> ## Documentation Index
> Fetch the complete documentation index at: https://irisdocs.prescientlabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Cost & Benefit Modelling

> Build a transparent commercial model that separates baseline, expected change, implementation cost and ongoing cost.

A useful business case makes the economics easy to challenge.

## Start with the baseline

Record the current state before modelling the future:

* annual transaction or case volume;
* current time per case;
* error or rework rate;
* delay or queue time;
* loaded labour cost;
* current software or supplier cost;
* current conversion, churn or service metrics where relevant.

For every important input, record its source and confidence level.

## Model the expected change

Translate the intervention into operational assumptions such as:

* minutes saved per case;
* percentage of cases affected;
* adoption rate;
* reduction in rework;
* increase in throughput;
* reduction in cycle time;
* improvement in conversion.

Do not jump straight from “AI is introduced” to a financial number. Explain the mechanism that creates value.

## Separate value types

### Capacity value

Useful time becomes available for other work. This is not automatically a cash saving.

### Cost avoided

The change prevents a future cost, such as an additional hire or external service.

### Cash saving

A cost genuinely leaves the P\&L.

### Revenue value

The change creates more sales, conversion or retention through a defensible causal mechanism.

## Include the full cost

Model:

* implementation;
* integration;
* licences and infrastructure;
* testing;
* training and change;
* ongoing support;
* internal client effort where material.

## Build scenarios

Use conservative, base and upside cases. Vary only the assumptions that materially drive the result.

<Warning>
  Avoid adding multiple benefit categories that describe the same underlying improvement. Capacity, avoided hiring and cash saving can easily be double-counted.
</Warning>