Hub for Investment Screening & Due Diligence

Early-stage deal sourcing is bounded by analyst hours, and the way firms cope is to look at fewer opportunities. What arrives is a stream of pitch decks — PDFs, mostly, with the substance in charts — that have to be read, assessed against criteria, and either progressed or dropped. Because each analyst applies the criteria slightly differently, the pipeline is not comparable across the team, and an investment committee ends up weighing memos written to different standards about companies screened on different thresholds.

• The number of opportunities reviewed is capped by how many decks a team can read.
• Screening criteria are applied inconsistently, so pipeline comparisons are unreliable.
• Due diligence questionnaires are completed from scratch for each opportunity.
• Investment committee reports vary in structure and depth depending on who wrote them.

Hub enriches the ontology with insights extracted from pitch decks and other document types using multimodal language models, then combines those with structured internal and third-party sources. Screening runs against user-configured criteria — revenue thresholds, growth over a stated period, geography — with rules that automate movement of an opportunity through the review process while retaining the ability to conduct a thorough human review and override at any point. Due diligence is accelerated in two directions: suggested answers are offered for preset questions, and a set of bespoke questions is proposed by the model based on the specific deck being analysed and the answers already generated, so gaps are surfaced rather than missed. Reports for investment committees are generated from templates whose predefined elements are populated from the ontology — a structured foundation that constrains generation and mitigates the risk of hallucination — and reviewed and edited by a person before they are exported or shared.

Pitch decks are the reason a model is required: the substance sits in charts, tables and design-heavy layouts, and a text extractor returns fragments. Multimodal extraction turns the deck into structured facts the screening rules can act on. The bespoke-question capability is the part that is genuinely additive rather than merely faster — proposing what else should be asked about this specific company is judgment work that scales badly with headcount. Report generation deliberately runs the other way: the ontology supplies the predefined elements so the model assembles rather than invents, and a human edits before anything leaves the firm.

EXTRACTION

Multimodal — pitch decks, PDFs, mixed document types

SCREENING

User-configured criteria with automated progression rules

DUE DILIGENCE

Suggested answers plus bespoke questions proposed per deck

REPORTS

Templated, populated from the ontology to constrain generation

METHODOLOGY

Standardised across the organisation, configurable per firm

AUTONOMY

Human review and adjustment available at every stage

1. 01Inbound pitch decks screened automatically against configured, firm-specific criteria.
2. 02Opportunities progressed through review by explicit rules, with human override retained.
3. 03Suggested answers for preset due diligence questions, drawn from the enriched ontology.
4. 04Bespoke follow-up questions proposed from the specific deck and the answers already generated.
5. 05Unanswered questions exportable, with follow-up material uploadable for fast completion.
6. 06Investment committee reports generated from ontology-backed templates, then human-edited.

One method

Applied across the team, against per-analyst variationdesign intent

Ontology-bound

Report generation, to mitigate hallucinationdesign intent

• SPEND INTELLIGENCEHub for Intelligent Purchasing→
• DUE DILIGENCE RESPONSEHub Response for RFPs & DDQs→
• RELATIONSHIP BANKINGHub for Real-time Customer Intelligence in Banking→