# MAPICS and Infor XA AI Assistant Sizing and Value Calculator

> Source: https://netray.co/tools/mapics-xa-ibm-i-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 GPU count, token volume, index size, hours saved and payback for a private on-prem AI assistant on MAPICS and Infor XA.

## Category

on-prem-ai

## Erp

Infor XA

## Audience

- CIO
- IT manager
- Finance controller
- ERP consultant

## Tool

**Kind:** calculator

**Currency:** $

**Inputs:**

#### Named users

**Key:** users

**Type:** number

**Default:** 250

**Min:** 1

**Step:** 5

**Unit:** users

**Help:** Licensed or named MAPICS and Infor XA users who could use the assistant.

#### Share active on a typical day

**Key:** active

**Type:** slider

**Default:** 35

**Min:** 5

**Max:** 100

**Step:** 5

**Unit:** %

**Help:** Percent of named users who ask at least one question per day.

#### Questions per active user per day

**Key:** q

**Type:** number

**Default:** 8

**Min:** 1

**Step:** 1

**Unit:** questions

#### Average prompt plus answer size

**Key:** tokens

**Type:** number

**Default:** 2200

**Min:** 200

**Step:** 100

**Unit:** tokens

**Help:** Include retrieved context. Typical range 1000 to 3000.

#### Documents to index

**Key:** docs

**Type:** number

**Default:** 80000

**Min:** 0

**Step:** 1000

**Unit:** documents

#### Average pages per document

**Key:** pages

**Type:** number

**Default:** 5

**Min:** 1

**Step:** 1

**Unit:** pages

#### Minutes saved per question

**Key:** mins

**Type:** number

**Default:** 4

**Min:** 0

**Step:** 0.5

**Unit:** minutes

**Help:** Measure in a pilot; this is the most uncertain input.

#### Loaded hourly cost

**Key:** rate

**Type:** number

**Default:** 42

**Min:** 0

**Step:** 1

**Unit:** per hour

#### GPU monthly cost

**Key:** gpucost

**Type:** number

**Default:** 2500

**Min:** 0

**Step:** 100

**Unit:** per GPU per month

**Help:** Amortised purchase plus power, space and support, or a private cloud rate.

#### Aggregate tokens per second per GPU

**Key:** tps

**Type:** number

**Default:** 1500

**Min:** 100

**Step:** 100

**Unit:** tokens/s

**Help:** Assumption: depends on model, quantisation and batching. Measure on your hardware.

#### GPUs needed to hold the model

**Key:** modelgpus

**Type:** select

**Default:** 2

**Options:**

##### 1 GPU

**Value:** 1

##### 2 GPUs

**Value:** 2

##### 4 GPUs

**Value:** 4

#### One-time setup cost

**Key:** setup

**Type:** number

**Default:** 90000

**Min:** 0

**Step:** 1000

**Unit:** once

**Help:** Integration, indexing, testing and rollout effort.

**Outputs:**

#### Active users per day

**Key:** active_users

**Expr:** users * active / 100

**Format:** integer

#### Questions per day

**Key:** q_day

**Expr:** active_users * q

**Format:** integer

#### Tokens per day

**Key:** tok_day

**Expr:** q_day * tokens

**Format:** integer

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

#### Peak concurrent requests

**Key:** peak_conc

**Expr:** max(1, ceil(q_day * 0.25 / 3600 * 20))

**Format:** integer

**Description:** Assumes 25% of daily questions in the busiest hour and 20 seconds per answer.

#### Peak load

**Key:** load_tps

**Expr:** max(peak_conc * 40, tok_day * 0.25 / 3600)

**Format:** integer

**Description:** Tokens per second: the larger of 40 tokens/s per concurrent stream and busy-hour token volume.

#### GPUs needed

**Key:** gpus

**Expr:** max(modelgpus, ceil(load_tps / tps))

**Format:** integer

**Primary:** true

**Description:** Larger of the GPUs needed to hold the model and load divided by throughput per GPU.

#### GPUs with a standby replica

**Key:** gpus_ha

**Expr:** gpus + modelgpus

**Format:** integer

**Description:** Adds one more model replica for resilience.

#### Vector index size

**Key:** index_gb

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

**Format:** decimal

**Description:** GB, assuming 2 chunks per page and about 6 KB per chunk including vectors and metadata.

#### Hours saved per month

**Key:** hours

**Expr:** q_day * 21 * mins / 60

**Format:** hours

**Description:** 21 working days per month.

#### Monthly value

**Key:** value

**Expr:** hours * rate

**Format:** currency

#### Monthly GPU cost (with standby)

**Key:** cost

**Expr:** gpus_ha * gpucost

**Format:** currency

#### Net monthly value

**Key:** net

**Expr:** value - cost

**Format:** currency

**Primary:** true

#### Simple payback

**Key:** payback

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

**Format:** months

**Description:** Setup cost divided by net monthly value. Zero means value does not cover running cost.

**Chart:**

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

**Type:** line

**X:** users

**From:** 10

**To:** 1000

**Steps:** 25

**Series:**

- value
- cost

**Disclaimer:** Planning estimate. Throughput, minutes saved and token counts vary widely; measure them in a pilot before committing budget.

## Content

**Intro:** This free calculator sizes a private AI assistant for MAPICS and Infor XA on IBM i: GPU count for the off-box model, token volume, index size, hours saved and payback. Use the chart to see cost against value as users grow.

**Sections:**

#### How to use the calculator

**Body:** Enter the XA named users and the share active on a normal day, then questions per active user, tokens per question and answer, and the documents you will index. Add minutes saved, loaded hourly cost, GPU monthly cost and a one-time setup cost. The default assumes a larger model that needs two GPUs just to hold it, because XA sites often have a few hundred users.

#### Formulas and assumptions

**Body:** Active users are named users times the active share and questions per day follow from that. Peak concurrency assumes 25 percent of questions in the busiest hour at 20 seconds each. Load is the larger of concurrency times 40 tokens per second and busy-hour tokens per second. GPUs are the larger of the model memory requirement and load divided by throughput per GPU, and the standby adds one model replica. Index size uses 2 chunks per page at about 6 KB. The IBM i is not sized here because inference runs off-box. A month is 21 working days.

#### Reading the result

**Body:** Expect the GPU count to be set by the model footprint. The larger driver of project cost is the integration effort to build a replica and a glossary of XA fields, which you capture in the setup cost. If payback is zero, value does not cover running cost under your inputs, so test a smaller model or a narrower user group.

**Cta:** If you want these numbers checked against your XA environment, Netray can help scope a pilot. The calculator is free either way.

**Faq:**

#### Why does this not size the IBM i?

**Answer:** Because inference runs on separate x86 GPU servers. The IBM i only serves data reads or journal entries, which are a small load compared with normal batch and interactive work.

#### Should I include replica hardware?

**Answer:** Include it in the setup cost or add it to the GPU monthly cost field if you prefer a monthly figure. The calculator treats GPU cost and setup as separate inputs you control.

#### How many documents should I index?

**Answer:** Start with the sets users search most, such as drawings, POs and quality records. Index size is small for modest counts, but ingestion effort scales with document variety.

#### Can I model a smaller pilot?

**Answer:** Yes. Lower named users and active share, and the chart shows how cost and value change with scale.

## Search terms

MAPICS and Infor XA AI cost, MAPICS and Infor XA AI ROI calculator, on-prem LLM GPU sizing calculator, private AI assistant payback, ERP AI sizing tool, GPU count for LLM users, vector index size estimate

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