Skip to content

AI Automation

Agents do the repeat work. A person signs off.

We build workflows that pull report numbers, add details to new leads and draft content. Each one has a step where a person approves the output before it goes anywhere.

Tools we work in
OpenAI
Claude
Gemini
Zapier
Make
n8n
Supabase
Vercel
Stripe
HubSpot
Notion
Airtable

What is AI automation? · Quick answer

AI automation for marketing means putting language models (Claude, GPT, Gemini) inside the workflows you already run, such as reporting, lead enrichment and content drafts, with a step where a person checks the output. We build these in n8n and connect them to the tools you already use.

The audit

Repeat work run by agents, checked by a person.

What we look for first

  • Reports rebuilt by hand from platform exports
  • Lead details copied from the form to the inbox to the CRM
  • Leads assigned to a rep from a spreadsheet
  • The same dashboard built in several tools
  • Status updates chased in meetings

What we set up

  • Report drafts built from the platform data, checked before they're sent
  • Form leads written to the CRM with the source attached
  • Routing rules that send each lead to the right person
  • One dashboard fed from the same data
  • A log of every run, so you can see what each workflow did

Example workflow

One lead, from form to follow-up.

Example

The same lead, two ways

Switch between the steps done by hand and the same steps with an agent doing the repeat work. It's an example. Your steps will differ.

Lead workflowExample
By hand8 steps
  1. 01A lead comes in through the form
  2. 02Someone spots the notification email
  3. 03They copy the lead's details into the CRM
  4. 04They look up which rep should get it
  5. 05They assign the lead to that rep
  6. 06The rep writes and sends a follow-up
  7. 07Someone adds the lead to the report in Sheets
  8. 08The report goes to the manager
With an agent5 steps
  1. 01A lead comes in through the form
  2. 02An agent checks it against your ICP rules and writes it to the CRM
  3. 03The right rep gets the lead with a short summary
  4. 04A follow-up draft waits for a person to approve it
  5. 05The lead shows up in the report draft, which a person checks

Tools we connect

The tools a workflow links up.

Tools a workflow links upThree tiers

Runs the workflow

n8nMakeZapierOpenAI

Reads and writes

SlackHubSpotGoogle SheetsClaudeSalesforceNotionAirtableWebflow

At the edge

GmailCalendlyStripeTwilioDiscordKlaviyo

The tech stack

What the workflows run on.

Our stack

Tools we build with

We pick the tool that fits the job and the tools you already pay for.

  • n8n

    Open-source workflow automation. We use it first for multi-step workflows, and it can run on your own server.

    Workflow builder Self-hosted AI nodes Error alerts

  • Make

    Visual automation for lead routing and data sync, when your team already works in it.

    Visual builder Webhooks Data transforms API calls

  • Zapier

    Quick trigger-and-action workflows for simple jobs across your tools.

    App library Instant triggers Multi-step Zaps Filters and paths

  • OpenAI / GPT

    Models for drafting, sorting messages and pulling fields out of messy text.

    Lead scoring Report drafts Email drafts Data extraction

  • Claude

    Models for long documents, careful reasoning and code.

    Document analysis Code Reasoning Writing

  • HubSpot

    CRM workflows, lead scoring and pipeline triggers.

    CRM workflows Lead scoring Email sequences Pipeline triggers

  • Slack

    Alerts and approval steps, so a person signs off before an action runs.

    Alert channels Approvals Bot commands Thread updates

  • Airtable

    A flexible table for structured data and workflow triggers.

    Structured data Project tracking Workflow triggers Custom views

  • Notion

    Docs and wikis that a workflow can read from and write to.

    Knowledge base Docs Team wiki Database views

  • Google Sheets

    Simple data pipelines and report tabs.

    Data pipelines Reports Dashboards Sheet updates

  • Perplexity

    Research and source lookup inside a workflow.

    Research Competitor notes Source lookup Fact checks

  • Salesforce

    CRM workflows and lead routing for teams on Salesforce.

    CRM workflows Lead routing Pipeline sync Field updates

Other tools a workflow can connect to

  • Google Sheets
  • Salesforce
  • HubSpot
  • Webflow
  • Klaviyo
  • Google Analytics
  • Google Ads
  • Meta Ads
  • Perplexity
  • Calendly
  • Stripe
  • Twilio
  • Gmail
  • Discord
  • Notion
  • Airtable

Use cases

Jobs we automate.

  • Report drafts

    Checked first

    An agent pulls the platform numbers into a draft. A person checks it before it goes out.

  • Content drafts

    Human-edited

    Blog, social and ad copy drafts that a person edits before anything posts.

  • Lead enrichment

    New contacts

    Company and role details added to new contacts, then scored against your ICP rules.

  • Attribution notes

    From your CRM

    Pipeline and revenue from the CRM set beside the ad platform numbers.

  • Customer ops

    Human approves

    Messages sorted and routed, with reply drafts a person approves.

  • Competitor watch

    Alerts

    An alert when a competitor changes its pricing, pages or ads.

How we work

From audit to live workflow.

  1. 01First

    Operations audit

    We list the tasks your team repeats: reporting, lead routing, content handoffs, enrichment and attribution.

  2. 02Next

    Automation design

    We pick what to automate and what stays with a person. Then we map the triggers, prompts, checks and who gets alerted when a run fails.

  3. 03Then

    Build and run in shadow

    The workflow runs next to the manual process first. We compare its output with the manual result, fix the prompts and write down the rules.

  4. 04Ongoing

    Run and maintain

    We keep checking the output after launch and update prompts or models when the results drift. You see the open tasks and the numbers each week.

FAQ

Questions about AI automation.

Getting Started

What types of processes can be automated with AI?
Lead scoring, email drafts, support replies, data entry and report drafts. If a task repeats and follows rules, it's a good candidate. A person still makes the judgment calls and approves the output.
How does AI automation differ from traditional workflow automation?
Tools like Zapier and Make follow fixed if-then rules. Adding a language model lets a workflow read text, sort it and draft a reply, which fixed rules can't do. It still needs checks, because models make mistakes.
What's your automation process?
We audit your existing workflows, find the bottlenecks, map the data flows and rank the work by impact. Then we build the workflows, run them next to the manual process to compare results, and switch them on. The timeline depends on complexity, and we scope it before work starts.
How do you handle my data?
We use established AI providers (OpenAI, Anthropic) under their business data terms, and we keep sensitive data on your systems where possible. We scope data handling and compliance needs, such as GDPR, before any build starts. If you have specific requirements, tell us on the call.

Pricing & ROI

How much does AI automation cost?
We quote a fixed price for the build after the first call, plus the running costs for the tools and models. We don't publish prices yet.
What's the ROI of AI automation?
It depends on the process. Before a build we write down the hours the task takes today, the error rate and what the workflow would change. We don't promise a payback period.
What if automation doesn't work as expected?
We start with a pilot, check the output against real cases, and change the prompts, rules and data until it holds up. If a workflow doesn't fit your use case, we say so and suggest another route.
Is AI automation a one-time cost or recurring?
Both. The initial build is a one-time cost. Ongoing maintenance, model updates and monitoring are a running cost that depends on system complexity. We agree both figures with you before the work starts.

Technical

What AI models and tools do you use?
We use models from OpenAI and Anthropic depending on the task, and n8n or Make to run the workflows. We connect them to the tools you already use.
How do you ensure the AI makes good decisions?
We set guardrails: a threshold for each decision, a person's review on anything high-stakes and a log of every decision the model makes. A person approves the critical actions. The workflow handles the routine ones.
Can AI automation integrate with our existing systems?
Yes. We connect workflows to tools like Salesforce, HubSpot, Shopify, Slack and Google Workspace, and to custom APIs. The more systems involved, the longer the build.
How do you monitor AI automation performance?
We set up a dashboard with accuracy, error rates, time saved and cost, plus an alert when a workflow fails or its output drifts. You see it each week next to the open tasks, and we adjust the prompts and models when the numbers slip.

Results & Timeline

When will automation go live?
It depends on scope and on how many systems need integrating. We scope the work first and give you a timeline before approval. Complex integrations take longer.
What are typical results from automation?
Results vary by process, so we don't quote averages. Before the build we measure the time a task takes, its error rate and the response time, and after it we report the same numbers against that baseline.
Can you share examples of automation you've built?
Not yet. No case studies are approved for publishing, and nothing goes on this site without a source. On a call we can walk you through the kind of workflows we build and what they changed.

What is included

What an AI automation engagement includes.

What we build depends on the audit. These are the pieces we set up most often. We build in n8n and connect it to HubSpot or Salesforce.

  • n8n on its own server

    Self-hosted n8n with backups. A failed run posts an alert to Slack, so a break doesn't go unnoticed.

  • Fit and intent scoring on inbound leads

    A model scores each lead against your ICP rules and writes the result to your CRM as structured fields.

  • Lead routing

    High-fit leads go to your sales channel with a short summary. The rest go to a queue or get a reply you've approved.

  • Report drafts

    Pulls numbers from Meta Ads, Google Ads, GA4 and HubSpot, flags the big changes and drafts the report. We check it before it's sent.

  • Checks against human review

    We compare a sample of the model's labels with a person's labels and add the misses to the prompt as examples.

  • CRM cleanup

    Duplicate checks across HubSpot or Salesforce, company details added through Clay, and stale leads flagged for you to archive.

  • Outbound reply sorting

    Cold email replies sorted by intent. Interested replies go to the owner, and the rest go to the right follow-up.

  • Content distribution

    When you publish a post, n8n drafts versions for LinkedIn, X, Slack and email. A person approves each one before it posts.

  • Model and cost review

    Token spend per workflow, accuracy drift and prompt updates. We move a workflow to a cheaper model when its output holds up.

Start with the tasks you repeat.

On a 30-minute call we ask which tasks your team repeats. After an audit we write down what we'd automate, what stays with a person and the build plan.

Book a call