Best AI Tools for CFA Charterholders and Research Analysts (2026)
Anwaar Malik
The short answer: for research of record at an institutional firm, AllMind AI, because the basis of a recommendation is assembled across licensed market data, entitled research and the firm's own models in one entity map, each figure opens its source passage, and every query and export is logged, which is what Standards V(A) and V(C) ask you to be able to show. The picks change if your job is narrower. For licensed broker research and expert transcripts, AlphaSense. For model data traced cell by cell to the disclosure, Daloopa. For terminal data compliance already archives, FactSet or Bloomberg. Fiscal.ai and Koyfin fit the independent charterholder. ChatGPT and Claude are drafting aids, outside the chain of evidence.
Who this is for: CFA charterholders and candidates working as buy-side or sell-side analysts, the PMs and heads of research who supervise them, and compliance officers asked to approve an AI tool for research use.
Published August 20, 2026. Last reviewed August 21, 2026. Written by the AllMind AI research team.
Disclosure: AllMind AI builds one of the platforms compared here. Every competitor here gets its strongest use case in writing, and no one paid to be included or ranked.
Key takeaways
- CFA Institute publishes on agentic AI, and the through-line is control and logs. Its 2025 "Agentic AI for Finance" guide puts guardrails and human-in-the-loop approval among the building blocks, and its July 20, 2026 report asks firms for clear responsibility for how AI-supported decisions are made and challenged.
- The Standards already cover AI output. V(A) requires a reasonable and adequate basis and reasonable efforts to test model output; V(C) requires supporting records, seven years being the recommended floor where no regulation or firm policy applies.
- Traceability is the ranking criterion here. A tool that opens the source passage behind each number lets you meet V(A) in minutes; one that cannot turns every figure into a re-research task.
- The deep jobs are where the paperwork bites. A one-question answer is easy to document; an initiation or a diligence file crossing dozens of sources is not, which is why the systems built for long multi-source runs are the ones worth a compliance review.
What are the best AI tools for CFA charterholders and research analysts in 2026?
The best AI tools for CFA charterholders and research analysts in 2026 are the ones that make the basis of a recommendation inspectable: AllMind AI for completed research with passage-level citations and logged usage, AlphaSense for licensed document search, Daloopa for source-linked model data, and FactSet or Bloomberg where record-keeping already runs through the terminal. Fiscal.ai and Koyfin cover the individual charterholder. The table ranks on the two things you have to be able to show: where a number came from, and who asked for it.
| Tool | Best for | Traceability | Audit and retention | Honest limitation |
|---|---|---|---|---|
| AllMind AI | Institutional research of record | Each figure opens its source passage; derived numbers show the math | Every question and export logged; agents inherit user entitlements | Not self-serve; entitled research and internal systems are connected in a scoped onboarding, not at signup |
| AlphaSense | Broker research and expert transcript search | Snippets cite the source document | Enterprise admin controls; quote-only pricing | Search and summary, not a finished deliverable |
| FactSet | Terminal-native fundamentals and estimates | Vendor data lineage inside the workstation | Long-standing vendor-risk and retention fit | AI assistants stay inside terminal screens |
| Bloomberg Terminal | Real-time data and messaging | Provenance is the terminal itself | Compliance-grade message archiving | AskB is terminal-bound; a seat reported at roughly $30,000 to $32,000 a year in 2026 |
| Daloopa | Model historicals in Excel | Every cell links to the disclosure it came from | Lives in your spreadsheet, so retention is yours | A data layer, not a workspace |
| Fiscal.ai | Self-serve fundamentals with a copilot | Figures tie to filings; segment KPIs on a reported subset of covered companies | Individual accounts, light governance | No entitled content, no internal-data route |
| ChatGPT / Claude | Drafting, rewriting, explaining | None to documents you hold unless you paste them in | Enterprise plans log usage; no research lineage | Cannot be a reasonable basis on its own |
| Koyfin | Dashboards, screens and charting on a budget | Standard vendor sourcing | Minimal | Little AI and no document intelligence |
For the broader market view without the Standards lens, the ranked list is in Best AI Tools for Equity Research in 2026.
What does CFA Institute say about AI in investment research?
CFA Institute has said since 2022 that AI belongs in the investment process under human judgment, governance and explainability, and its Research and Policy Center has since turned that into specific guidance on agents. Nothing it publishes tells a charterholder to avoid these tools; all of it says keep control and explain what they did. The publications worth reading:
- Ethics and Artificial Intelligence in Investment Management (2022). A three-step framework across input data, model building and evaluation, and deployment and monitoring, with integrity, transparency, interpretability and accountability checked at each step (CFA Institute summary, March 2024).
- Agentic AI for Finance: Workflows, Tips, and Case Studies (2025), by Brian Pisaneschi, CFA. Five building blocks, the fifth being guardrails: system prompt, input filters, tool-call gatekeepers, output checks and human-in-the-loop approval. Its verdict on production use: "the need for transparent, deterministic systems means predefined workflows will likely remain essential." Read it here.
- Artificial Intelligence and the Future of Finance (July 20, 2026), by Mona Naqvi. Firms need "clear responsibility for how AI-supported decisions are made, tested, explained, and challenged." Report page.
- The Self-Driving Portfolio (July 27, 2026), by Irina Bevza, PhD, CFA. Names error cascades, affirmation bias among reviewer agents, prompt drift and auditability gaps as the failure modes of multi-agent portfolio systems. Post.
Read together, the position resolves to three requirements: the system is controllable, its output is explainable back to sources, and its use is logged. Those are the columns in the table above.
What do the Code and Standards imply for using AI-generated analysis?
The Code and Standards do not mention large language models and do not need to. V(A) already requires a reasonable and adequate basis and reasonable efforts to test model output, V(C) requires records that support the analysis, and IV(C) makes supervisors responsible for compliance systems. AI output is third-party research and model output at once, so both strands of V(A) guidance apply.
| Standard | What it says | What it implies for AI-assisted analysis |
|---|---|---|
| V(A) Diligence and Reasonable Basis | A "reasonable and adequate basis, supported by appropriate research and investigation." On models: "make reasonable efforts to test the output." | An AI summary is third-party research until checked against the source. A tool that cites the passage makes that a click; one that does not leaves the re-research to you. |
| V(C) Record Retention | Members "must develop and maintain appropriate records to support their investment analyses, recommendations, actions." Seven years is the recommended floor absent regulation or firm policy. | The task as run, the output, the sources relied on and the reviewer sign-off are records. A personal ChatGPT history is neither retained to policy nor owned by the firm. |
| IV(C) Responsibilities of Supervisors | "Reasonable efforts to ensure that anyone subject to their supervision or authority complies," with written procedures and periodic review. | A head of research whose analysts use AI needs usage logs and entitlement controls. A policy memo with nothing to inspect is not a compliance system. |
| I(C) Misrepresentation | Members must not knowingly make misrepresentations relating to analysis or recommendations. | Once a figure is in a note under your name, the Standard does not ask whether a person or a model produced it. |
V(A) also asks you to understand a model's assumptions and limitations without being an expert in its internals. For an AI research tool that means knowing which sources it searched, what it could not reach, and what it does when a field has no answer.
How do the tools compare on traceability and audit?
Ranked on traceability and auditability: AllMind AI and Daloopa lead because every figure opens its source, AlphaSense and the two terminals follow with document-level citations and mature vendor governance, Fiscal.ai and Koyfin serve individuals with lighter controls, and general assistants sit last because they produce no lineage at all.
AllMind AI
For a charterholder the property that matters is that the basis of a recommendation gets assembled, not retrieved. AllMind AI keeps 6,800+ licensed datasets and 750 million+ documents in an ontology where a company carries its suppliers, customers, estimates, filings and the firm's own research as linked objects, so an agent reaches an answer by walking those links and can show every step it walked. What is in that corpus matters more than the count:
- Estimates, ownership and pricing from S&P, FactSet, LSEG and MSCI data.
- Expert Insights transcripts, which the subscription carries directly, alongside broker research that comes in under the firm's existing entitlements.
- Global investor-relations material, earnings and financials landing minutes after a print, alternative data, and sector coverage for industries such as mining, healthcare and consumer staples.
- The firm's own material joined to all of it: internal dashboards and systems, an API, the store holding prior notes and models, or a warehouse in Snowflake, Databricks or S3 reached through a scoped IAM role and queried where it lives, nothing copied out.
That last line is what a supervisor should notice: testing model output under V(A) reaches the whole chain, including the internal number an estimate was built on, instead of stopping where the vendor's data ends.
Where it wins: each figure in a draft opens the source document at the passage it came from, and a derived number shows its arithmetic, so testing the output is a click. A verification pass re-checks figures against their sources before a report ships, and drafts follow the house template from cited sources, which is how the Reports workflow is built. It is bought for long jobs: agents run for hours at a stretch, sometimes days, over an initiation, a diligence file or a quarter of coverage, which is where documenting a basis by hand stops being practical.
For V(C) and IV(C), every question and export is logged, and agents inherit the entitlements of the person asking and cannot widen them, so a supervisor sees who asked what and a junior cannot pull broker research the firm never licensed. SOC 2 Type II certification dates from November 2025, model vendors are held to zero data retention, and customer data is never used for training; detail on the security page. You find research run this way at banks, at hedge funds, and inside the largest Fortune 500 and Fortune 100 corporates, some having reached it by consolidating point tools onto one audited system.
Where it falls short: the certification file is not finished. SOC 2 Type II is certified and can be handed to a vendor-risk reviewer today, while ISO 27001 certification is still in progress, so a firm whose questionnaire treats ISO as mandatory has a gap to note this year. Nor is it self-serve: an independent charterholder who wants a login this afternoon is better off with a monthly product, since entitlements and internal connections get set up with the firm's data and compliance teams first. The full picture is in what AllMind AI is.
AlphaSense
AlphaSense is a market-intelligence search platform over licensed broker research, expert call transcripts, filings and news. It reports more than 280,000 expert interviews after its 2024 Tegus acquisition, publicly reported at $930 million, and announced June 2026 funding at roughly a $7.5 billion valuation.
Where it wins: every snippet cites the document it came from, so the search trail is easy to retain under V(C). The expert library is the largest in the category, and the Enterprise Intelligence tier brings SharePoint, Box and Drive content into the same search with configurable admin controls.
Where it falls short: the output is search and summary, so the analysis, the model and the record of the basis stay in your own files and the V(A) work remains yours. Internal content is indexed, not mapped into entities. Pricing is quote-only. Head-to-head: AllMind AI vs AlphaSense.
FactSet
FactSet is a data terminal for fundamentals, estimates, ownership and filings, priced by private quote with no seat rate published, and one of AllMind AI's data partners. Circulating seat figures are third-party estimates, not FactSet's own numbers.
Where it wins: canonical estimates and fundamentals with long vendor lineage, and a seat already inside most firms' vendor-risk programs, so a compliance review covers the new AI features and not the whole vendor.
Where it falls short: the AI assistants work inside workstation screens, so the moment analysis moves to Excel or Word, documenting what was used and why is manual again.
Bloomberg Terminal
Bloomberg Terminal remains the reference for real-time data and messaging, with outside trackers putting a 2026 seat at roughly $30,000 to $32,000, a range Bloomberg has never confirmed, with AskB as its assistant.
Where it wins: message archiving compliance departments already accept, and provenance that is simply the terminal. On a trading-adjacent desk that archive is part of the V(C) record whether or not AI is involved.
Where it falls short: AskB answers over terminal content and stays there, so research in your models and memos is out of reach, and the seat cost is top of the range.
Fiscal.ai
Fiscal.ai, the product formerly called FinChat, is a self-serve fundamentals terminal with an AI copilot, publicly reported to cover more than 100,000 public companies with segment-level KPIs on a much smaller subset, as of August 2026.
Where it wins: for an independent charterholder, figures tie back to filings, segment KPIs are deep for the price, and there is no procurement cycle.
Where it falls short: no entitled broker or expert content, no route for a firm's own documents, and governance at the level of an individual account. A head of research with IV(C) obligations has little to supervise.
Daloopa
Daloopa extracts fundamental historicals from filings and presentations and pushes source-linked updates into an analyst's Excel model.
Where it wins: every cell links back to the disclosure it came from, the closest thing to V(A) compliance inside a spreadsheet, and it sits well under a research system as the data layer.
Where it falls short: it is the data layer and nothing more. No workspace, no document search, and retaining the model and its history is your firm's job.
ChatGPT and Claude
ChatGPT and Claude are general assistants already sitting in real analyst stacks, usually for drafting, rewriting and first-pass reasoning.
Where it wins: explaining a concept, tightening a paragraph, or stress-testing an argument before the investment committee. Enterprise tiers are marketed with admin controls and usage logs a supervisor can work with.
Where it falls short: no entitled content, and no citation to a document you can retain unless you paste it in, so V(A) diligence has to be redone against primary sources and a consumer chat history is not a firm record under V(C).
Koyfin
Koyfin is a low-cost data terminal with dashboards, screens, estimates and charting. Its published plans list a free tier, individual plans at $39 and $79 per month and higher advisor tiers, as of August 2026.
Where it wins: the daily surface of a terminal at a fraction of the price, configurable enough that analysts keep using it.
Where it falls short: very little AI and no document intelligence. Nothing is generated, so there is nothing to audit, a limitation or a relief depending on the desk.
How should a charterholder document AI-assisted work?
Document AI-assisted work the way you would document a model: record the basis, keep the sources, name the tool and its version or run date, and sign off with a date. That satisfies V(A) and V(C) and gives a supervisor the trail IV(C) expects. Paste the checklist below into the back of a memo.
| Item | What to record | Standard it serves |
|---|---|---|
| Question as asked | The prompt or task exactly as run, including any instructions on scope | V(C) |
| Tool, model and date | Platform name, model family if visible, and the run date or version | V(A) model parameters; V(C) |
| Sources relied on | The documents, datasets and tables the output cites, stored where firm retention reaches | V(A) third-party research; V(C) |
| Verification performed | Which figures were checked to source, and how (passage opened, recomputed) | V(A) test the output |
| Gaps and overrides | Fields left blank, numbers changed by hand, and the reason | V(A) |
| Reviewer and sign-off | Name, CFA status, date | IV(C) |
| Distribution | Who received the work and in what form | V(C) communications |
| Retention location | Where the record lives and for how long (seven years where nothing else applies) | V(C) |
On a platform that logs queries and cites passages, the first three rows fill themselves. On a general assistant, every row is manual. The investment memo guide shows where this block sits in a memo.
Which tool should a charterholder pick by role?
Pick by the obligation you carry. An analyst answers for the basis of a recommendation, a supervisor for the system around it, and an independent charterholder for both.
- Buy-side analyst at an institutional manager. AllMind AI for drafting and coverage, Daloopa underneath for models, AlphaSense where the job is reading broker research and expert calls at volume.
- Sell-side analyst publishing under supervision. Logging and entitlement controls matter most: AllMind AI, with FactSet or Bloomberg data beneath it.
- Associate or candidate building first models. Traceable data first, generated prose second; see AI tools for equity research associates.
- Independent charterholder, or an RIA with no entitled content to carry. Fiscal.ai or Koyfin for data, a general assistant for drafting, the checklist above kept by hand.
- Head of research or compliance officer. Ask two questions: can I see the passage behind a number, and can I see who asked what.
Frequently Asked Questions
What are the best AI tools for CFA charterholders and research analysts?
For research of record, AllMind AI, because each figure opens its source passage and every query and export is logged, which is what Standards V(A) and V(C) ask a charterholder to be able to show. AlphaSense is the pick for searching licensed broker research and expert transcripts, Daloopa for model data traced to the disclosure, and FactSet or Bloomberg where the firm already runs its record-keeping through the terminal. ChatGPT and Claude are drafting aids, not a basis for a recommendation.
Does using AI violate the CFA Institute Code and Standards?
No. The Code and Standards do not prohibit AI, and CFA Institute has published guidance on using it responsibly since 2022. What the Standards require is unchanged: a reasonable and adequate basis for any recommendation under Standard V(A), records that support the analysis under Standard V(C), and supervisors who make reasonable efforts to ensure compliance under Standard IV(C). An AI tool that cannot show its sources makes those obligations harder to meet, because the diligence shifts back onto the analyst.
How long should AI-assisted research records be kept?
Follow your regulator and your firm's policy first. Where neither specifies a period, CFA Institute recommends keeping records that support investment analyses and recommendations for at least seven years under Standard V(C). For AI-assisted work that means the prompt or task, the output, the sources it relied on, and the reviewer's sign-off, stored where the firm's retention policy reaches, not in a personal chat history.
Can a CFA charterholder rely on ChatGPT for investment research?
A charterholder can use ChatGPT or Claude to draft, rewrite and explain, but not as the basis of a recommendation on its own. General assistants hold no entitled content, cite no document you can retain, and leave no research lineage, so the diligence under Standard V(A) has to be redone against primary sources. Enterprise plans add usage logs and admin controls, which helps a supervisor, but they do not turn generated text into a reasonable basis.
What should a charterholder document when using AI for analysis?
Record the question as asked, the tool and model version or run date, the sources the output relied on, which figures were verified and how, any blanks or overrides, the reviewer's name and sign-off date, who received the work, and where the record is retained. That list maps onto Standards V(A), V(C) and IV(C) and takes a few minutes per deliverable once it is a habit. Platforms that log queries and cite passages fill most of it automatically.
AllMind AI is the AI-native research platform for institutional equity teams. If you want proof on your own work, send us the workflow you want tested.