# QAD EE AI Assistant Sizing and Value Calculator

> Source: https://netray.co/tools/qad-ee-ai-assistant-sizing-and-value-calculator
> Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems.
> Section: Free Tools

## Short answer

Estimate peak concurrency, GPU count, index size, hours saved and payback for a private AI assistant on QAD EE.

## Category

on-prem-ai

## Erp

QAD Enterprise Edition

## Audience

- CIO
- IT manager
- Plant manager
- Controller

## Tool

**Kind:** calculator

**Currency:** $

**Inputs:**

#### Named users

**Key:** users

**Type:** number

**Default:** 220

**Min:** 1

**Step:** 5

**Unit:** users

**Help:** Licensed or named QAD EE users who could use the assistant.

#### Share of users active on a typical day

**Key:** dau

**Type:** slider

**Default:** 35

**Min:** 5

**Max:** 100

**Step:** 1

**Unit:** %

#### Questions per active user per day

**Key:** q

**Type:** number

**Default:** 8

**Min:** 0

**Step:** 1

**Unit:** questions

#### Average prompt plus answer size

**Key:** tokens

**Type:** number

**Default:** 1800

**Min:** 100

**Step:** 100

**Unit:** tokens

**Help:** Includes retrieved rows or document chunks placed in the prompt.

#### Average seconds to produce an answer

**Key:** sec

**Type:** number

**Default:** 12

**Min:** 1

**Step:** 1

**Unit:** seconds

**Help:** Wall time per question under load; used for concurrency.

#### Share of daily questions in the busiest hour

**Key:** peak

**Type:** slider

**Default:** 20

**Min:** 5

**Max:** 60

**Step:** 1

**Unit:** %

#### Aggregate tokens per second per GPU

**Key:** tps

**Type:** number

**Default:** 1200

**Min:** 50

**Step:** 50

**Unit:** tokens/s

**Help:** Batched throughput of your model on your GPU. Measure with your own model; this is an assumption.

#### Documents to index

**Key:** docs

**Type:** number

**Default:** 250000

**Min:** 0

**Step:** 1000

**Unit:** documents

#### Average pages per document

**Key:** pages

**Type:** number

**Default:** 3

**Min:** 1

**Step:** 1

**Unit:** pages

#### Minutes saved per question

**Key:** mins

**Type:** number

**Default:** 4

**Min:** 0

**Step:** 0.5

**Unit:** minutes

#### Loaded hourly cost

**Key:** rate

**Type:** number

**Default:** 48

**Min:** 0

**Step:** 1

**Unit:** $/hr

#### Monthly cost per GPU

**Key:** gpu

**Type:** number

**Default:** 1800

**Min:** 0

**Step:** 50

**Unit:** $/month

**Help:** Depreciation, power, space and support, or rental rate.

#### One-time build cost

**Key:** setup

**Type:** number

**Default:** 60000

**Min:** 0

**Step:** 1000

**Unit:** $

**Outputs:**

#### Active users per day

**Key:** active

**Expr:** users * dau / 100

**Format:** integer

#### Questions per day

**Key:** qpd

**Expr:** active * q

**Format:** integer

#### Tokens per day

**Key:** tokens_day

**Expr:** qpd * tokens

**Format:** number

**Description:** Questions per day times average tokens per question.

#### Peak concurrent requests

**Key:** peak_conc

**Expr:** max(1, ceil(qpd * peak / 100 / 3600 * sec))

**Format:** integer

**Description:** Little's law: busiest-hour arrival rate times seconds per answer.

#### GPUs needed (before redundancy)

**Key:** gpus

**Expr:** max(1, ceil(qpd * peak / 100 * tokens / 3600 / tps))

**Format:** integer

**Primary:** true

**Description:** Peak tokens per second divided by assumed tokens per second per GPU.

#### Vector index size

**Key:** index_gb

**Expr:** docs * pages * 2 * 6 / 1000000

**Format:** decimal

**Description:** About 2 chunks per page at roughly 6 KB each including embedding and metadata.

#### Hours saved per month

**Key:** hours

**Expr:** qpd * mins / 60 * 21

**Format:** hours

#### Monthly value of time saved

**Key:** value

**Expr:** hours * rate

**Format:** currency

#### Monthly GPU run cost

**Key:** run_cost

**Expr:** (gpus + 1) * gpu

**Format:** currency

**Description:** Includes one spare GPU for redundancy.

#### Net monthly benefit

**Key:** net

**Expr:** value - run_cost

**Format:** currency

**Primary:** true

#### Simple payback

**Key:** payback

**Expr:** if(gt(net, 0), setup / net, 0)

**Format:** months

**Description:** Zero means no payback at these inputs.

**Chart:**

#### Monthly value and GPU cost vs named users

**Type:** line

**X:** users

**From:** 10

**To:** 660

**Steps:** 30

**Series:**

- value
- run_cost

**Disclaimer:** Planning estimate only. Tokens per second per GPU depends on model, quantization, context length and serving settings; measure your own. Time-saved figures are assumptions to be validated in a pilot.

## Content

**Intro:** This calculator estimates what a private AI assistant on QAD EE needs in GPUs and index storage, and what it might be worth in saved time. Enter your usage and assumptions; the formulas are stated so IT and finance can check them.

**Sections:**

#### How to use it

**Body:** Start with named users and the share active on a typical day, then estimate questions per active user. Set average prompt plus answer tokens high enough to include retrieved rows or document chunks, since those dominate. Enter the tokens per second your chosen model achieves on your GPU; if you do not know it, run a short benchmark in vLLM or Ollama.

**Bullets:**

- Defaults reflect a typical QAD EE site
- Change one input at a time
- Use the chart to see how cost and value scale with users

#### The formulas

**Body:** Questions per day equals users times active share times questions per user. Peak concurrency is the busiest-hour questions divided by 3600 times seconds per answer. GPUs equal peak tokens per second divided by tokens per second per GPU, rounded up, with one spare added to cost. Index size assumes about 2 chunks per page at 6 KB each. Hours saved use 21 working days per month.

**Bullets:**

- Concurrency = (questions per hour / 3600) x seconds per answer
- GPUs = ceil(peak tokens per second / GPU throughput)
- Value = hours saved x loaded hourly cost

#### Reading the results

**Body:** If net monthly benefit is negative, either usage is too low for the hardware or the time-saved assumption is optimistic. Small deployments often fit on one GPU, so the spare GPU doubles the cost; a smaller model or shared hardware can change the answer. Treat payback as directional and confirm minutes saved by timing a few real tasks in a pilot.

#### Common mistakes

**Body:** Sizing on named users instead of concurrency overstates GPUs. Ignoring retrieved context understates tokens. Counting every minute saved as cash overstates value; time saved only becomes money if it is redeployed.

**Bullets:**

- Assuming all saved time is cashable
- Using vendor peak throughput instead of measured throughput
- Forgetting redundancy and maintenance
- Forgetting document volume when sizing storage

**Cta:** Netray can run a short pilot on your own QAD EE data to replace these assumptions with measured numbers. Ask for a review of your sizing if you want a second opinion.

**Faq:**

#### How is the GPU count calculated?

**Answer:** Peak tokens per second needed in the busiest hour, divided by the aggregate tokens per second one GPU delivers with batching, rounded up to at least one. Measure throughput with your own model; the default is an assumption.

#### Why is a spare GPU included in cost?

**Answer:** A single GPU is a single point of failure for the assistant. The cost line adds one spare so you see a redundant design; remove it mentally if you accept downtime or run a smaller pilot.

#### Is the value figure real savings?

**Answer:** It is the loaded cost of time saved. It becomes cash only if the time is redeployed or avoids hiring or overtime. Validate minutes saved per question with a pilot.

#### How accurate is the index size?

**Answer:** It is a rough estimate based on about 6 KB per chunk and 2 chunks per page. Embedding dimension, chunking and metadata change it, but for most deployments it is within a factor of two.

## Search terms

QAD EE AI sizing, how many GPUs for on-prem LLM, on-prem AI assistant ROI calculator, LLM concurrency calculator, AI payback calculator ERP, vLLM GPU sizing, private AI cost QAD EE, RAG index size estimator

---
Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json