FROM FIRST SCREEN TO INVESTMENT COMMITTEE
This personal prototype turns CIMs, management calls and quality-of-earnings reports into cited screening notes, NBOs and IC drafts.
Demonstration only. Synthetic data.
Not offered for sale or subscription.
Westgate operates across the South Coast and London. FY25 adjusted EBITDA is , with of revenue recurring through Denplan memberships and NHS contracts.
| Cell | A / Bridge item | B / GBP m |
|---|---|---|
| B3 | Reported EBITDA | 2.40 |
| B4 | Owner remuneration normalisation add-back | 0.18 |
| B5 | One-off practice integration costs add-back | 0.09 |
| B6 | Non-recurring legal settlement add-back | 0.03 |
| B7 | Adjusted EBITDA | 2.70 |
INFORMATION MEMORANDUM / PAGE 14
Group revenue grew from £11.4m in FY23 to £14.2m in FY25, reflecting four practice acquisitions and like-for-like growth of 4% a year.
Recurring revenue represented 72% of FY25 revenue: Denplan memberships 41% and NHS GDS contracts 31%.
Private revenue is billed monthly in advance through Denplan. NHS revenue is contracted annually under GDS terms and paid monthly.
MANAGEMENT CALL / 14 MARCH
110AHAnd the estate today — how many sites are you at?
111CEOThe Bournemouth acquisition completed in February, so the count is current.
112CEOEighteen practices: fourteen along the South Coast and four in London.
113AHAnd all eighteen are on the same practice management system?
Workbook source open: Adjustments, cell B7.
THE WALKTHROUGH / 3:35
Follow Westgate from source documents to an IC decision. Open the evidence, review the findings, and see where a human steps in.
Explore the synthetic deal
Westgate Dental Group · 3:35 A synthetic deal, start to finish Play In the demo
Synthetic deal data · no client or confidential information Prefer to click around? Open live demo →
01 / DEAL WORKFLOWS
From the first screen to the investment committee, the agents work from the same evidence. Explore a second synthetic deal: Foundry Systems, a UK MSP.
FOUNDRY SYSTEMS / MANDATE SCREEN
Sector and size fit. Revenue durability needs a closer look.
| Criterion | Deal | Mandate | Result |
|---|---|---|---|
| Market growthFoundry CIM · Market | 9% CAGR | ≥ 5% | Pass |
| Recurring revenueFoundry CIM · KPIs | 66% | ≥ 70% | Fail |
| Top-5 concentrationFoundry CIM · Customers | 49% | ≤ 40% | Fail |
| EBITDA marginFoundry CIM · Financials | 20% | ≥ 15% | Pass |
| Net revenue retentionRequest cohort schedule | — | ≥ 100% | Gap |
| Equity ticketTarget profile | £20.3m | £10–60m | Pass |
02 / REVISION HISTORY
Every agent reads the same current answer. New evidence can improve it, but the earlier answer, its source and the reason for the change stay on the record.
Contracted revenue · renewal profile
Current record
verified currentNHS activity is on rolling GDS contracts. Private members renew monthly on Denplan.
The fuller answer becomes current without deleting the first.
Source excerpt · one conversation “Renewal terms are mixed: NHS activity is on rolling GDS contracts, and private members renew monthly on Denplan plans.”
Nothing overwritten
03 / PROTOTYPE SAFEGUARDS
What reaches the model. How records are deleted. Current safeguards and planned capabilities, clearly separated.
TODAYModel-agnostic by design: the evidence record and verification rules sit outside the model layer. Provider integrations determine which models can be used. Document text reaches the model during extraction and drafting; ingest itself is local parsing.
RESEARCH IDEAZero-retention and UK/EU regional inference agreements.
TODAYThe record and its sources live in the deployment’s filesystem and database. The audited hard-delete command removes the deal room and its database rows and writes a signed audit line; zero residue is drilled in the test suite, with one named exception — shared counterparty stubs are cleared by a second audited sweep.
RESEARCH IDEAAn append-only (WORM) audit chain for the deletion record itself.
TODAYThe app is passcode-gated and has no write path to the record: every change goes through one audited command-line path, and every fact carries who, what source, when and why.
RESEARCH IDEAPer-user roles and insider-list enforcement (MAR Art. 18).
TODAYNo. Every extracted value carries a citation checked deterministically — the cited page or cell must exist and the cited text must appear there verbatim. A failed check rejects the fact before it is recorded.
TODAYDeal rooms are scoped directories behind a symlink-escape guard, and firm-wide memory learns only from deals that were explicitly promoted after closing.
RESEARCH IDEAContainer-level isolation per deal.
What the demonstration explores
The example deal uses fictional source material: a CIM, QoE and management transcripts.
The prototype extracts values with citations and verification states.
Open the deal, click any figure through to its source, and ask the only question that matters: what did it miss?
An audited hard-delete removes the deal room and its database rows and writes the signed audit line that proves it; a second audited sweep clears any shared counterparty stubs.
Where this goes
Research ideas include isolated deal rooms, insider lists and an append-only audit chain. These are architectural concepts beyond the personal prototype, not available services.
About this project
Built by Oliver Shewan to explore how AI can organise deal evidence and support human review. This is a personal software prototype, shown with synthetic data. It is not offered for sale or subscription, and no pilots or client engagements are available.
Prefer to click around first? Open the live demo — passcode required, synthetic data only.
Synthetic demo · no client or confidential data
Personal project · demonstration only