What can a 20-person company actually automate?
At twenty people, the automations that reliably pay for themselves are lead intake and routing, quote and proposal generation, client or customer onboarding, invoice capture and matching, recurring reports, and inbound email triage. These share a common shape: they run often, they follow rules, and a person is currently moving data between systems that do not talk to each other.
What does not pay for itself at this size is anything that runs a handful of times a month, anything still changing shape, and anything requiring genuine judgement. A useful test: if nobody could write down the rules, it is not ready to automate yet.
Why twenty people is the turning point
Under about ten people, everyone can see everything. Work gets handed over by turning around in a chair, and the informal system genuinely works.
Somewhere around twenty, that breaks. There are now enough people that nobody has the whole picture, enough volume that things fall through gaps, and enough systems that somebody is spending real hours acting as the connective tissue between them. But there is still no operations function, no internal developer, and nobody whose job is to fix it.
So the work lands on whoever is most conscientious. That person becomes the integration layer — and their time is the most expensive way possible to move data from a form into a CRM.
The six that consistently pay back
1. Lead intake and routing
What happens now: an enquiry arrives through a form or shared inbox. Someone checks it a few times a day, searches the CRM for an existing record, creates or updates it, works out who owns the territory or service line, assigns it, pastes the enquiry text into a note, and sends a holding reply.
What it becomes: the form posts straight into a workflow that matches on company domain, creates or updates the record, applies the routing rules, notifies the owner, and sends the acknowledgement — in under a minute, including overnight and weekends. Anything the rules cannot classify goes to a human instead of being guessed at.
Why it pays: the labour saving is real but secondary. The genuine value is response time. Answering in sixty seconds rather than four hours materially changes conversion on inbound enquiries, and the automation works at 2am.
2. Quote and proposal generation
What happens now: someone copies last month's proposal, changes the client name, updates the pricing table, hopes they caught every instance of the previous client's name, exports a PDF and emails it.
What it becomes: an intake form produces a populated document from a maintained template, logs the opportunity in the CRM, and routes it for signature. Pricing comes from one source rather than from whichever old proposal was nearest.
Why it pays: speed to quote, plus the elimination of a specific and embarrassing error class. Most firms at this size have sent a proposal with the wrong company name in it at least once.
3. Client or customer onboarding
What happens now: a signature triggers a mental checklist. Create the folder. Set up the project from the template. Change the CRM status. Raise the first invoice. Send the welcome email. Add them to the right Slack channel. Five or six systems, done by whoever is free, slightly differently each time.
What it becomes: one trigger, one sequence, every step every time — with the genuinely human parts, like the welcome message, pre-drafted for one-click sending rather than removed.
Why it pays: consistency more than hours. Onboarding is where clients form their impression of whether you are organised, and it is the process most degraded by the person who normally does it being on leave.
4. Invoice capture and matching
What happens now: supplier invoices arrive as PDF attachments. Someone opens each one, reads the numbers, keys them into QuickBooks or Xero, codes them to an account, and checks them against the purchase order.
What it becomes: invoices are read automatically on arrival, line items extracted, matched against the PO, and posted. Anything that fails validation — a price mismatch, a missing PO, an unrecognised supplier — lands in a review queue with the reason attached. Everything else clears itself.
Why it pays: this is usually the single largest pool of hours in a twenty-person business with any physical supply chain. It is also error-prone in a way that costs money directly.
5. Recurring reports
What happens now: three exports, a pivot table, and forty minutes of formatting every Monday to produce numbers that were already sitting in the source systems.
What it becomes: figures pulled from source on a schedule, assembled, and delivered to email or Slack already readable. Or a live dashboard that removes the need for the report entirely.
Why it pays: the hours are modest. The real gain is decisions made on this morning's numbers rather than last week's, and the end of the quiet argument about whose spreadsheet is correct.
6. Inbound email triage
What happens now: a shared inbox that two or three people all half-watch, with the attendant duplication and the messages that everyone assumed someone else had picked up.
What it becomes: each message classified on arrival by intent, key details extracted, priority set, and routed to the right person or queue. This is one of the few places where an AI model genuinely earns its cost, because the input is unstructured language.
Why it pays: it removes the daily sorting session and, more importantly, the dropped message. See whether you need an agent or a workflow for where the line sits.
A four-question test for any process
Before you consider automating anything, answer these. Three or four yeses means it is a strong candidate.
- Does it run at least weekly? Frequency is what turns a small per-run saving into a number worth paying for. Monthly processes rarely justify a build.
- Could someone write down the rules? Not perfectly — but if the person doing it cannot explain when they do X instead of Y, the process is not ready. Document it first.
- Does it move data that already exists somewhere? Re-typing information your systems already hold is the clearest possible signal.
- Is it stable? If the process is changing monthly because the business is still working out how it wants to operate, wait. You will automate a shape you are about to abandon.
What to leave alone at this size
- Anything monthly or rarer. A quarterly board pack taking two hours is four hours a year. A build will not pay back this decade.
- Genuinely relational work. Client check-ins, difficult conversations, negotiation. Automating the appearance of these is worse than not doing them.
- Hiring decisions. Automate the scheduling and the paperwork around hiring, not the assessment.
- Anything you are about to change. New system going in next quarter? Wait for it.
- Your one weird competitive advantage. If a process is genuinely differentiating and depends on judgement, keep the judgement and automate the admin around it.
More on this in what should not be automated.
What order to do them in
Not the biggest one first. The most legible one first.
The first automation in a business does two jobs: it saves time, and it teaches everyone what this actually looks like. Pick something with a clear before and after that a sceptical colleague can see working within a fortnight. Lead routing and onboarding are usually the best candidates, because the improvement is visible daily.
The invoice-processing project — often the largest pool of hours — is a better second or third project, once people trust the approach and you have discovered which of your exceptions actually matter.
One more sequencing rule: if a process depends on data that is currently a mess, fix the data first or automate something else. Automating on top of duplicate records propagates the problem at speed.
Not sure which of these applies to you?
That is exactly what the audit works out. Thirty-minute scoping call first, no charge.