Infor XA AI assistant sizing and value calculator
Use this calculator to estimate how many GPUs a private Infor XA assistant needs, how large the retrieval index will be and whether the time saved justifies the cost. Every assumption is an input you can change.
Your numbers
People with a Infor XA login who could use the assistant.
Share of named users who use the assistant on a given day.
Include retrieved context. Retrieval-heavy questions often run 2,000 to 6,000.
Amortised purchase plus power, space and support, or a private cloud rental rate.
Aggregate throughput under batching for your chosen model. Measure on your hardware.
Integration, data dictionary, hardware purchase if not in monthly cost, testing.
Your results
Planning estimate only. Throughput, tokens per question and minutes saved vary widely; measure them in a pilot before committing to hardware.
Monthly value vs GPU cost by named users
Sweeps Named users from 20 to 1000, holding your other inputs fixed.
Sizing review for your Infor XA assistant
Optionally share your inputs for a short review of GPU count, model choice and the value assumptions.
Optional. The tool and every export work without this. No spam.
How to use it
Start with named users and the share who use the assistant daily. Estimate questions per user per day from a pilot or from similar tools; 5 to 15 is common. Set tokens per question to include retrieved context, not just the visible answer. Enter your GPU cost as a monthly figure and set tokens per second to a value you have measured, or leave the default for a first estimate. Read the GPU estimate and payback together.
- Peak load drives GPU count, not daily totals.
- Use measured throughput when you have it.
- Run a few scenarios using the chart.
The formulas
Questions per day equal named users times active share times questions per user. Tokens per day equal questions times tokens per question. Peak requests per second equal questions per day divided by an 8-hour day in seconds, multiplied by a peak factor of 4. GPU count is peak token rate divided by tokens per second per GPU, rounded up. Index size multiplies documents, pages and an assumed chunk density. Monthly value is hours saved times loaded cost.
- Peak factor 4 and 22 working days are fixed assumptions.
- Concurrency assumes about 25 tokens per second per stream.
- Index size is an order-of-magnitude estimate.
Reading the results
If payback is zero, the monthly GPU cost is higher than the value at your inputs; increase adoption or questions, or reduce model size and GPU count. A modest model on one GPU can serve a surprising number of users if questions are short. Long-context retrieval and big models push GPU count up quickly. Keep a safety margin of at least one spare GPU or a failover plan if the assistant becomes business critical.
- Compare against the TCO calculator for a fuller picture.
- Treat minutes saved as a hypothesis to test.
- Do not count savings that are not realised in practice.
Common mistakes
People often size from average load rather than peak, forget retrieved context in token counts, or assume large time savings per question without evidence. Another mistake is ignoring the one-time work: for Infor XA, the data dictionary and permission mapping take real effort and belong in setup cost. Finally, remember that a simpler existing report may already answer the question for free.
- Size for peak, not average.
- Include context tokens.
- Count setup effort honestly.
- Validate with a pilot.
Frequently Asked Questions
What tokens per second should I use?
It depends on the model, quantisation, GPU and batching. Values from several hundred to a few thousand aggregate tokens per second are common. Measure with your own model and typical prompts, then update the input.
Why does the calculator use a peak factor of 4?
Usage clusters around shift start, after meetings and at period end. A factor of 4 over the daily average is a cautious planning figure. If your usage is flatter, the GPU count will be lower than the estimate.
Is the value figure reliable?
It is only as good as minutes saved per question and loaded cost. Treat it as a hypothesis and test it in a pilot by timing real tasks with and without the assistant.
Does it include power and rack space?
Only if you include them in the GPU monthly cost. Enter an amortised figure that covers hardware, power, cooling and support.
If you want these numbers validated on your own Infor XA data and hardware, Netray can run a short sizing pilot on-prem.
Related Tools
Infor XA Private AI Layer Blueprint Generator
Fill in your Infor XA version, database, hosting and use cases and get a shareable architecture plan for a private AI layer.
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.
On-Prem AIPrivate AI Total Cost of Ownership Calculator
Model the full 3-year cost of an on-prem AI deployment, including amortized hardware, power, staff time, and support, against comparable API spend.
More Other ERPs tools
See all- Infor VISUAL Work Order Status and Structure Reference
- Infor VISUAL Manufacturing Scheduling Checklist
- Infor XA (MAPICS) Period End Close Checklist
- Infor VISUAL Database Table Reference
- Infor XA (MAPICS) Application Module Reference
- IBM i Commands for Infor XA Administrators Reference
- Acumatica Generic Inquiry Design Template
- Acumatica Manufacturing Edition Setup Checklist
- Sage X3 Import Template Worksheet
Go Deeper
On-Prem LLM Deployment Architecture: Reference Guide
Reference architecture for on-prem LLM deployment: inference servers, GPU sizing, RAG pipelines, and security zones for regulated manufacturers.
Security Hardening for On-Prem AI Stacks
Security hardening for on-prem AI stacks: container isolation, model supply chain, prompt injection defenses, and audit logging for defense contractors.
Keep exploring
Free tools
- SyteLine AI Assistant Sizing and Value CalculatorSize the GPUs and estimate the monthly value and payback of a private, on-prem AI assistant for SyteLine.
- MAPICS and Infor XA AI Assistant Sizing and Value CalculatorEstimate GPU count, token volume, index size, hours saved and payback for a private on-prem AI assistant on MAPICS and Infor XA.
- Infor VISUAL AI Assistant Sizing and Value CalculatorSize the GPUs and estimate the monthly value and payback of a private, on-prem AI assistant for Infor VISUAL.
ERP fixes and how-tos
- How to add a private AI layer to Infor XA on-premReference architecture to add AI to Infor XA on-prem: read-only data path, retrieval, permissions, local GPUs, write-back control and rollout.
- Infor XA costing: cost types, roll-up and standard cost setupHow product costing works in Infor XA (MAPICS): standard versus current cost, how roll-up builds cost from BOM and routing, setup order, and common pitfalls.
- Infor XA Factory Track: What It Does and How to Plan Shop Floor Data CollectionWhat Factory Track does for Infor XA shop floor data collection, which transactions it captures, how to plan a setup, and alternatives when it is aging.
Integrations
- How to add an AI chatbot to Infor XAHow to add an AI chatbot to Infor XA on IBM i: grounded answers from Db2 for i, internal and customer portals, role-based access, and a fully on-prem option.
- How to add AI agents to Infor XA on IBM iHow to add AI agents to Infor XA on IBM i: a tool layer over Db2 for i and RPG programs, approval gates, on-prem models, and realistic expedite and exception flows.