AURICS STUDIO
Customer Proof AI Layer
Rebuilding Customer Advocacy & Marketing to be AI-Native.
An open-source layer that turns rough customer material into governed proof that Product Marketing, Customer Marketing, and the wider GTM team can reuse - without anyone overstating the claim.
The problem
Every GTM team rewrites the same customer story.
Sales writes an outreach line. Marketing writes a case study. The website writes a proof block. AR/PR writes a briefing note. Each works from a different reading of the same call, each lands on a slightly different claim, and none of them is governed.
Typically, teams don't lack proof. What's missing is anything that controls how far a claim may go. So proof either sits unused because nobody is sure it's cleared - or it gets used, and overreaches.
What's needed is a clear, accountable decision about what is true, current, commercially useful and safe to say. That's the part of marketing that can't be delegated to a model.
What this is
A governance layer that sits between raw customer material and everything produced from it.
Transcripts, QBR notes, call excerpts and existing case studies go in. What comes out is a Customer Proof Record - a single governed unit that Sales, CS, Marketing, Website, AR/PR and Events all draw from, instead of each team writing its own version of the same story.
AI structures and adapts customer proof. Humans approve what is true, current, commercially useful, and safe to use.
Whoever owns Customer Advocacy owns the record - usually Customer Marketing or Product Marketing. Everyone else consumes it.
The Customer Proof Record
Five fields. Everything downstream is generated from the approved proof only - never from ungoverned raw notes.
| Field | What it holds |
|---|---|
| Customer context | Who the customer is, and the scope of what they actually deployed |
| Proof priority | What this proof is best used to demonstrate |
| Buyer problem | The pre-purchase problem, in the buyer's terms |
| Approved proof | The conservative, defensible outcome statement |
| Use controls | Evidence strength, approval status, permission type, customer load |
That last field is the one that does the work. Customer load - how often you've already gone back to this customer is tracked as a first-class property. Overuse wears down the customer relationship, and it wears down the external credibility of the proof itself.
The governance check
Every generated output is checked against the record's use controls before it is produced:
- Source attached - the output traces to material
- Permission checked - public, private or internal use, honored
- Approval checked - internal review vs. customer-approved
- Claim scope preserved - nothing quantified beyond what was approved
Anything held out of an output is listed with its reason. The exclusion is visible, not silent. You can see what the system refused to say, and why - which is the difference between a governance layer and a filter.
What the human actually decides
This is the moment the whole system is built around.
AI EXTRACTION
HUMAN-APPROVED VERSION - WITH CONTEXT
During a six-week pilot, AI-referred sessions converted 12% above the site average. Referral volume remained limited, so this is classified as an early commercial signal, not validated enterprise ROI.
Same underlying fact. Completely different claim. The second one survives contact with a prospect who asks a follow-up question.
AI can surface evidence. Humans decide the claim scope.
The defaults enforce this in code rather than leaving it to the model's judgment: any quantified business outcome the source material doesn't explicitly approve is stripped from approved proof and moved to the excluded list, and every new record starts at private use and internal review regardless of what the notes claim.
What one approved record produces
- Proof summary for Sales outreach
- Business expansion note
- Marketing case study brief
- Website proof block
- External approved proof note
Five outputs, one governed source, every one carrying the same claim scope.
Try it
The demo runs a full workflow end to end.
| Step | What you do |
|---|---|
| Start | Pick a sample record, or paste your own customer notes |
| Create record | Watch raw material get structured into the five fields |
| Review record | See the record, and the human correction moment |
| Use this proof | Generate role-specific outputs, each governance-checked |
| Proof library | Search and filter by priority, problem, vertical, approval, permission, load |
| Metrics | Proof system health - ready-to-use proof, source coverage, reuse, customer load |
| Expansion | Where this goes next |
The demo runs entirely in your browser and makes no network calls. Nothing you paste is stored or sent anywhere. The full version adds live model extraction on your own notes.
Scope
- The sample records are synthetic. Five category-referenced examples - enterprise payments, agentic commerce, AI infrastructure, CRM, enterprise AI. Illustrative, not real customer claims, and no one of them is the primary use case.
- This is a prototype that demonstrates the workflow and sample outputs. Future versions could add CRM integration, approval routing, and persistence.
- It doesn't replace the approval conversation. It makes the conversation faster and gives it a consistent shape. Someone still has to have it.
- There are two versions. The browser demo runs the full workflow over sample records with no network calls at all; on your own notes it applies the governance defaults - which live in code, rather than calling the model. The full version adds live model extraction.
Why I built it
Buyers increasingly discover products through AI assistants, and AI systems reuse whatever proof they can find. If that proof is ungoverned, what gets repeated back is whatever the strongest phrasing happened to be - usually the one Legal didn't approve.
Marketing already had this problem before AI; AI just made the blast radius larger and the loop faster. So the fix isn't to generate proof faster. It's to make the governed version the one that's easiest to reach for.
That's what AI-Native GTM & Marketing means here: not bolting a model onto the existing workflow, but redesigning the workflow so the governed path is the path of least resistance.
Where it goes next
Integrations with CRM, Gong and transcript sources · approval routing · semantic search across records · proof gap analysis · field usage tracking · revenue influence attribution.
Further Reading
FAQ
- Who is this for?
- Customer Marketing and Product Marketing teams who own customer proof and are tired of every other team rewriting it. It's most useful if you have more customer material than you have approved claims.
- How is this different from a case study library?
- A case study library stores finished assets. This stores the governed unit - the approved claim and its use controls - and then generates the assets from it. When the claim changes, everything downstream changes with it, instead of six stale documents staying wrong.
- Does the AI decide what's approved?
- No, and that's the point. AI extracts and adapts; it never sets claim scope. Conservative defaults are enforced in code - anything quantified that the source material doesn't explicitly approve is excluded automatically and surfaced for a human to rule on.
- What happens to notes I paste into the demo?
- Nothing is stored. Text is sent for extraction, structured into a record, and held in browser state for the session only.
- Can I run this on my own data?
- Yes - it's MIT licensed and the source is public. It's a prototype, so expect to adapt it rather than deploy it as-is.
- Why open source?
- Same reason as the rest of Aurics Studio: GTM tooling claims are easy to make and hard to check. Shipping the source means the method is inspectable, not just the marketing.