# Cost
> **What you will learn**
> How to judge the second question, why this area suits a first project, and why
> reading "cost reduction" as "headcount reduction" wrecks transformations.
## The question
> **Do the operating costs of making the product go down?**
## What falls under this principle
### 1. Automating repetitive work to cut labour cost
The most direct form. If human time goes into something that repeats the same
way every month, it is a candidate.
A typical example:
```
Work: receive supplier invoices and enter them into accounting
Today: 400/month × 6 minutes each = 40 hours
one person spends a quarter of the month on this
Errors: 12/month on average (wrong amount or account) → rework
```
That table matters because **all three numbers are there** — volume, time,
errors. You can judge later whether it worked.
### 2. Moving senior people onto high-value work
This is **the essence of cost reduction**, and the most frequently missed.
An accountant with ten years of experience spends 40 hours a month on data
entry. What they are good at is not entry — it is **spotting anomalies and
making judgements.** Remove the entry and the same salary produces more valuable
output.
> Labour cost did not fall. **What you get for the same labour cost went up.**
> On the books the cost is unchanged; in substance it is a reduction.
### 3. Cutting rework cost by reducing errors
Rework is an **invisible cost.** It appears as a line item nowhere. In practice
it is substantial.
```
What one wrong entry causes:
discovery (usually days later) → 15 min
tracing the cause → 30 min
correcting it and related documents → 30 min
contacting the supplier → 20 min
─────────────────────────────────────────────
about 1.6 hours × 12/month = 19 hours/month
```
You were looking at 40 hours of entry, and **there were another 19 hours of
rework.** The real cost is 59 hours.
**Impact: 70–90% reduction in operating cost**
## Why it suits a first project
Of the three principles its impact is in the middle, but **it is first in
sequence.** Three reasons.
| Reason | Explanation |
|---|---|
| **Fast to prove** | time and volume are immediately measurable. Results next month |
| **Low data prerequisite** | it is work you already do, so the records exist |
| **Little opposition** | few people object to "let's reduce this tedious job" |
To keep a transformation going, an organisation needs **one visible early
result.** Cost reduction usually plays that role.
## The common trap — being read as "headcount reduction"
Let that reading spread and **the transformation quietly stops.**
### What happens
```mermaid
graph TD
A["Cost reduction project announced"] --> B["The floor: 'are they cutting us?'"]
B --> C["No cooperation
information about the work withheld"]
C --> D["Nobody knows how the work really runs
so the wrong thing gets built"]
D --> E["Nobody uses it"]
```
AI work **cannot be built without cooperation from the people doing the job.**
They are the only ones who know how it actually runs. Nobody cooperates in
eliminating their own job.
### How to handle it
**1. Say where the freed time goes, first.**
"We're cutting 40 hours of entry" sounds like 40 hours of people are surplus.
Say "we're cutting 40 hours of entry and spending that time on anomaly review."
**2. Put the people doing the job into the design.**
When the person doing the work participates in designing the project, they stop
being **someone whose job is being removed** and become **someone redesigning
their own job.**
**3. Choose a first project nobody wants to do.**
Start with data entry, file collation, copy-and-paste — **work nobody will miss**
— and there is almost no resistance.
> This is not a political technique; it is **a probability-of-success issue.**
> If the floor does not cooperate, the project fails no matter how good the
> technology is.
## The question that finds candidates
Ask each team:
> **"Of the things you do now, what would you not miss at all if it vanished?"**
Far better answers than "what should we use AI for?" People know exactly which
parts of their job they hate.
## Real cases
- [invoice processing automation](/ax-consulting/cases/invoice-processing-automation)
- [HR onboarding automation](/ax-consulting/cases/hr-onboarding-automation)
- [resume screening automation](/ax-consulting/cases/resume-screening-automation)
- [marketing campaign automation](/ax-consulting/cases/marketing-campaign-automation)
## Common misconceptions
### "If payroll doesn't actually drop, is it really a reduction?"
A reduction does not require a line item to shrink. If the same people produce
more valuable output, **cost per unit of output** fell. And in practice this
form is more common and more sustainable.
### "Isn't 70–90% an exaggerated figure?"
That figure is about **the specific work step**, not a department's total cost.
If 40 hours of invoice entry becomes 4, that step fell 90% — the accounting
team's total cost did not. **Always state what the percentage is a percentage
of** and there will be no misunderstanding later.
---
## Check yourself
**1. Why is cost reduction middling in impact but first in sequence?**
Answer
Because it is **fast to prove, has a low data prerequisite, and attracts little
opposition.** Time and volume are immediately measurable, so results arrive next
month. An organisation continuing a transformation needs one visible early
result, and this fits that role best.
**2. Why was the real cost in the invoice example 59 hours rather than 40?**
Answer
Because **19 hours of rework** were hidden. Each wrong entry costs about 1.6
hours in discovery, tracing, correction, and contact, and there were 12 a month.
Rework appears as a line item nowhere, so it is easy to miss.
**3. What must you say first to get cooperation on a cost project?**
Answer
**Where the freed time goes.** "We're cutting 40 hours" alone sounds like
headcount reduction. You have to add "and spending that time on anomaly review."
And bringing the person doing the job into the design turns them from someone
losing a job into someone redesigning one.
---
Next, the area with the largest impact and the most prerequisites →
[Revenue and margin](/guide/ax-revenue)