"Systemise your business" usually means "write down how everything works". It produces a shared drive full of procedures and a business that still runs on the founder, because the documents describe work that people have to remember to do. The version that works starts from a different question: of everything that recurs in this business, what should a system do, and what genuinely needs a person? Only the second list needs documenting. The first list needs building.
Step 1: Audit the recurring work
For two weeks, every recurring task gets a line: what it is, minutes each time, times a month, who does it, which systems it touches. Not the one-offs; the things that come back. Most founders find between 40 and 80 of them, and that the founder personally does a third. Our Scaleability Audit does this in 48 hours with the founder; the task list is the deliverable.
Step 2: Score each task
Two scores per line. Value: hours a month it consumes. Risk: what happens if it is done wrong (a mis-sent client email is high; a mis-filed receipt is low). Then three buckets:
- Automate: high value, low risk, consistent inputs. Reporting, chasing, data entry, scheduling, status updates.
- Automate with a human gate: high value, higher risk. Invoicing, client communications, anything touching money. A system prepares; a person approves.
- Keep with a person, and document it: judgement calls, relationships, the sales conversation. This is the only bucket that needs an SOP.
Step 3: Build in layers, not all at once
The order we use for every business, including our own:
- Context. Write down, once, what the business is, who it serves, what the numbers mean. Every automation and every AI draft reads from this, so it stops being generic.
- Data. Pull the live numbers from every system into one place daily. You cannot delegate what you cannot see.
- Intelligence. A daily brief: what moved, what needs a decision, what is slipping. The founder reads one message instead of checking seven tools.
- Automation. The "automate" bucket, one task at a time, highest value first. Each one is bandwidth returned for good.
- Build. The freed time goes on growth, or on life. This is the point of the previous four.
That sequence is what an AI Operating System is. We run Anco AI on one: the bookkeeping queries, the receipts, the proposals and their signatures, the content pipeline and the morning brief are all in the "automate" bucket and have been for months.
Step 4: Document only what a person still does
Now the SOPs are short, because most of the recurring work is no longer done by a person. What remains is the judgement work, and the documentation for that is a checklist and a decision rule, not a 12-page procedure. Keep it where the work happens (in the CRM, in the task template, in the brief), not in a folder.
Step 5: Measure the one number
How many hours can you step away before something falls over? Call it away-from-desk autonomy. Measure it honestly now (for most founders it is a working day, at best), then after each layer. When it reaches a week, the business is systemised in the only sense that matters. Two supporting numbers: the percentage of recurring tasks that run without a person, and revenue per team member, which rises when the team stops doing work a system should.
What goes wrong
- Documenting first. You write down a process, then automate it, then the document is wrong. Automate first; document what is left.
- Automating a mess. A process that is done differently by every person cannot be automated. Standardise it for a month, then build.
- Buying a platform. A new tool does not systemise a business; it moves the manual work somewhere more expensive. Connect what you have.
- Skipping the human gate. Fully automated money or client communication is how a small error becomes a client loss.
A real example: how we systemised our own bookkeeping
Every week our accountant sends a bookkeeping-queries spreadsheet: unexplained bank lines, "invoice needed" rows, "receipt?" rows. For a long time a person read it, raised the sales invoices in Xero by hand, hunted the mailbox for receipts, and typed the answers back. Recurring, universal, mostly mechanical: the definition of the "automate" bucket. Now a system reads the sheet, drafts a plan (these invoices will be raised, these receipts were found and will be filed, these three rows need a human), a person reads the plan in plain English and approves it, and the system does the rest and writes back into the sheet what it did. The weekly job went from an afternoon to the few minutes it takes to read the plan.
What made it work was not the AI. It was the human gate (nothing touches the ledger until the plan is approved), exact matching for money (an early version matched £8.00 against anything containing an 8, never again), and the write-back so nobody redoes the work. The same three rules apply to every automation in the "with a human gate" bucket.
The task audit, as a table
Six columns, one row per recurring task. It is the only template the method needs.
| Task | Minutes each | Times a month | Who | Systems | Risk if wrong | Bucket |
|---|---|---|---|---|---|---|
| Monthly client report | 120 | 10 | Account manager | Meta, GA4, CRM | Low | Automate |
| Raise invoices | 15 | 20 | Founder | Xero | High | Automate with gate |
| Chase documents | 10 | 40 | Admin | Low | Automate | |
| Scope a proposal | 90 | 4 | Founder | Docs, CRM | High | Keep, document |
Multiply minutes by times a month, sort descending, and the first three builds choose themselves.
If you want the audit done with you, that is the discovery call and then the Scaleability Audit; the AI consultancy page has the prices and the method in full.
Founder, Anco AI. Builds the systems this site talks about, and runs his own business on them.
