Evidence-to-decision platform / Built in Australia

Make better decisions, faster. Keep the evidence at your fingertips.

Convergence takes a complex business question from research to a decision-ready answer in one governed workflow. It finds and tests the evidence, builds the analysis, exposes what remains uncertain, and keeps every conclusion traceable to its source. AI does the heavy reading. People hold authority over what counts.

Built initially for advisory, investment and risk teams producing evidence-heavy work that must withstand client, committee and regulatory scrutiny.

One workflow from question to decision: scope, source, test, decide, monitor.
Evidence, assumptions, disagreements and review history stay attached to every conclusion.
The work is reused: the same record supports monitoring, reassessment and adjacent decisions.

Request a controlled evaluation See how it works

A working, founder-operated product currently being evaluated on live, buyer-defined questions for three external organisations.

Assertion CC-0231 Held: measure mismatch

"Australian childcare occupancy fell 2.9 percent year on year in the June quarter 2026."

Source: Department of Education, Child Care Subsidy data report, June quarter 2026 (published 18 September 2026), child-count table. Role: official statistics.
Tension: T-044: the source counts unique children using centre-based care. It does not measure occupancy, which needs a places denominator the source does not carry.
State: Held corrected wording offered: "the number of children using centre-based care fell 2.9 percent"

The claim is excluded from the decision output until its measure matches its source. The correction, and who accepted it, travel with the record wherever it is cited.

Illustrative assertion record, modelled on a control run in a live evaluation (Australian childcare, September 2026). The platform's core object: a claim that carries its receipts.

3External evaluations

3Unrelated domains

2 daysOfficial-data refresh completed

Working MVPFounder-operated today

The problem

AI output is scaling faster than institutional verification can handle.

In 2025 Deloitte delivered a A$440,000 assurance review to a federal department. It cited academic references that did not exist and included a quotation absent from the cited Federal Court judgment. The revised report disclosed use of Azure OpenAI GPT-4o. The deliverable was assurance work. The evidence record still failed. An outside academic found the errors, the report was reissued and the final instalment repaid.

Source: departmental correspondence released publicly; corrected report, October 2025.

This is not only a model failure. It is a control failure: fluent output arrives faster than anyone can check it. Disagreement between sources is averaged into false consensus. And when a decision is questioned later, nobody can show which evidence was relied on, by whom, or why. AI accelerates synthesis. High-stakes decisions also require conflicts, definitions and uncertainty to remain visible.

Convergence does not make hard decisions easy. It makes them faster, more precise and defensible. The difficulty in complex evidence is information, and the platform is built to preserve it: visible, structured and reviewed, so that people can make the best call available and show exactly how they made it.

Worked example

From a buyer's question to a refreshed answer, in one governed record.

A deals-strategy team asked when Australian childcare would recover, and why. The same evidence record produced the decision brief, preserved its limits and re-tested the answer against the next quarter's official release within two days of publication.

Question

"At what point does the sector turn around, and why?"

Buyer-defined. Reader: a partner-level deals team.

Answer

The June quarter weakens the case for a near-term national turnaround. Children using care fell 2.9 percent while operating services grew 2.9 percent; child numbers now need to grow about 2 percent a year just to stabilise activity per service.

Tested

Every figure is re-derivable from Department of Education, ABS and listed-operator disclosures. The earlier stabilisation case was checked against the new release and the check's own limits remained visible.

Held

Sector proxies and listed-operator occupancy were not treated as equivalent: definitions, periods and portfolios differ. No earnings-recovery date was released without catchment and centre data.

Candidate output built from public sources: eleven primary pages and five supporting pages, with sources and calculations attached throughout. A public sample will be linked once source-rights review is complete.

External evaluations

Tested on buyer-defined work across three unrelated domains.

Convergence is being evaluated on live questions for three external organisations. The platform remains founder-operated and none is yet a paid deployment. In each case, the output is judged by the people who would use it.

Commercial due diligence · Australian childcare

For a Big Four deals-strategy team

A buyer-facing decision brief on sector recovery, revised on partner feedback and refreshed within two days of the June-quarter Child Care Subsidy release with every figure re-derived from public sources.

Energy-transition pathways · 14 Pacific island countries

For an impact-communications firm

A country-by-country assessment of the renewable build, capital requirement and financing gap behind national 100 percent renewable commitments, with unsupported commitments held rather than inferred.

Manuscript evidence review · global eye health

For an international eye-health research team

An evidence adjudication for a five-country caregiver study, with suggested wording and checkable excerpts. The review surfaced a potentially material correction to the manuscript's account of the literature, now under expert review.

How it works

One governed workflow, from question to decision.

AI supports discovery, extraction, evidence classification, claim and tension mapping, synthesis, conditional forecasting and decision composition. Deterministic controls and named reviewers govern what can be relied on and released.

1 / Scope

The question, the decision, the reader

What must be answered, for whom, and what would change the view. The obligations of the question are compiled before any source is read.

2 / Source

Evidence with custody

Sources are retrieved, versioned and hashed. Each claim is extracted with an exact source locator, and the source's role is classified on the record.

3 / Test

Disagreement on the record

Contradictions, incompatible definitions and boundary conflicts are mapped as first-class records. Comparisons that are not permitted are held, with the reason typed.

4 / Decide

A decision-ready answer

Answer first, then what it rests on, what remains uncertain and what would change it. Rendered fail-closed from accepted state only.

5 / Monitor

The record keeps working

New releases are tested against the earlier case. The same record supports reassessment and adjacent questions without starting again.

Decision control

Every claim carries its review state.

Scaffolded

Machine-produced, structured and traceable. Useful for orientation; barred from decision outputs.

Reviewed

Inspected by a named reviewer. Provenance, classifications and known tensions have been checked.

Governed

Released under defined rules and fit to carry decision weight, with the review history intact.

Financial services

Initial applications for decisions banks, insurers and funds must defend later.

Financial-services regulation increasingly assigns named accountability for how key functions are managed. Convergence preserves the evidence, assumptions and review trail as the work is performed, rather than reconstructing them after a challenge.

Investment and credit committee papers

Committee decisions that must survive later review by boards, auditors and regulators, with reliance, contested points and reviewer identity on the record from the start.

Financial Accountability Regime: banks from March 2024; insurers and superannuation from March 2025.

Risk, model and AI governance

Evidence that AI-assisted analysis was checked, by whom, against which sources. A record the risk function can inspect, rather than an assurance it has to take on trust.

APRA CPS 230 Operational Risk Management: current amended standard in force from 1 July 2026.

Disclosure and customer-facing decisions

Product, disclosure and remediation positions traceable to source, with disagreement visible before publication rather than after enforcement.

ASIC greenwashing penalties: A$34.7 million across three funds, 2024 to 2025.

Transaction diligence and transformation assurance

Commercial due diligence, portfolio reviews and programme assurance produced faster and more precisely, with the audit trail delivered as part of the work.

In live evaluation: commercial due diligence with a Big Four deals-strategy team.

The initial financial-services scope is evidence-heavy analysis and decision support. Fraud detection, transaction monitoring and identity are outside the current product scope.

Enterprise readiness

Governed by design. Hardened in stages.

The governance architecture is in place and exercised. The enterprise controls around it are a roadmap, stated as one. Models operate through bounded interfaces: they propose, while governed records and named reviewers hold authority over what can be released.

In place

Source custody: versioned documents, SHA-256 hashes, exact passages
Typed assertions with source roles and comparison permissions
Immutable review events with reviewer identity
Release gates; fail-closed rendering from accepted state only
Typed holds with adversarial and negative-control testing
Reproducible builds: the same accepted state yields the same record

In progress

Identity federation and role-based access
Tenant isolation and secure source handling
Audit logging to institutional standard
Australian-hosted deployment
Independent security and architecture review
Second-model semantic checks alongside the deterministic controls

Founder

Twenty years making and defending material decisions.

James Drury

Founder

"The biggest problems are always people problems."

Decision discipline. Ten years as a derivatives lawyer, then strategy consulting on commercial due diligence at Strategy&: every seat where a decision had to be defended after the fact.

Operator. Director of APAC Strategy and Operations at DoorDash, running ANZ supply strategy across a 70,000-plus Dasher network and a cost base of about A$1 billion with a fifteen-person team. Before that, General Manager of Customer Experience at Endeavour Group, leading more than 150 product, design, engineering and analytics staff.

Builder. Designed and shipped Convergence's governed core alone in six months: SQL, Python, hands-on architecture. Convergence is what that experience builds once you accept that the failures in AI-assisted decisions are people failures, of structure, incentive and attention, and that structure rather than more training is the fix.

Team and partners

Founder-led today. The product is presently built and operated by James. The first planned hire is a senior technical leader for backend, AI infrastructure and enterprise deployment.

Research pathway. The University of Technology Sydney is the proposed research partner in a submitted CRC-P grant application focused on governed AI and climate-transition decision-making.

External validation. Current evaluations are buyer-defined and reviewed by the advisory and research teams described above. None is represented as a customer deployment.

Company. FJD Consulting Pty Ltd, trading as Convergence. Sydney, Australia.

Platform transfer

Start with one valuable workflow. Reuse the decision infrastructure.

The initial entry point is specialist advisory, investment and risk work. The same governed primitives transfer when another domain has contested evidence, accountable reviewers and a decision that must withstand scrutiny.

Financial services and investment

Can we show our work to the board and the regulator?

Committee, risk, disclosure and diligence decisions supported by a record that documents what was relied on, what was contested and who reviewed it.

Research and advisory firms

Can we deliver rigour at commercial speed?

Evidence work with systematic-review discipline: explicit inclusion, source roles, documented disagreement, at a pace and price that fits live engagements, with the audit trail as part of the deliverable.

Policy and government teams

Can this advice survive estimates, audit and FOI?

Briefings and regulatory positions backed by evidence traceable to source, with disagreement on the record rather than smoothed over.

Contact

Bring a question that matters.

A controlled evaluation runs on your question, with your sources where possible and under your review. You see the answer, the record behind it and what was held before any commitment is made.

james@convergencehq.io