# Dynamics AX 2012 AI Assistant Sizing and Value Calculator

> Source: https://netray.co/tools/dynamics-ax-2012-ai-assistant-sizing-and-value-calculator
> Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems.
> Section: Free Tools

## Short answer

Size the GPUs, retrieval index and monthly value of a private AI assistant for AX 2012 from your user count and question volume.

## Category

on-prem-ai

## Erp

Microsoft Dynamics AX 2012

## Audience

- CIO
- IT manager
- ERP architect
- Finance

## Tool

**Kind:** calculator

**Currency:** $

**Inputs:**

#### Named AX users

**Key:** users

**Type:** number

**Default:** 180

**Min:** 1

**Step:** 10

**Unit:** users

#### Share who use the assistant on a given day

**Key:** dau_pct

**Type:** slider

**Default:** 30

**Min:** 5

**Max:** 100

**Step:** 5

**Unit:** %

#### Questions per active user per day

**Key:** q_per_user

**Type:** number

**Default:** 10

**Min:** 1

**Step:** 1

**Unit:** questions

#### Average prompt plus answer tokens per question

**Key:** tokens_per_q

**Type:** number

**Default:** 3500

**Min:** 200

**Step:** 100

**Unit:** tokens

**Help:** Includes retrieved context. Document questions run higher than database lookups.

#### Documents to index

**Key:** docs

**Type:** number

**Default:** 150000

**Min:** 0

**Step:** 10000

**Unit:** documents

#### Average pages per document

**Key:** pages

**Type:** number

**Default:** 4

**Min:** 1

**Step:** 1

**Unit:** pages

#### Minutes saved per question

**Key:** mins_saved

**Type:** number

**Default:** 4

**Min:** 0

**Step:** 1

**Unit:** min

#### Loaded hourly cost

**Key:** hourly

**Type:** number

**Default:** 42

**Min:** 0

**Step:** 1

**Unit:** per hour

#### GPU monthly cost (amortised plus power)

**Key:** gpu_month

**Type:** number

**Default:** 2200

**Min:** 0

**Step:** 100

**Unit:** per GPU

#### Sustained tokens per second per GPU

**Key:** tps_gpu

**Type:** number

**Default:** 1500

**Min:** 100

**Step:** 100

**Unit:** tokens/s

**Help:** Aggregate throughput with batching for your model and card. Measure it; do not trust a datasheet.

#### One-time build and integration cost

**Key:** build_cost

**Type:** number

**Default:** 60000

**Min:** 0

**Step:** 5000

**Unit:** one-time

**Outputs:**

#### Daily active users

**Key:** active

**Expr:** users * dau_pct / 100

**Format:** integer

#### Questions per day

**Key:** q_day

**Expr:** active * q_per_user

**Format:** integer

#### Tokens per day

**Key:** tokens_day

**Expr:** q_day * tokens_per_q

**Format:** integer

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

#### Peak concurrent requests

**Key:** peak_conc

**Expr:** q_day * 0.2 / 3600 * 20

**Format:** decimal

**Description:** Assumes 20% of daily questions land in the busiest hour and each takes about 20 seconds.

#### GPUs needed (before redundancy)

**Key:** gpus

**Expr:** max(1, ceil(q_day * 0.2 * tokens_per_q / 3600 / tps_gpu))

**Format:** integer

**Primary:** true

**Description:** Peak-hour tokens per second divided by sustained tokens per second per GPU, minimum one.

#### Retrieval index size

**Key:** index_gb

**Expr:** docs * pages * 2 * 4096 * 1.5 / 1000000000

**Format:** decimal

**Description:** Two chunks per page, 1024-dimension float32 vectors, 50% overhead for text and metadata.

#### Hours saved per month

**Key:** hours_month

**Expr:** q_day * 22 * mins_saved / 60

**Format:** hours

#### Monthly value of time saved

**Key:** value

**Expr:** hours_month * hourly

**Format:** currency

#### Monthly value minus GPU cost

**Key:** net

**Expr:** value - gpus * gpu_month

**Format:** currency

**Primary:** true

#### Simple payback

**Key:** payback

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

**Format:** months

**Description:** One-time cost divided by monthly net value. Zero means no payback at these inputs.

**Chart:**

#### Monthly net value vs named users

**Type:** line

**X:** users

**From:** 20

**To:** 1000

**Steps:** 25

**Series:**

- net

**Disclaimer:** Planning estimate. Measure real throughput on your model and hardware, add redundancy for production, and treat time-saved figures as hypotheses until a pilot confirms them.

## Content

**Intro:** Use this calculator to turn your AX 2012 user count and question volume into a GPU count, token load, index size and a simple payback. Every formula is shown so finance and IT can challenge the assumptions.

**Sections:**

#### How to use it

**Body:** Enter named users and the share who actually use the assistant in a day. Estimate questions per user from the pilot or from how often people ask colleagues for data today. Tokens per question include retrieved context, so document-heavy use needs a larger number. Replace the default throughput with a figure measured on your own card.

**Bullets:**

- Start with 20 to 30% daily active users
- Use a measured tokens-per-second value
- Add redundancy for production
- Run low, expected and high scenarios

#### The formulas

**Body:** Questions per day equals named users times active share times questions each. Peak-hour tokens per second equals 20% of daily questions times tokens per question divided by 3600. GPUs equal that peak divided by sustained throughput, rounded up. Index size multiplies pages by two chunks per page by 4096 bytes per vector with 50% overhead. Hours saved use 22 working days.

#### Reading the result

**Body:** The GPU count is a floor for capacity, not a recommendation to run a single card in production. The payback figure is only as good as minutes saved per question, which is the weakest input; test it with a pilot group and timed tasks. If net value is negative at realistic inputs, a simpler tool such as reports or Power BI may serve you better.

**Cta:** If you want these numbers checked against your AX 2012 environment, Netray can run a short sizing review with your IT team.

**Faq:**

#### Why 20% of questions in the busiest hour?

**Answer:** It is a conservative planning assumption for office workloads where use clusters in the morning. Replace it with your own peak share if you have usage data.

#### Does throughput depend on the model?

**Answer:** Strongly. Model size, quantisation, context length and batch size all change tokens per second. Measure on the model you intend to run.

#### Why is the index so small?

**Answer:** Vectors are compact. Most storage in real deployments is the original documents, not the index, which this tool does not size.

#### Can I trust the payback number?

**Answer:** Treat it as a hypothesis. Time saved per question is an estimate and adoption varies. Validate with a pilot before committing budget.

## Search terms

AX 2012 AI sizing calculator, private LLM GPU count calculator, AI assistant ROI calculator ERP, on-prem AI payback calculator, LLM concurrent requests calculator, vector index size estimate

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