Pick one process that costs you more than it should. Six weeks later you have a governed AI agent working it inside SAP, Salesforce, ServiceNow or whichever system runs it today, with guardrails your risk team can sign off on, for one fee agreed before we start. And you'll know exactly what it changed, because we measure the process before and after.
Agent demos are easy to produce and hard to trust. The questions that matter in production are about governance: who approved the action, what the agent is allowed to touch, and what happens when its confidence drops. Your CFO has a different question: did it actually improve anything? The pilot exists to answer both, using your own data.
Running on your process, in your own system, with guardrails configured and your team trained to operate it.
Week-one and week-six measurements side by side, showing what improved and what that is worth over a year. See the sample format →
Permissions model, thresholds, review-queue design, BPMN process models, the LeanIX map of every system the agent touches, and the audit-trail spec. Everything risk and compliance will ask for, already written.
Based on what the numbers show. If they say stop, we will tell you to stop, and you will have learned that at week six rather than after a large rollout.
Every agent targets at least one of these four measurements. The pilot report shows the movement in each, taken from your own event data.
The share of work done in the system rather than around it. Agents turn the emails, spreadsheets and workarounds into governed system actions.
What it costs to process one case. Agents take the routine touches out, so your people only handle the cases that need them.
How much of the elapsed time is actual work. Agents act in minutes, so cases stop waiting in queues between steps.
Cases completed without rework. Agents apply the rules the same way every time, so fewer cases bounce back.
Illustrative movements. The four measurements come from our Portfolio Process Mining methodology, and your pilot report shows the real ones.
The pilot runs on a simple rhythm: capture how the process performs today, put the agent to work, and finish with results you can verify. The numbers come straight from the event logs your systems already produce, so nobody has to estimate the benefit.
We measure how the process actually runs today, straight from the source system.
The agent goes to work inside the system, with guardrails agreed before it handles a single live case.
We run the same measurement again and review it with you: what improved, by how much, and what that is worth over a year.
Generic AI consultancies assert ROI. We measure it in your event logs. The measurement stays in place after the pilot, so if you scale, every new agent is held to the same standard.
An agent that can't explain itself has no place in your core systems, so every pilot includes the governance an auditor would expect to see.
The agent gets the minimum access the process needs, defined action by action, and you can revoke it in one place.
Business rules and confidence scores decide what the agent completes on its own and what routes to a person.
Edge cases land in a queue with full context, so your team stays in charge of judgement calls while the routine work keeps moving.
The agent's role, hand-offs and escalation paths are modelled in BPMN alongside your human process, so one model covers both people and agents.
Every action the agent takes is logged with its inputs, confidence, and outcome. When someone asks "why did it do that?", there's an answer.
We keep tracking exception rates, interventions and throughput after go-live, using the same process intelligence that measured the pilot.
An agent working blind in a complex landscape is a risk. We keep a live model of your architecture in SAP LeanIX, covering applications, interfaces, data ownership and any agents already running, and make it available over MCP so agents and developers can check what exists before they act.
Before touching a process, the agent queries the model to learn which systems, APIs and data are involved, so nothing depends on hard-coded assumptions that go stale.
Dependencies, ownership and blast radius live in the model. The agent knows what sits downstream of a record before changing it, and your architecture team can see the same picture.
Each new agent inherits the map the last one used and adds itself to it, so the model becomes more useful with every deployment.
The pilot is not tied to one vendor. If the system that runs your process produces event data, we can measure it and build against it.
Measurement is SAP Signavio or Celonis, depending on what fits your stack, including landscapes where one of them is already deployed. The architecture map is SAP LeanIX, queried by agents over MCP. On SAP, agents build on Joule and BTP; elsewhere we bring the agent runtime and connect it through your platform's native APIs.
Pilots start from $40,000, with the exact fee set at the scoping call once we have seen the process and the systems involved. It covers all six weeks, with no day rates and nothing billed on top. If you go on to roll out or scale to more processes, the pilot fee is credited toward that work.
You don't need to. Salesforce, ServiceNow, Oracle, Workday, Dynamics: if the system that runs your process produces event data, the pilot works there. We choose the measurement tool, Signavio or Celonis, to fit your systems.
Then you have learned that for a fixed fee in six weeks instead of after a large rollout, and the baseline data usually shows where the real value is hiding.
No. The baseline is built from a read-only event-log extract. The agent earns its access progressively: scoped permissions, agreed with your security team, before it touches anything live.
A 30-minute scoping call is enough to choose it, confirm the systems involved and agree the price.
Book a pilot scoping call