SAP BW/4HANA AI assistant sizing and value calculator
This calculator estimates how many GPUs, how much index storage and what monthly value a private AI assistant on SAP BW/4HANA is likely to need. Enter your user counts, question volumes and costs; it returns peak concurrency, GPU count, hours saved and payback, with a chart sweeping named users.
Your numbers
Users with access to the assistant.
Share of named users who ask at least one question on a typical day.
Includes retrieved context. Retrieval-heavy answers often run 2,000 to 6,000 tokens.
End to end, including retrieval and generation.
Be conservative: time to find the answer today minus time with the assistant.
Batched serving throughput of your model on your GPU. Measure with your own model; this default is a planning figure.
Amortised hardware, power and support, or your private cloud rate.
Integration, security review, indexing and pilot work.
Your results
Planning estimate only. Measure real throughput with your model and hardware, and validate time saved with a pilot before committing budget.
Monthly value vs GPU cost by named users
Sweeps Named users from 10 to 600, holding your other inputs fixed.
Check my sizing
Optional: send your inputs and we will sanity-check the GPU count and value assumptions. The calculator works fully without it.
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 on a typical day, then estimate questions per active user. For documents to index (report definitions, kpi docs), count what you really intend to index, not everything on the file server. Set tokens per answer from a prototype if you have one; retrieval context makes answers longer than chat. The tool is aimed at SAP BW/4HANA cases like BW teams where analysts spend time finding the right query and explaining numbers. Change one input at a time and watch the chart.
- Use measured tokens per second for your own model
- Keep minutes saved conservative
- Remember the spare GPU for failover
- Revisit after the pilot with real usage data
The formulas
Daily active users equal named users times the active share. Questions per day equal active users times questions per user. Tokens per day equal questions times average tokens. Peak concurrent requests equal questions per day times 0.2 (the busiest-hour share) divided by 3,600, times seconds per answer, rounded up. GPUs equal peak-hour tokens per second divided by aggregate tokens per second per GPU, rounded up, with a minimum of one; the calculator then adds a spare. Hours saved per month equal questions per day times minutes saved divided by 60, times 21 working days.
Reading the result
If net monthly value is negative, the user base or question volume is too low to justify dedicated GPUs, and a smaller model, shared hardware or a narrower use case is the better answer. If payback is under a year with conservative minutes saved, the case is reasonable. Index size is usually small compared with model weights; the model and its GPU memory dominate hardware planning. Treat value as an upper bound until a pilot shows people actually save the time.
Common mistakes
Using the licensed user count instead of real daily users inflates GPUs. Counting every question as saving the same time overstates value, because many questions replace a thirty-second lookup. Using vendor peak throughput numbers measured on short prompts understates GPU need for long retrieval prompts. And ignoring the one-time work of permissions, indexing and security review makes payback look better than it is.
Frequently Asked Questions
How do I find tokens per second per GPU?
Benchmark your chosen model on your GPU with a serving stack such as vLLM, using realistic prompt lengths and many concurrent requests. Use aggregate throughput, not single-stream speed. The default in this tool is a planning figure only and will vary widely by model size and quantisation.
Why add a spare GPU?
A single GPU is a single point of failure, and you also need capacity during maintenance and model updates. The spare is a simple allowance. Some sites accept degraded service on failure and skip it; set the cost input accordingly if you do.
Is the 20% busiest-hour share right?
It is a reasonable planning default for a single-site working day. Multi-shift or multi-timezone sites spread load more evenly and need fewer GPUs; sites with a morning rush may need more. Adjust by scaling questions per day or adding headroom.
Does this include the cost of people?
Only through the one-time setup input, which you should fill with integration, security review and pilot effort. Ongoing support and model upgrades are not modelled, so add a margin to the monthly GPU cost if you expect meaningful run effort.
If you want these numbers checked against a real pilot on SAP BW/4HANA, Netray can run a scoped proof of value on your own hardware.
Related Tools
SAP BW and BW/4HANA Private AI Layer Blueprint Generator
Fill in your version, database, hosting and use cases to get an architecture plan for adding a private, on-prem AI assistant to SAP BW/4HANA, with components, security controls and a three-phase rollout.
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 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 AIOn-Prem AI Deployment Checklist
A 30-point pre-deployment checklist covering use cases, hardware, security, model operations, and rollout for self-hosted enterprise LLMs.
More SAP tools
See allKeep exploring
Free tools
- SAP BW and BW/4HANA Private AI Layer Blueprint GeneratorFill in your version, database, hosting and use cases to get an architecture plan for adding a private, on-prem AI assistant to SAP BW/4HANA, with components, security controls and a three-phase rollout.
- SAP S/4HANA On-Prem AI Assistant Sizing and Value CalculatorEstimate peak concurrent requests, GPUs, tokens per day, index size, hours saved and payback for a private AI assistant on SAP S/4HANA.
- SAP Business One AI Assistant Sizing and Value CalculatorEstimate peak concurrent requests, GPUs, tokens per day, index size, hours saved and payback for a private AI assistant on SAP Business One.