Stop the wrong work before AI automates it. An open-source agent pipeline that reads Celonis-shaped event logs, scores against the AI Business Value Framework, and emits MuleSoft Anypoint calls when the ruling is Stop.
Every senior leader I worked with this year arrived with the same shape on their AI portfolio. They had scored initiatives, ambitious roadmaps, and a quietly growing change-management line item absorbing the cost of workflows that should not have existed.
The workaround was always the same. More change-management spend, more governance committees, more pilots. Nobody asking the upstream question, should the workflow exist at all.
Prosci's data says drop change spend below 10 percent of programme budget and failure runs six times higher. That is the engine behind every CIO budgeting up the change line against a workflow that should have been deleted six months earlier.
Every bundled consultancy platform shipping this year defaults to automate. The Coroner defaults to Stop. Score the workflow against the AI BVF, eliminate the work that drained ROI under the old shape, only then sequence the agentic redesign across what survives.
My positioning is the one reducing friction between strategy, execution, and decision-making. The friction is the workflow itself. Kill the workflow before you hire agents against it.
The AI BVF is the decision layer, the scoring engine and public protocol that grade any AI initiative against four pillars and return Stop, Fix, Accelerate, or Keep. The Workflow Coroner is what happens when that decision layer wraps an actual process estate, the same scoring engine applied across an end-to-end pipeline.
The Workflow Coroner is the public reference build that shows how the AI Business Value Framework (AI BVF) lands inside an enterprise process estate. Pulse ingests workflow data shaped like Celonis Process Intelligence exports, Verdict scores against the AI BVF, Architect drafts the target state, Wire emits MuleSoft Anypoint calls, and Sequencer ranks the portfolio into a 90-day plan. The Coroner orchestrates the five and writes the artefacts to disk, the buyer reads a single envelope.
The pipeline is open source, MIT-licensed, and reproduces deterministically. Anyone can run it against the included fixtures or their own Celonis-shaped JSON, the verdict comes back with EUR drag and saving denominated, and the source code carries the whole rule set.
Two real fixtures ship with the Coroner. The same five-agent pipeline produces opposite rulings depending on the workflow signals, the AI BVF decides which way each one goes.
Score an AI initiative on the AI BVF first, the four pillars and your business inputs come back here in the URL. Answer one workflow-shape question on the next section, the Coroner runs the same AI BVF rules with workflow drag layered on top, and returns Stop, Fix, Accelerate, or Keep with EUR drag and saving denominated. Your workflow inputs never leave your browser, the verdict runs client-side.
Once you complete the AI BVF and copy the share link, paste it into the address bar on this page and the Coroner picks up the four pillar scores plus your business inputs, asks one shape question on the workflow itself, and returns the verdict. Until then the form below stays hidden.
The workmate runs a typical workflow signature for your industry and function, drawn from a pre-generated profile table the workmate was seeded with. Confirm or adjust the four sliders below, then run the verdict. Your workflow inputs never leave your browser.
This is the senior-leader on-ramp, fast, plain English, no Python. The Coroner's actual Celonis-shaped ingestion and MuleSoft Anypoint actuation lives in the technical on-ramp below, reads your real event log, runs on the open-source pipeline.
The honest current state, the integration contracts are real but the wires to a customer's tenant are not yet plugged in. Pulse reads JSON in the exact shape a Celonis Process Intelligence export produces, the Coroner runs the pipeline end to end, and Wire emits Anypoint-shaped DELETE / PATCH / POST calls that mirror what a real MuleSoft connector would accept. The contracts match production, the calls do not leave the host.
Two reasons for shipping the mock first. Buyer trust, the rule engine, the verdict logic, and the saving calculation are all auditable on GitHub before any tenant is connected. Build discipline, contracts harden against fixtures before they touch live data, which is the cheapest place for the contracts to fail.
The live API adapter sits in weeks 1 to 2 of the platform build path below. The Coroner's design is built for the swap, the live adapter slots into the same Pulse interface that the fixture uses today, and the same Wire interface that the mock uses today.
The platform around the Coroner is private, in flight, and built for a CIO self-serve buyer. The roadmap is in the open here so any reader can see what is shipped and what is coming.
This section is for developers and technical operators. The senior-leader on-ramp is the workmate above, which picks up your AI BVF inputs from the scorer at bvf-app.vercel.app/protocol/scorer and runs the verdict in browser against a typical workflow profile.
The technical path below is the deepest way to interact with the Coroner. Open source on GitHub, Python in 30 seconds, two fixtures included, reads your actual Celonis-shaped event log rather than a typical profile.
npm install -g aibvf-mcp. Six stdio tools any Claude agent can call.Thirty minutes on the calendar, you bring the workflow data shape, I bring the AI BVF and an honest verdict. No deck, no pitch, the engine does the work.
Book a Call