The AI layer
Meet Vera. She reads the documents, cites her sources, and never invents a number.
Vera is Acquiror AI's M&A analyst. She classifies what you upload, extracts the financials with a confidence score on every figure, answers questions with citations, and drafts memo narrative from figures a reviewer approved. Prefer your own model? Bring your own key and she steps aside.
What Vera does
| Task | What you get |
|---|---|
| Document classification | CIM, P&L, Balance Sheet, Cash Flow, Tax Return or Other, detected on upload and confirmed by you |
| Optical text recovery | Scanned and photographed pages transcribed, and marked as machine-read wherever the text travels |
| Financial extraction | Revenue, EBITDA, margins, growth and risk flags by year, each with a confidence score, each cross-checked against the source table |
| Document Q&A | Grounded answers with numbered citations that open the exact passage; "the excerpts do not show that" when they do not |
| Risk summary | A deal-level risk read from the aggregated financials |
| Revenue durability narrative | A written assessment behind the durability score |
| Vera Valuation | A base estimate with a low-high range and a confidence label on the Summary tab before any module runs; once Valuation unlocks, when a QoE period is locked, the AI Valuation sub-tab shows the assumptions and rationale |
| Investment memo | Executive summary, business overview and market analysis, drafted from approved figures only |
| Drafting | LOI drafts at the LOI stage; email drafts to counterparties |
| Playbook customization | Industry question sets tailored to the deal |
The rules Vera works under
- Approved figures only. Memo narrative uses approved QoE add-backs and the selected valuation and financing scenarios. Draft lines never reach the narrative. If no financing scenario is selected, the memo says so instead of printing a stale DSCR.
- Confidence, shown. Every extracted data point carries a confidence score computed from what was found. The same number appears on every screen that quotes it.
- Citations, or nothing. A Q&A answer lists only the sources it cites, numbered from [1] in the order cited. An unsupported answer cites nothing.
- Refusals, surfaced. If the model refuses or cannot complete, you see that. You never get a half-finished memo presented as whole.
- One account of the truth. The deal context an external assistant reads through MCP states the same figures in the same words as the memo.
Where AI runs
Customer data at rest lives in Canada; platform AI is processed in the United States today, on a deployment type under which an individual request may be processed in another region where the model is available, under terms that prohibit training on your prompts and outputs, with processing in Canada on the roadmap (full detail on Security).
Governance
- Per-plan monthly AI analysis ceilings (100 / 500 / 1,500 / uncapped) with a usage dashboard that shows where you stand.
- Prompts are versioned; the catalogue is documented.
Your key, your model, your bill.
Switch the organization from the platform AI to your own provider under Admin → AI Settings.
- Providers: OpenAI or Anthropic Claude.
- Models: the list is fetched live from your provider with your key, so it shows exactly what your account can call, including models released after this page was built. Or type any model id. Or leave it on Recommended.
- One model, every feature. The model you choose runs Q&A, classification, extraction, durability, valuation, decisions, LOI drafting and memo narrative.
- Your key is stored once, in a secure key vault, referenced from then on, and never displayed again. Replace it to rotate; remove it to return to Vera.
- Changing provider or gateway always asks for the key again. If you ask for the list of available models before you save a key, the key is sent once for that listing and held only in memory.
- OpenAI-compatible gateways are supported on a public https address, because your key and your deal content travel to that address.
- BYO tenants never see Vera. Every AI surface shows a neutral indicator with your provider and model, or the name you give it.
What comes out of it
Every upload classified, the financials extracted with a confidence score on each figure, answers with numbered citations that open the passage, and the memo's executive summary, business overview and market analysis drafted from approved figures only.
Questions
Does Acquiror AI train on my data?
No. The platform does not train models, and the platform's provider contractually does not train on prompts or outputs.
Can I turn AI off?
Extraction and Q&A are how documents become figures, so they stay on. You choose the provider, the model and the key that run them.
Which model is Vera?
Vera runs on Acquiror AI's managed models. You never see or configure the underlying provider; that is the point of the platform option.
Join the waitlist for the 14-day trial.
Sign-up is not open yet; we will email you when it opens.