# Worked counts (/docs/method/worked-counts)

---
type: Document
title: Worked counts
description: The same counting method applied to five different businesses, each with its own step table, its own number and its own human gates.
status: stable
order: 4
generated: { by: "process:claude-code", at: 2026-08-27T12:00:00Z }
verified:
  - { by: "human:rrizwan98", at: 2026-08-28T14:30:00Z }
ksor:
  audience: [public]
  owner: human:rrizwan98
  approval: { by: "human:rrizwan98", at: 2026-08-28T14:30:00Z }
trust_tier: human-reviewed
build_id: sha256:d9f989bee8fa4c15cf0f7f1572f9e9a18305ef4e2ce0c3cfb6a5a421ead6dc1d
dirty: true
---

Five runs from five different businesses, counted with the
[same procedure](count-your-own-workflow.md). They are here so the method is
visible as a method: the tables differ, the numbers differ, and the reasoning
is identical.

These five are **illustrations, not client work**. They show how a run gets
tagged, not what any particular business paid for. Your own run will have
different steps, and it is worth counting rather than copying the closest one
here.

Every percentage below is a **could be**: the share of steps a system could own
once the work is done. In each of these businesses today is 0%, because none of
them has built anything yet. The two numbers are defined on
[count your own workflow](count-your-own-workflow.md).

## 1. A web-design studio: landing page delivery

| # | Step | Tag |
| --- | --- | --- |
| 01 | Enquiry arrives, goal and budget confirmed | DECISION |
| 02 | Brand, audience and reference questions asked | AI |
| 03 | Competitor and market scan for the sector | AI |
| 04 | Copy and section plan drafted | AI |
| 05 | Design direction chosen | DECISION |
| 06 | Page built to the chosen direction | AI |
| 07 | Responsive and accessibility checks | AI |
| 08 | Fix loop until the checks pass | AI |
| 09 | Client revision request read and applied | AI |
| 10 | Client sign-off | DECISION |
| 11 | Deploy, handover note, analytics wired | AI |

**8 AI, 3 DECISION: 73%.** Gates at 05 and 10. The taste call stays human, the
production does not.

## 2. An agentic AI service: building a client's agent

| # | Step | Tag |
| --- | --- | --- |
| 01 | Discovery call, problem and constraints agreed | DECISION |
| 02 | Follow-up questions on data, tools and red lines | AI |
| 03 | Architecture plan and acceptance criteria written | AI |
| 04 | Plan approved | DECISION |
| 05 | Tests written before the code | AI |
| 06 | Agent implemented until the tests pass | AI |
| 07 | Evaluation set built and scored | AI |
| 08 | Guardrails and permissions checked | AI |
| 09 | Packaging and deployment prepared | AI |
| 10 | Delivery report written | AI |
| 11 | Go-live approved by the client | DECISION |

**8 AI, 3 DECISION: 73%.** Gates at 04 and 11. Note that the scope call at 01
is a decision even though it looks like a conversation: it sets price and
liability.

## 3. A recruitment agency: screening for one role

| # | Step | Tag |
| --- | --- | --- |
| 01 | Role brief taken from the client | DECISION |
| 02 | Scorecard written from the brief | AI |
| 03 | Applications gathered from every channel | AI |
| 04 | Each application scored against the scorecard | AI |
| 05 | Missing information requested from candidates | AI |
| 06 | Shortlist assembled with reasons | AI |
| 07 | Shortlist chosen | DECISION |
| 08 | Interviews scheduled | AI |
| 09 | Interview held | DECISION |
| 10 | Feedback written and sent | AI |
| 11 | Offer decision | DECISION |

**7 AI, 4 DECISION: 64%.** Lower than the first two, correctly: judging a
person is a decision, and three of the four gates involve one.

## 4. An e-commerce team: adding a supplier's catalogue

| # | Step | Tag |
| --- | --- | --- |
| 01 | Supplier feed received | AI |
| 02 | Feed validated against the schema | AI |
| 03 | Products matched against existing catalogue | AI |
| 04 | Duplicates and conflicts flagged | AI |
| 05 | Titles and descriptions written to house style | AI |
| 06 | Images checked, resized and named | AI |
| 07 | Pricing rules applied | AI |
| 08 | Margin exceptions reviewed | DECISION |
| 09 | Published to the store | AI |
| 10 | Spot check of a sample | DECISION |

**8 AI, 2 DECISION: 80%.** The highest of the five, because almost every step
is repeatable and checkable. Margin is a decision because it is price.

## 5. A SaaS company: first-line support

| # | Step | Tag |
| --- | --- | --- |
| 01 | Ticket arrives | AI |
| 02 | Classified and prioritised | AI |
| 03 | Account history and logs gathered | AI |
| 04 | Known-issue match attempted | AI |
| 05 | Reply drafted from the documented answer | AI |
| 06 | Reply sent for routine cases | AI |
| 07 | Sensitive cases escalated | DECISION |
| 08 | Refund or credit decided | DECISION |
| 09 | Resolution logged and the article updated | AI |

**7 AI, 2 DECISION: 78%.** Step 07 is a gate rather than a step: the system
recognises the exception and stops, which is what makes 06 safe.

## What the five have in common

| | Number | Human steps |
| --- | --- | --- |
| Web design | 73% | brief, direction, sign-off |
| Agentic AI service | 73% | scope, plan approval, go-live |
| Recruitment | 64% | brief, shortlist, interview, offer |
| E-commerce catalogue | 80% | margin exceptions, spot check |
| SaaS support | 78% | escalation, refund |

Every human step is a **decision**: taste, judgement of a person, price, or a
final yes. Not one of them is typing. That is the whole result of the exercise,
and it is why the number lands between 60% and 80% in most businesses rather
than at 100% or at zero.

Now count your own: [count your own workflow](count-your-own-workflow.md).