# Hyperion and Essbase AI Assistant Sizing and Value Calculator

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

## Short answer

Estimate peak load, GPUs, index size, hours saved and payback for an on-prem AI assistant on Hyperion and Essbase.

## Category

oracle

## Erp

Oracle Hyperion and Essbase

## Audience

- CIO
- IT architect
- Finance leader
- ERP administrator

## Tool

**Kind:** calculator

**Currency:** $

**Inputs:**

#### Named users

**Key:** users

**Type:** number

**Default:** 140

**Min:** 1

**Step:** 10

**Unit:** users

**Help:** Users with a login to Hyperion and Essbase who could use the assistant.

#### Daily active share

**Key:** active_pct

**Type:** slider

**Default:** 30

**Min:** 1

**Max:** 100

**Step:** 1

**Unit:** %

**Help:** Share of named users who ask at least one question on a typical day. 10 to 30 percent is common after launch.

#### Questions per active user per day

**Key:** q_per_day

**Type:** number

**Default:** 4

**Min:** 0

**Step:** 1

**Unit:** questions

#### Prompt plus answer tokens per question

**Key:** tokens_q

**Type:** number

**Default:** 4000

**Min:** 100

**Step:** 100

**Unit:** tokens

**Help:** Includes retrieved context sent to the model. 2,000 to 6,000 is typical with retrieval.

#### Documents to index

**Key:** docs

**Type:** number

**Default:** 20000

**Min:** 0

**Step:** 1000

**Unit:** documents

**Help:** For Hyperion, count board packs, commentary documents and supporting detail exports. The numbers in the cube are small; documents dominate the index.

#### Average pages per document

**Key:** pages

**Type:** number

**Default:** 12

**Min:** 1

**Step:** 1

**Unit:** pages

#### Minutes saved per question

**Key:** mins

**Type:** number

**Default:** 6

**Min:** 0

**Step:** 0.5

**Unit:** minutes

**Help:** Be conservative. Count only time that would otherwise be spent searching or running a report.

#### Loaded hourly cost

**Key:** rate

**Type:** number

**Default:** 70

**Min:** 0

**Step:** 1

**Unit:** $/hr

#### GPU monthly cost (each)

**Key:** gpu_cost

**Type:** number

**Default:** 4200

**Min:** 0

**Step:** 100

**Unit:** $/month

**Help:** Hardware amortised over 36 months plus power, hosting and support for one data-centre GPU.

#### Tokens per second per GPU

**Key:** tps_gpu

**Type:** number

**Default:** 1500

**Min:** 50

**Step:** 50

**Unit:** tokens/s

**Help:** Aggregate batched throughput for your model and GPU. Measure it; 500 to 3,000 is a wide range.

**Outputs:**

#### Daily active users

**Key:** active

**Expr:** users * active_pct / 100

**Format:** integer

#### Questions per day

**Key:** q_day

**Expr:** active * q_per_day

**Format:** integer

#### Tokens per day

**Key:** tokens_day

**Expr:** q_day * tokens_q

**Format:** integer

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

#### Peak concurrent requests

**Key:** peak_conc

**Expr:** max(1, ceil(q_day * 0.2 * 20 / 3600))

**Format:** integer

**Description:** Assumes 20 percent of daily questions land in the busiest hour and an average of 20 seconds per answer.

#### Peak tokens per second

**Key:** peak_tps

**Expr:** tokens_day * 0.2 / 3600

**Format:** integer

#### GPUs required

**Key:** gpus

**Expr:** max(1, ceil(peak_tps / tps_gpu))

**Format:** integer

**Primary:** true

**Description:** Peak tokens per second divided by tokens per second per GPU. Add a second node for availability.

#### Vector index size

**Key:** index_gb

**Expr:** docs * pages * 1.2 * 6 / 1000000

**Format:** decimal

**Description:** About 1.2 chunks per page and 6 KB per chunk (embedding plus text and metadata). Result in GB.

#### Hours saved per month

**Key:** hours

**Expr:** q_day * 22 * mins / 60

**Format:** hours

**Description:** 22 working days per month.

#### Monthly value of time saved

**Key:** value

**Expr:** hours * rate

**Format:** currency

#### Monthly GPU cost

**Key:** cost

**Expr:** gpus * gpu_cost

**Format:** currency

#### Net monthly value

**Key:** net

**Expr:** value - cost

**Format:** currency

**Primary:** true

#### Simple payback on GPU hardware

**Key:** payback

**Expr:** if(gt(net, 0), gpus * gpu_cost * 36 / net, 0)

**Format:** months

**Description:** Hardware treated as 36 times the monthly figure divided by net monthly value. Zero means no payback at these inputs.

**Chart:**

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

**Type:** line

**X:** users

**From:** 50

**To:** 5000

**Steps:** 30

**Series:**

- value
- cost
- net

**Disclaimer:** Planning model. Peak hour share (20 percent) and 20 seconds per answer are fixed assumptions. Hours saved is the most uncertain input: measure it in a pilot before relying on the value figures. Excludes implementation, licences, staff and storage costs.

## Content

**Intro:** This calculator sizes an on-prem AI assistant for Hyperion and Essbase and turns the numbers into a monthly value and a simple payback. Enter users, usage, document volume and GPU cost and read off peak load, GPU count and index size.

**Sections:**

#### How to use it

**Body:** Start with named users and the share who will ask questions on a typical day. Then estimate questions per active user and tokens per question (a retrieval-based answer often uses 2,000 to 6,000 tokens including context). Enter the number of documents and average pages to be indexed, the minutes saved per question and a loaded hourly cost. Adjust the tokens per second per GPU to match a benchmark of your own model on your own hardware.

**Bullets:**

- Use pilot data for questions per user as soon as you have it
- Keep minutes saved conservative, 2 to 5 for most lookups
- For Hyperion, count board packs, commentary documents and supporting detail exports. The numbers in the cube are small; documents dominate the index.

#### The formulas

**Body:** Questions per day = users x daily active share x questions per user. Tokens per day = questions x tokens per question. Peak tokens per second = tokens per day x 0.2 / 3,600, assuming 20 percent of traffic in the busiest hour. GPUs = peak tokens per second / tokens per second per GPU, rounded up with a minimum of one. Index size = documents x pages x 1.2 chunks x 6 KB. Hours saved per month = questions per day x 22 x minutes saved / 60. Value = hours x loaded rate. Payback = 36 x monthly GPU cost / net monthly value.

#### Reading the result

**Body:** A single GPU often covers a few hundred users because traffic is bursty and answers are short, so the GPU count is usually driven by availability and model size more than throughput. The sweep chart shows value rising linearly with users while GPU cost rises in steps, which is why value per GPU improves as you add users. If net value is negative, check whether minutes saved is realistic or whether the use cases are too narrow.

#### Common mistakes

**Body:** People often assume every named user will ask questions daily; real adoption is lower. Others count the cheap tokens of a short question but forget the retrieved context that is sent with it. Do not forget implementation effort, staff time, licensing and storage, which this model excludes, and measure savings with a pilot rather than trusting a guess.

**Cta:** If you want these numbers validated against your Hyperion and Essbase environment, Netray can run a short sizing review and a pilot on your own hardware.

**Faq:**

#### Why only 20 percent of traffic in the busiest hour?

**Answer:** It is a common planning rule for office workloads, where use clusters in a few hours. If your usage is more concentrated, such as shift starts or month-end close, increase the factor by scaling questions per day or the tokens per second margin.

#### How do I get tokens per second per GPU?

**Answer:** Run your chosen model on your GPU under a serving stack such as vLLM with realistic concurrency, and record aggregate output plus input tokens per second. Published figures vary widely with model size, quantisation and batch size, so measure.

#### Does the model include licences and staff?

**Answer:** No. It covers GPU cost and time saved only. Add implementation, database licensing, storage, security review and support staff to get a full business case. Use the TCO and ROI calculators for a broader view.

#### What does a zero payback mean?

**Answer:** It means net monthly value is zero or negative with your inputs, so the hardware never pays back in this model. Raise realistic adoption or time saved, narrow the scope, or reconsider whether the use case justifies a dedicated GPU.

## Search terms

hyperion and essbase ai roi calculator, hyperion and essbase llm gpu sizing, hyperion ai business case, on-prem ai assistant sizing calculator, how many gpus for erp chatbot, private llm payback calculator, hyperion assistant cost

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