Limited published evidence · Asset management and investment research
AI investment research: Franklin Templeton example
Franklin Templeton is developing AI workflows to compare equity-research views and surface risks. OpenAI describes experiments and a path toward production, not a verified production outcome.
Develop connected equity-research workflows for comparing perspectives and surfacing risks
Franklin Templeton
OpenAI’s 5 November 2026 account describes ongoing development, hackathons and Investment AI champions converting experiments toward production-ready workflows. It does not establish a deployed production research workflow or measured investment outcome.
A published third-party example, not a Nuvepro customer case or evidence of a partnership.
The workflow at a glance
Franklin Templeton’s development-stage research approach
Illustrative summary of the published account
Reported work components. Their arrangement does not imply a fixed execution order or simultaneous work.
- Current human workflow
Analyst research
First-pass stock analysis, earnings-call review and competitor analysis by business line are described.
- Build-stage work
Develop connected intelligence
Codex and other tools are used with agents and a semantic layer to develop investment-research workflows.
- Stated ambition
Compare perspectives + risks
The firm wants diverse research, risk, compliance and trading inputs surfaced for portfolio managers.
- Source boundary
Production status not established
Investment AI champions are working to convert experiments into production-ready workflows with controls.
What the source reports
OpenAI reports that Franklin Templeton is developing AI workflows beginning with equity research. The firm added agents and a semantic layer after a direct database connection did not provide enough understanding of relationships in its data. Its stated aim is to combine research, risk, compliance and trading perspectives for portfolio managers and preserve diversity of research rather than reinforce an individual’s preferred sources. The existing analyst workflow described in the account includes first-pass stock analysis, earnings-call review and competitor analysis by business line. Hackathons, security review, expert support and data controls are part of the development path. The source does not establish that the envisioned connected workflow is in production or quantify an investment result.
Reported technology: Codex and other AI tools with agents and a developing semantic layer. Scope and outcomes are source-reported.
What the enterprise environment would need
The following is a proposed project design for an enterprise doing similar work. Published details are marked; the remaining systems, data and scenarios would be agreed for your project.
Proposed practice environment
Build a balanced research briefing exercise
Illustrative project design to scope with your enterprise.
- 1 · Set up the environment
Dated fictional source set
Company filings, earnings-call extracts, competitor facts and distinct research-provider views.
- 2 · Preparation and agent contributions
Agent drafts comparison + risks
Summarize views, link each claim to a dated source and preserve disagreements.
- 3 · Independent checks
Check identity, dates + diversity
Independently verify issuer, reporting period, calculations and source representation.
- 4 · People decide acceptance
Analyst + PM accept bounded brief
Review risks and unresolved evidence; no recommendation execution or trading.
A reviewable evidence package
Versioned briefing with citations, perspective comparison and unresolved questions.
If a check fails → withhold acceptance.
Several reports all reflect the same source and a material downside is omitted. Hold acceptance, add independent perspectives, recheck citations and return to owner review.
Failure or owner rejection returns to review
- Hold
- → 2 · Revise work
- → 3 · Rerun independent checks
- → 4 · Owner review
Operating boundary. Fictional research exercise; no client data, investment recommendation or portfolio action.
- Systems and tools
- Published: Codex and other tools are being used to develop investment-research workflows; agents and a semantic layer are part of the evolving approach. Proposed: a mock research workspace with versioned company, earnings-call and competitor-source documents.
- Data and state
- Proposed: pin dated source excerpts, issuer identity, reporting period, research-provider identity and portfolio context. Preserve conflicting views rather than ranking by one analyst’s past preference.
- People and handoffs
- Published: analysts and portfolio managers retain human judgment; investment experts participate in hackathons and Investment AI champions develop workflows. Proposed: an analyst checks source attribution and a portfolio manager accepts the bounded briefing.
- Rules and authority
- Proposed: prepare research summaries and risk questions only. No trading instructions, portfolio changes, external publication or investment recommendation execution.
- Normal and exception cases
- Proposed: balanced source perspectives, conflicting earnings signals, missing risk disclosure, stale filing, incorrect issuer match, duplicated research and a source set dominated by one provider.
- Reset and replay
- Proposed: restore the same dated source set, semantic mappings and portfolio context; log all sources and revisions so the briefing can be independently reproduced.
A proposed simulation
An illustrative practice design derived from this work, not an executed assessment.
Proposed: prepare a source-linked fictional equity-research briefing that compares independent views and surfaces risks for human review.
Review gate: A separate checker verifies issuer, period, citation, calculation and source diversity; an analyst and portfolio manager accept the exact versioned briefing.
Challenge the result: The briefing includes several reports but all derive from one preferred research source and omits a material downside. The diversity and risk checks must hold acceptance and request a broader source set.
What needs checking
- Source boundary: the workflow is in development; security review, expert support and data controls are described as part of the path toward production-ready workflows.
- Proposed: verify every claim against the correct issuer and source date; independently check calculations against source data.
- Proposed: measure source diversity and representation of material risks using a separately prepared reference set, not the summarizing agent.
- Proposed: an analyst and portfolio manager accept the exact briefing and unresolved risk list. These checks are designed, not executed.
Review environment fidelity, human capability and agent performance separately. The enterprise’s accountable owners decide acceptance of the shared workflow and any operating use.
How an enterprise could build similar work
A proposed build order for an enterprise doing similar work, written by Nuvepro from the published account and our interpretation of the environment and checks above. It is not how Franklin Templeton built or operates this work.
- Choose a fictional research questionDefine one issuer and period using a bounded, fictional research brief.
- Prepare a diverse source setVersion filing, earnings-call and distinct-provider excerpts, preserving source date and issuer identity.
- Draft a comparison and risk listSummarize differing views with claim-level citations and keep unresolved conflicts visible.
- Check evidence independentlyVerify citations, issuer, reporting period and calculations; check that material views are not all derived from one source.
- Repair missing perspectivesHold the brief if a source is stale, misidentified or materially one-sided. Add the needed evidence and rerun checks.
- Analyst and PM acceptThe analyst checks source meaning and the portfolio manager accepts the exact briefing. No investment or trading action is included.
Your systems, data, rules and accountable owners decide the real project. Agree acceptance criteria, escalation owners and operating responsibility with the people accountable for the work before anything runs on live records. Treat the first run as a rehearsal on test records: compare the draft output with an independently prepared answer, and record what the people reviewing it change.
Questions about this example
- What investment work is Franklin Templeton exploring?
- The OpenAI account describes development of equity-research workflows that could bring research, risk, compliance and trading perspectives together for portfolio managers.
- Is the connected research workflow in production?
- The source describes ongoing development, hackathons and Investment AI champions working toward production-ready workflows with security review, expert support and data controls. It does not establish production deployment.
- Why did the firm add a semantic layer?
- Franklin Templeton reports that a direct database connection did not give models enough understanding of how the data fit together. The firm added agents and a semantic layer while developing its connected intelligence approach.
- What would a similar practice environment require?
- A proposed exercise uses dated, fictional research sources and issuer context. Independent checks validate identity, citations, calculations, source diversity and risk coverage; an analyst and portfolio manager review the exact briefing.
Shape this around your enterprise
Bring a similar workflow and the people responsible for it. Together, define the records and systems to use, the result your team will build or configure, and how its owners will check it. Nuvepro scopes the simulation, environment and preparation. Agree any existing-agent evaluation, implementation assistance and production responsibilities separately.
Sources and limitations
Curation review: . This is the date this source example was reviewed, separate from source publication dates.
- OpenAI is the publisher and technology provider; no independent evaluation is provided.
- The account does not establish production deployment, workflow coverage, measured accuracy or investment outcomes.
- The semantic layer, connectors and future shared intelligence foundation are described as development or ambition; detailed system architecture is not disclosed.
- The proposed environment is fictional and unexecuted; it does not reproduce Franklin Templeton data, research systems or investment controls.
- Franklin Templeton uses OpenAI to support investment researchSource publication: 2026-11-05