Proving it on a real, regulated corpus — energy-renovation advisory.
Our first reference is a governed public agent for an energy-renovation advisory practice (Energieberatung / iSFP) under BEG, BAFA and KfW funding rules — a real, regulated, document-heavy domain where customers ask the same hard questions repeatedly. We’re proving deflection, the correction loop, and zero internal-corpus leakage before we generalise.
An honest note up front. This is our Phase 0 dogfood — the first live reference is being built with a design-partner practice right now. There are no fabricated testimonials or invented figures on this page. The metrics panel below is the real frame; the numbers land as the channel runs.
Why energy-renovation advisory is the corpus to prove on
We win one opinionated channel completely before opening the platform. Energy-renovation advisory is real, regulated and repetitive — the ideal place for a governed public agent backed by a private superset.
A regulated, document-heavy corpus
BEG, BAFA and KfW funding rules, eligibility tests and renovation roadmaps (iSFP) change often and reward precision. Getting an answer wrong has real consequences — exactly where a governed agent earns its keep.
The same hard questions, on repeat
Homeowners and SMEs ask the same funding and eligibility questions over and over. That repetition is what a public agent can deflect accurately — freeing the advisor for the judgement only they can give.
Trust already runs the relationship
The advisor is the trusted face for the homeowner and the regional SME. The agent extends that trust online — it never replaces the human, and it hands back the moment it is unsure.
The same governance need shows up in any document-heavy, compliance-sensitive SME service. We prove it here first — then it becomes the playbook for the consultants we arm next.
Five proof points, measured — not claimed
These are the numbers the Phase 0 reference exists to capture. The frame is fixed; the values are filled in as the live channel runs. We’d rather show you an honest placeholder than a flattering invention.
Deflection rate
— in measurementShare of customer questions answered accurately by the public agent without an advisor in the loop.
Escalation → correction-loop closing
— in measurementHow reliably an uncertain question reaches the advisor and the human answer is folded back as new knowledge.
Internal-corpus leakage
Zero · targetPrivate superset content that ever reaches a public customer. The target is, and stays, zero — by design, not by hope.
Advisor hours saved
— in measurementHours the advisory practice no longer spends retyping the same funding and eligibility answers each week.
Customer satisfaction (CSAT)
— in measurementHow homeowners and SMEs rate the answers they get — straight reply or fast hand-off to a person who knows them.
Design-partner in progress. Want the numbers the moment they land? Ask to follow the reference, or become the practice we measure.
Two of these are the spine of the product: every answer is a governed projection of a private superset, and every uncertain one runs through the correction loop back into the knowledge base.
Private superset in, public agent out, the loop around it
The mechanism is the same one the whole platform rests on — we’re just exercising it on a live, regulated corpus instead of a sample one. Governed projection decides what goes out; the correction loop makes it better with every escalation.
A private superset, curated
The advisory practice’s full knowledge — funding rules, internal notes, case history, scattered PDFs — becomes one structured, default-private knowledge base. The superset, never exposed wholesale.
A governed projection out
The public agent is a governed projection of that superset: sensitivity labels and policy decide exactly what a customer can ever see. Approved, cited answers go out; the private corpus stays in.
Uncertainty escalates to the advisor
When the agent isn’t sure, it doesn’t guess — it hands off to the human advisor who actually knows the funding landscape and the customer. The escape hatch is the point, not an edge case.
The answer becomes new knowledge
The advisor’s resolution is captured back into the knowledge base, so the next person asking the same question gets it instantly. The system improves exactly where it was weak.
That is what makes the proof honest: governed projection is why leakage can be zero, and the correction loop is why deflection compounds rather than plateaus.
This is our Phase 0 dogfood — the first live reference is being built with a design-partner practice.
We’re recruiting the first handful of design partners, not running a mass program. If you’re an energy-efficiency advisor, a Steuerberater, or any advisory practice running on a document-heavy corpus, you can be the reference — or just follow along and see the numbers as they land.
Want to be a reference — or see the numbers as they land?
Become a design partner and we’ll build the live reference with your practice and your corpus. Prefer to watch first? Talk to us and we’ll keep you posted as the proof points come in.