— AI for small business / Automation
Take the copying and pasting out of the week.
Most small businesses lose a day a week to moving information from one place to another: email to spreadsheet, receipt to bookkeeping, notes to CRM. We automate that work, use a language model only where it earns its place, and keep a person approving anything a customer or the tax office will see.
What changes in a normal week.
What we automate.
-
Inbox triage
Incoming email and web-form enquiries classified, the useful details pulled out (name, need, location, timeline, budget signals), logged to your CRM or sheet, and a reply drafted for review.
-
Quote and proposal drafts
Drafts built from your own price list, standard terms, and past quotes for similar work. The model writes; your price book sets the numbers; a person approves before anything is sent.
-
Receipts, bills, and invoices
Documents read and entered into QuickBooks Online or Xero with a suggested category, matched to the right job, and queued for approval. Nothing is filed or paid automatically.
-
CRM and job-tool upkeep
Call notes, meeting summaries, and email threads turned into updated contact records, next steps, and tasks, so the CRM reflects reality without anyone typing into it.
-
Follow-ups that actually go out
Review requests after a finished job, reminders for quotes that went quiet, and renewal notices, sent on a schedule you set and paused the moment a customer replies.
-
Reports nobody has time to build
A Monday summary of last week: enquiries, quotes sent and won, overdue invoices, and anything stuck. Pulled from the systems you already use.
How we build it.
One workflow at a time. The first one runs for a month and gets measured before we build the second.
-
Map the week
We sit with the people doing the work and time the repetitive parts. The automation with the most hours behind it goes first.
-
Pick the plainest tool
Fixed rules get plain automation. A model is used only where the input is messy, such as free-text email or a crumpled receipt.
-
Build with an approval step
Anything a customer, supplier, or the tax office will see waits for a person to approve it. That step comes out only when you decide it has earned trust.
-
Measure for a month
Minutes saved, how often a person changed the draft, and errors caught. You get the numbers in writing.
-
Hand over or keep running
You get the documentation and the logins. We can keep it maintained under ongoing care, or your team can own it.
Automations break quietly. Ours report it.
A broken automation keeps running. It just files invoices in the wrong place or stops sending follow-ups, and you find out a month later. Every workflow we build is watched.
- A person approves anything that leaves the business until you decide otherwise.
- Every run is logged, and failures alert a named person rather than failing silently.
- Money never moves automatically: no auto-paid bills, no auto-issued refunds.
- Client data goes only to providers on business terms that exclude training on your data.
- Each workflow has a one-page description of what it does and how to switch it off.
- Credentials live in a password manager or the platform’s secret store, never in a spreadsheet.
Is it for you?
A good fit when
- Someone on your team spends hours a week retyping information between systems.
- The work follows a pattern, even if the inputs are messy.
- Your core tools have an API or integration: Microsoft 365, Google Workspace, QuickBooks, Xero, HubSpot, Jobber, and most modern tools do.
- You can name the person who will approve drafts for the first month.
Probably not yet when
- The process changes every time, or nobody agrees on how it should work. We fix the process first.
- The work happens a few times a month. A checklist is cheaper.
- You want it fully hands-off from day one on anything involving money.
What it costs.
Each workflow is quoted as a fixed-price build, with any monthly platform or model costs listed separately so you can see them. We will also tell you when a hosted tool’s per-task pricing will cost more than a small custom build within a year.
What moves the number
- How many systems the workflow touches, and whether they have usable APIs.
- Volume: tens of documents a week or thousands.
- Hosted tools (Zapier, Make, Power Automate) versus self-hosted n8n or custom code.
- How messy the inputs are: clean web forms or scanned handwriting.
- Whether we maintain it afterward or hand it to your team.
— Questions
Before you write.
Do you use Zapier, Make, or n8n?
Whichever fits. Zapier and Make are quick to start and easy for your team to read. Power Automate makes sense in Microsoft 365 shops. Self-hosted n8n or a small custom service wins when volume makes per-task pricing expensive or the data should stay in your own account. We explain the trade-off in the quote.
Where does AI actually help, versus plain automation?
Plain automation handles anything with fixed rules: when a form is submitted, add a row. A language model helps when the input is unstructured: a rambling email, a photo of a receipt, a call transcript. We use the cheaper, more predictable option wherever it works.
Will it work with QuickBooks or Xero?
Yes, both have solid APIs. We draft entries for approval rather than posting them directly, so your bookkeeper stays in control and nothing reaches your books unreviewed.
What happens when something changes, like a new form or a new price list?
Small changes are part of maintenance if we look after it. If your team owns it, the documentation shows where each rule and prompt lives so they can change it without us.
How do we know it is saving time?
We measure before and after: minutes per task, tasks per week, and how often the draft needed changes. If a workflow is not saving time after a month, we say so and switch it off.
The other three AI services.
- AI receptionist A voice and chat agent on your own number and site that answers, books, takes orders, and hands the hard calls to a person.
- Copilot, ChatGPT & Claude rollout The right assistant for your team, set up on business plans with sane data controls, a written use policy, and training built around real jobs.
- Private AI on your documents An assistant that answers from your own policies, manuals, and past work, cites the page it used, and stays inside your account.
Further reading Five AI automations worth building firstWhen not to use an LLMGetting reliable structured output from an LLM
— Engage
Which job would you hand off first?
Tell us the task, roughly how often it happens, and which tools it touches. We reply in writing within one business day with whether it is worth automating and what it would take.