal checklist, and produces a structured report flagging anomalies. Time per client review: down from 45 minutes to 10. Quality of attention: improved because the obvious items are pre-flagged.

These are custom builds. Tools like Make.com, Zapier, n8n provide the workflow plumbing; the AI is called via API. Costs are typically low hundreds of pounds for the build and a small monthly recurring fee for the AI API usage.

The build vs buy question

Three options when you decide a workflow is worth automating.

Buy off-the-shelf. A specialist tool exists for your need. Subscription cost, faster to start, less control. The vendor evaluation guide has the framework.

Build with no-code tools. Zapier and Make.com let you wire AI calls into existing software. £20-50 a month plus AI API costs. Possible without a developer for simple workflows.

Custom build. A developer or AI consultant builds a bespoke automation. Most expensive upfront but most fitted to your specific work. Worth it for processes that are central to the business and not well-served by off-the-shelf tools.

Most North Wales SMEs do not need custom builds. The combination of built-in AI in existing software plus one or two no-code automations covers most needs.

How to know what to automate

The signal is repetition. If the same workflow happens more than five times a week, it is a candidate. If it happens once a week, it is not.

Three other signals.

The bottleneck is your time, not skill. A task that requires your judgment cannot be automated. A task that requires you to do the typing because no one else has time can be partly automated.

The output has a clear template. Quotes, follow-up emails, summaries, reports - things with consistent structure - automate well. Open-ended creative work does not.

The cost of an error is recoverable. If a wrong AI output goes out and the cost is a 5-minute correction, automation is appropriate. If the cost is significant client damage, keep the human in the loop tightly.

The honest summary

The first 6-12 months of AI adoption is about ad-hoc use - prompts, drafts, individual time savings. The second year is about embedding. By year three, the businesses that committed to this are noticeably more productive than those that stayed on the plateau.

You do not have to do all three patterns at once. Start with the built-in AI in the software you already use. Add one no-code automation when you find a workflow that screams for it. Consider custom builds only when off-the-shelf has been outgrown.

If you would like to walk through where embedding makes sense for your business, that is exactly the kind of work AI strategy consulting and intelligent automation are for. Or book a free discovery call for a 30-minute scope.