A business process often spans forms, emails, document stores and different tools. Before automating it, I map out who enters the data, which checks are needed and who handles exceptions. AI can help with some steps, such as classifying text or preparing a draft; for precise rules, a deterministic integration may be enough. The examples below are possibilities to assess, not results already achieved for a client.

For example, an enquiry from the website can reach the CRM and the person responsible, along with a draft response to review. The workflow's usefulness depends on the manual steps it removes and the review work that remains.

To see where these automations fit into my work, start with the digital solutions page. To discuss a specific case, you can contact me here.

What AI automations are

An automation performs actions in response to an event, known as a trigger. It becomes an AI workflow when one of its steps uses a model, for example to interpret text or prepare a draft. A standard integration may be enough to move data or apply precise rules. Possible triggers include:

  • a form submission;
  • a new email;
  • an uploaded file;
  • a new lead;
  • a CRM update;
  • the publication of an article.

Artificial intelligence comes into play when the workflow goes beyond copying data from A to B and performs tasks such as:

  • classify requests;
  • summarise texts;
  • generate a draft;
  • extract information;
  • create tags or priorities;
  • suggest the next action.

For an SME, the initial assessment covers available data, access, volumes and responsibilities. Timelines, costs and benefits depend on the process and the integrations required.

How I choose the workflow architecture

Technical choices follow the requirements: systems to connect, available access, data handled, volumes and maintenance needs. A dedicated integration may be more suitable than an orchestration platform; the decision comes after the workflow has been clarified.

When designing an integration, I consider these aspects:

  • connect different services;
  • create conditional logic;
  • use APIs;
  • integrate AI models;
  • keep control of the workflow.

For each step, I define responsibilities, data access and what happens if an error occurs. Before sending something externally or making a sensitive change, I can add human approval. The n8n documentation on reviewing AI tool calls provides a practical example of this control. Tool selection comes after workflow design.

Five examples for an SME to consider

1. Leads from your website to email and CRM

An enquiry arrives through the website’s contact form. The workflow can:

  • send an immediate notification;
  • save the lead in a spreadsheet or CRM;
  • classify the message by topic;
  • assign priorities;
  • generate an initial follow-up draft.

The aim is to make enquiries traceable and check whether manual steps decrease. Response times and missed enquiries need to be compared against a measured baseline.

2. New articles and an automated newsletter

When you publish a new article, the system can:

  • update feeds or the content archive;
  • generate an excerpt;
  • prepare the email newsletter;
  • send it to subscribers.

A workflow could monitor the blog’s RSS feed, prepare a draft in the newsletter tool already in use and request a review before sending. It requires a list of subscribers who have opted to receive the newsletter, unsubscribe management and duplicate checks.

3. Incoming email classification

A workflow could suggest email categories, such as sales enquiries, support requests or general messages. Before activation, access permissions, required data and category quality are assessed. Ambiguous cases must be able to go back to a person.

4. Quotes or preparatory documents

A workflow could collect initial project information and prepare an internal draft quote, checklist or summary. Prices, commercial commitments and content intended for external use require review by the person responsible.

5. Automated sales follow-up

If someone completes a form but does not commit straight away, you can plan for:

  • internal reminders;
  • follow-up emails;
  • updating lead status;
  • automated tasks.

The workflow must respect the contact's preferences, avoid duplicates and distinguish internal activities from messages intended for external recipients.

How to measure the results

The comparison starts with the time spent on the current process. A purely illustrative example: 10 minutes per request × 20 requests per month = 200 minutes, or 3 hours and 20 minutes of manual work per month. This is not a promise of savings: any time recovered must be offset by the time spent on checks, exception handling and maintenance, while also accounting for tool costs.

I compare manual effort, errors, exceptions and costs before and after introducing the workflow. For the AI component, I prepare representative cases and check the actual output, including after a model change, as suggested by Anthropic’s work on agent evaluations. The budget accounts for service calls, retries, review and maintenance. The comparison may also show that simplifying the process is the better option.

How to integrate your website, CRM and automation

Integration allows data collected through the site to be used in other business tools. Possible connections include:

  • website form → integration → notification + CRM + internal email;
  • new article → integration → newsletter + content archive;
  • qualified lead → CRM → sales task;
  • quote requested → automatic summary for the team.

I explore process design in more detail on the how I work page.

AI in development and business processes

I use Codex, Claude and Cursor every day in development. I use them to explore solutions, work on code and support reviews. Integrating an AI model into a business process is a separate decision, which needs to be assessed against the specific requirements.

The code and results require review and testing. I remain responsible for technical decisions and for verifying behaviour, including error conditions.

When it makes sense to get started

It is worth getting started when:

  • you spend time on the same tasks every week;
  • your website already generates enquiries or content;
  • you have several channels that do not communicate with each other;
  • you want to respond faster without adding manual work;
  • you want a more structured approach to growth.

It is best to start with a clearly defined step, a measurable success criterion and a clear approach to handling exceptions. The actual complexity determines the timeline and scope.

Conclusion

Automation is useful when it addresses an observable problem and its behaviour can be verified. The starting point is the process: the people involved, the data needed, checks and responsibilities. Only then are integrations and tools selected and results measured.

To understand whether it makes sense for your situation, you can discuss your project or read How to rank on Google as an SME to see how SEO and automation can work together.

Related article

An outdated website: the cost of leaving it unchanged An outdated website is more than a visual issue: it holds your business back and reduces trust and enquiries. Find out when to refresh a business website and what a thorough redesign involves.