Estimator

Size your migration

Describe your company and pick a path. We size the program the way a systems integrator would, then show what a small senior team running AI agents takes — in time and money — and what you would pay us.

Your company

Customization
Path
If you have a quote for your current path, enter it. It replaces our benchmark and becomes the base for the success fee.

Execution model

Agent assumptions
How much of the work agents absorb, how productive the senior team is, and how much inference a unit of work takes.
Our working figure is 60–80 billion: many agent lanes running in parallel per engineer.

Delivery cost

Same scope, four ways to execute it

Months to complete

Including calendar time AI can't compress

Show as a table

What you would pay

Assessment and delivery are billed at our cost. We make money only from the success fee, paid on acceptance in milestone tranches.

Inside our estimate

All paths at a glance

Inference cost per AI-enabled engineer, by provider

List prices checked 3 November 2026, before negotiated discounts. Long-context price tiers are used where providers have them, because agent prompts are large. Where a provider charges no cache-write fee, uncached tokens are billed at the input price.

All assumptions

How this is calculated

Everything on this page comes from one model, which also produces the numbers on the rest of the site. Defaults marked sourced come from published benchmarks; those marked assumption are ours, and you can change all of them above.

Your current path

We start from an indicative all-in program budget for an SI-led conversion to S/4HANA at your size, adjusted for scope and customization, then take the share that is delivery services (SI fees and contractors). That is what we compare against. Licenses, infrastructure and your own staff's time are excluded from the comparison; they are similar whoever executes.

SI services = budget(size) × scope × customization × path × 60% SI hours = SI services ÷ blended rate SI months = months(size) × √(scope × customization) × path

Our team, with agents

We split the SI's hours into nine workstreams. Agents produce part of each workstream; senior people do the rest, steer the agents and review what they produce. The team is sized to finish near the calendar floor. Labor runs for the whole program; inference runs at the per-engineer rate you set.

agent work = Σ SI hours × share × absorb human hours = Σ SI hours × share × (1 − absorb) ÷ productivity + agent work × review ratio months = max(human hours ÷ team capacity, agent work ÷ agent capacity, calendar floor) cost = (team × months × 150 h × rate + engineers × months × inference/month + tooling) × (1 + contingency)

What you pay

assessment = Phase 0 at our cost fixed fee = delivery at our cost (above) success fee = 15% × current path's delivery cost you pay = assessment + fixed fee + success fee

The 5-year view

For the paths that change what you run, we add one-time software and infrastructure and five years of run cost. The annual SAP run cost is an assumption (a share of the program budget); the freedom paths replace some or all of it with the cheaper cost of running what you own.

What the model leaves out

  • Your own staff's time (vendors estimate it adds 20–35% to program cost).
  • Software and infrastructure, except in the 5-year view.
  • Overruns on the traditional path, unless you switch them on under All assumptions.
  • Faster or slower agent progress than the preset you choose. The conservative preset shows what happens if we are wrong.

Sources

  1. VendorBenchmark, “SAP S/4HANA Migration Cost Benchmarks 2026” (Apr 2026) — cost bands, durations, SI rates, SI share of spend. Indicative, vendor-published. vendorbenchmark.com
  2. McKinsey & University of Oxford (2012) — large IT projects average 45% over budget. mckinsey.com
  3. GitHub Copilot coding-agent telemetry, arXiv 2608.00101 (2026) — ~94% of agent prompt tokens are cache reads; output under 1%. Basis for the token mix.
  4. Provider pricing pages, checked 3 Nov 2026: Anthropic, OpenAI, Google, xAI, DeepSeek, Together AI, DeepInfra.
  5. SAP Joule for Developers claims (up to 20% less coding effort, 25% faster testing) — basis for the “SI with AI copilots” scenario. Vendor claim.
  6. Kagool, “SAP ECC to S/4HANA Migration Cost: 2026 Budget Guide” — brownfield conversion ≈ 3–4× annual running cost. Vendor estimate, used for the run-cost assumption.