IBM i (AS/400) Custom RPG Applications AI Assistant Sizing and Value Calculator
This calculator sizes a private AI assistant for IBM i (AS/400) Custom RPG Applications and shows whether the value of time saved covers the GPU cost. Enter users, usage and your own throughput number; the result is GPU count, token volume, index size, hours saved and payback.
Your numbers
People with access to the assistant.
Share of named users who ask something on a given day.
Includes retrieved passages in the prompt. 1,000 to 4,000 is common with retrieval.
Be conservative; measure in the pilot.
Amortized hardware, power and hosting per GPU.
Aggregate throughput with batching for your model. Measure with a benchmark.
Your results
Planning estimate. Time saved is the weakest input: haircut it by 30 to 50 percent until the pilot measures it. Excludes staff, licences, networking and storage beyond the index.
Monthly value vs GPU cost by named users
Sweeps Named users from 50 to 1500, holding your other inputs fixed.
Validate my sizing
Optional review of your inputs and a plan for measuring throughput and time saved in a pilot.
Optional. The tool and every export work without this. No spam.
How to use it
Start with named users and the share who ask something on a typical day. Set questions per active user and the tokens per question, remembering that retrieved passages count. Enter the documents to index with an average page count, then your estimate of minutes saved per question and the loaded hourly cost. Use a measured tokens-per-second figure for your model and GPU if you have one. Small user base, high question rate.
- Use realistic daily active share, often 20 to 40 percent
- Include retrieved context in token counts
- Measure GPU throughput on your own model
- Haircut time saved until the pilot proves it
The formulas
Questions per day equal named users times active share times questions each. Tokens per day equal questions times tokens per question. Peak tokens per second assume 15 percent of the day's questions land in the busiest hour, so peak rate is daily tokens times 0.15 divided by 3600. GPUs equal peak rate divided by tokens per second per GPU, rounded up with a minimum of one. Index size equals documents times pages times 2 chunks times about 6 KB. Hours saved use 22 working days.
- Peak share of 15 percent is an assumption
- Chunks per page and bytes per chunk are assumptions
- Throughput depends on model, quantization and batching
- Concurrency assumes about 30 tokens per second per stream
Reading the result
If net monthly value is negative, the deployment is too small for the GPU cost or the time saved per question is too low. If payback is under 12 to 18 months with a conservative saved-time figure, a pilot is justified. Watch the chart: value grows with users while GPU cost steps up in whole GPUs, so the gap widens at scale. The real cost is the semantic layer over short field names, not GPU hardware.
Common mistakes
The usual errors are counting every named user as a daily user, forgetting that retrieved text multiplies tokens per question, trusting a vendor throughput figure for a different model, and counting time saved on questions people would not have asked. Also remember that build effort, integration and support are real costs; the one-time build input is there so payback is not flattering.
- All named users counted as active
- No retrieval tokens in the prompt
- Borrowed throughput numbers
- Ignoring build and support cost
Frequently Asked Questions
What tokens per second should I use?
Benchmark your chosen model on your GPU with realistic prompt lengths and batching, then enter the aggregate throughput. Published figures vary widely by model size, precision and engine, so a measured number beats a default.
Why does the calculator assume a 15 percent peak hour?
Usage clusters at the start of shifts and after meetings. Fifteen percent of daily questions in the busiest hour is a common planning assumption; change the formula in your own copy if your logs show more.
Is the minutes-saved figure realistic?
Treat it as a hypothesis. Search and lookup questions often save a few minutes; complex tasks vary. Measure with a sample of tasks in the pilot and haircut the estimate before presenting payback.
Does it include staff and licence costs?
No. It covers GPU cost and a one-time build input. Add operations, storage, networking and any software licences separately for a full total cost of ownership.
We can run a measured pilot on your IBM i RPG data and replace these estimates with your own numbers. Ask for a sizing review if you want a second set of eyes.
Related Tools
IBM i (AS/400) Custom RPG Applications Private AI Layer Blueprint Generator
Fill in your IBM i RPG environment and export an architecture plan for a private, on-prem AI assistant: components, security controls, rollout and open questions.
On-Prem AIOn-Prem AI ROI Calculator
Turn hours saved per employee into annual net benefit, payback months, and 3-year ROI for an on-prem AI investment.
On-Prem AIGPU Sizing Calculator for LLM Inference
Work out how many GPUs you need to serve a given open-weight model to your user base, based on memory footprint and token throughput.
On-Prem AIOn-Prem LLM Total Cost of Ownership Calculator
Model the full multi-year cost of running LLMs on your own hardware, including GPU capex, power, cooling, support contracts, and operations staffing.
Developer ToolsIBM i (DB2 for i) SQL Data Extraction Script Generator
Generate DB2 for i SQL to inspect a file, count rows, and extract records with Julian date filters for JD Edwards World or other IBM i data.
More On-Prem AI tools
See all- On-Prem vs Cloud AI Cost Comparison Calculator
- Self-Hosted LLM Hardware Estimator
- Air-Gapped AI Readiness Assessment
- LLM API vs Self-Hosted Cost Calculator
- RAG Infrastructure Sizing Calculator
- Open-Weight Model Selector
- On-Prem AI Deployment Checklist
- LLM Token Cost Calculator
- AI Data Center Power and Cooling Calculator
Keep exploring
Free tools
- IBM i (AS/400) Custom RPG Applications Private AI Layer Blueprint GeneratorFill in your IBM i RPG environment and export an architecture plan for a private, on-prem AI assistant: components, security controls, rollout and open questions.
- Custom MES on SQL Server AI Assistant Sizing and Value CalculatorEstimate GPUs, index size, hours saved and payback for a private AI assistant over Custom MES on SQL Server.
- SharePoint Server On-Prem AI Assistant Sizing and Value CalculatorEstimate GPU count, token volume, index size and payback for a private AI assistant on SharePoint Server from users, usage and time saved.
ERP fixes and how-tos
- How to Add AI to an IBM i (AS/400) RPG Application Without Moving ItHow to add a private AI assistant to custom RPG applications on IBM i: Db2 for i access, field names, journals, RCAC security, local LLM, sizing and rollout.
- Infor XA REST API: expose XA through IBM i Integrated Web Services or IONHow to expose Infor XA (MAPICS) data and functions as REST on IBM i using Integrated Web Services or ION, with an example call and the safety rules for writes.