What problem does it solve?
Morgan Stanley alone publishes more than 70,000 proprietary research reports a year. No advisor or salesperson can read that, so client conversations lean on the few notes someone happened to see, and the firm's own view reaches clients unevenly. Analysts, in turn, spend much of their time on the mechanical parts of writing: sifting through financial statements, regulatory filings and industry reports, and summarizing earnings releases and investor transcripts.
Rewriting research for segments and languages multiplies the work, and every rewrite is a chance for a number to drift from the approved report. The job is to make the research usable at the moment of need without changing what it says.
How does it work?
- Ingest approved research. Published reports, the house view, earnings summaries and market notes are indexed with their publication date, author and distribution rules.
- Answer and summarize on demand. An advisor or salesperson asks a question or requests a briefing; the assistant retrieves the relevant passages and writes a short answer with links to each source report.
- Produce standard briefings. A morning note, a sector summary or a "what changed" digest is generated from a template, with figures and price targets copied from the source rather than generated.
- Adapt for audience and language. Approved summaries are rewritten for a client segment or translated, and the adapted version is checked against the source before use.
- Review before anything branded goes out. Research or compliance signs off on any client facing summary; internal answers carry citations so the reader can verify them.
- Audience
- Employee facing
- Autonomy
- Copilot
- Adoption
- Early adopters
- Channels
- Internal tools, Email, Microsoft Teams
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Users served | Not pooled | about 5000 | 1 | 1 organization |
| Time saved per task | Too few to pool | Not pooled: up to 120 minutes | 0plus 1 up to | 1 organization |
Value drivers: Employee productivity, Speed and cycle time, Customer experience, Compliance quality.
Indicative value
A research and advisory team of 100 analysts and strategists
USD 250,000 to USD 1.5 million
Value of analyst time released from summarizing and drafting per year
How this is calculated
Formula: analysts * documentsPerYear * hoursSaved * hourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Analysts and strategists producing research analysts, people | 100 | 100 | The reference team. |
| Notes and reports per person per year documentsPerYear, documents per person per year | 50 | 100 | Editorial assumption, replace with your own publication volumes. |
| Hours saved per document hoursSaved, hours per document | 0.5 | 0.75 | In line with the Deutsche Bank figure of 30 to 45 minutes saved on earnings note templates. The reported ceiling of up to two hours applies to full research reports and roadshow updates and is not assumed here. |
| Fully loaded analyst cost per hour hourlyCost, USD per hour | 100 | 200 | Editorial assumption, replace with your own fully loaded cost. |
What it leaves out: Covers production time only. It leaves out the value on the distribution side (faster answers to client questions), the cost of running the tool and the review effort for client facing output.
Who already uses it?
4 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Citi
United States · Wealth and asset management · 2025
Citi Wealth launched two AI tools built by its Data, Analytics and Innovation team. AskWealth is a generative AI assistant that gives service teams, advisors and managers answers across the wealth business, so that advisors can reach market insights and research when clients ask questions; after a launch in Asia it became available to Citi Wealth colleagues worldwide. Advisor Insights is a dashboard of timely messages about market moves, portfolios and events, including Chief Investment Office insights, piloted with Citigold and Citi Private Client advisors in North America with a wider rollout planned for Q4 2025 and Q1 2026. Citi says the tools will save hours of time but published no figures.
No outcome disclosed.
Morgan Stanley
United States · Capital markets · 2024
Morgan Stanley Research launched AskResearchGPT, a GPT-4 based assistant that lets investment banking, sales and trading and research staff search and summarize the firm's research (more than 70,000 proprietary reports a year), with hyperlinks to the source reports and a one click transfer of findings into an email draft that staff edit before sending to clients. It is available in the browser, Microsoft Teams and Outlook. Morgan Stanley's global director of research told CNBC that a salesperson needs one tenth of the time to answer the average client inquiry with the tool, and the bank said staff ask three times as many questions as with the traditional AI tool it had used since 2017.
No outcome disclosed.
Deutsche Bank
Germany · Capital markets · 2024
Deutsche Bank Research built DB Lumina, a research agent on Google Cloud and Gemini models that helps analysts ingest documents, summarize earnings releases and investor transcripts, answer questions with inline citations and edit notes, with guardrails, access control and audit logging. It went live in September 2024 and, when the bank described it in September 2025, was used by around 5,000 people across Deutsche Bank Research and divisions such as Investment Bank Origination and Advisory and Fixed Income and Currencies. Analysts report saving 30 to 45 minutes on earnings note templates and up to two hours on research reports and roadshow updates. The bank evaluates it with stable test sets, automated metrics and human review.
- Users served: about 5000, users at publication in September 2025
"Currently, DB Lumina is already in the hands of around 5,000 users across Deutsche Bank Research, specifically in divisions like Investment Bank Origination & Advisory and Fixed Income & Currencies."
Claimed by: organization - Time saved per task: up to 120 minutes, per research report or roadshow update
"Time savings: Analysts reported significant time savings, saving 30 to 45 minutes on preparing earnings note templates and up to two hours when writing research reports and roadshow updates."
Claimed by: organization
UBS
Switzerland · Wealth and asset management · 2024
UBS built two domain specific assistants, together called UBS Red, on Azure AI Search and Azure OpenAI Service to give client advisors fast, multilingual access to the bank's investment advice and product content during client work. UBS digitized about 60,000 investment advice and product documents into a queryable knowledge base, which it says saves considerable time in meeting preparation and research. Within 10 months the wider Azure OpenAI footprint reached key wealth, banking and operations divisions in the Switzerland, Hong Kong and Singapore booking centres. No usage or time saving figure specific to UBS Red is published.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Research archive with publication dates, authors, ratings and distribution permissions
- Current house view and CIO publications
- Market data feeds where briefings reference prices or moves
- Style guides and disclaimers per client segment and market
Systems to integrate
- Research publishing platform and archive
- Market data provider
- Document management and translation workflow
- Advisor or sales desktop, email and collaboration tools
Complexity: Medium
Retrieval and summarization are straightforward. The effort is in distribution rights (which research may be shown to whom), keeping numbers exact, and a review workflow for anything that reaches clients under the firm's name.
- 1
Start internal, with citations
Launch as an internal question answering tool over published research, where every answer links to the source report. Internal use builds trust and shows which questions matter.
- 2
Copy numbers, never generate them
Extract figures, ratings and price targets from the source with structured extraction and insert them into the text, then check the final output against the source automatically.
- 3
Respect distribution rules
Filter research by audience, market and embargo before retrieval so restricted or institutional only content never appears in an answer for the wrong reader.
- 4
Add templated briefings
Once answers are reliable, generate recurring digests (morning note, weekly house view changes) from templates owned by the research team.
- 5
Put client facing output through review
Any summary sent to clients goes through the same approval as other research or marketing communications, with the AI draft and the source retained.
Guardrails
- Answers only from published, approved research with citations and dates
- Numbers, ratings and price targets copied from source and verified, never generated
- Distribution and embargo rules applied before retrieval
- Human sign off before any branded or client facing summary is sent
- Refusal to give personalized recommendations; the assistant summarizes the firm's view
KPIs to instrument
- Share of answers with a valid citation and a matching figure check
- Time from report publication to advisor ready summary
- Weekly active users by desk
- Error rate on a monthly sample checked by analysts
- Client facing summaries rejected at review, with reasons
Human in the loop
Analysts approve summaries of their own work before external use; research management or compliance signs off on client facing templates; readers of internal answers verify through the cited source.
Common failure modes
- Drifting numbers
- A price target or percentage in the summary differs from the report. Copy numbers from structured extraction and check them automatically.
- Stale view presented as current
- An older note outranks the latest update. Weight recency, show dates and retire superseded views.
- Distribution breach
- Institutional or embargoed research reaches a retail audience. Filter by entitlement before retrieval.
- Summaries that read as advice
- A generic summary is sent to a client as if it were personal advice. Keep client facing use behind review and templates.
What are the risks and rules?
EU AI Act
Depends on design
Summarizing research for staff is not an Annex III use and is not a practice prohibited by Article 5, so the tier turns on the firm's role under Article 50. It is minimal for a purchased internal tool with no client or public facing exposure. Article 50 transparency applies when the firm builds the generating system itself, which brings the Article 50(2) duty to mark synthetic text in a machine readable format; when the assistant is offered to clients as a chatbot, which brings the Article 50(1) duty to tell them they are interacting with AI; or when AI generated text is published to inform the public on matters of public interest, which brings the Article 50(4) disclosure duty unless the text has gone through human review or editorial control and a person holds editorial responsibility for it.
Rules that apply
Guidance
- ESMA public statement on the use of AI in the provision of retail investment services (European Securities and Markets Authority, Europe). Names hallucination and overreliance as risks and expects accuracy controls and records when AI supports investment services.
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). Providers must mark AI generated text in a machine readable format (paragraph 2); the disclosure duty for text published on matters of public interest does not apply after human review under editorial responsibility (paragraph 4).
- Principles to Promote Fairness, Ethics, Accountability and Transparency (FEAT) (Monetary Authority of Singapore, Asia Pacific). Principles for the responsible use of AI and data analytics by Singapore financial firms, including accountability for AI driven outputs and transparency to customers about AI use.
Controls to put in place
- Source traceability for every summary, retained with the output
- Automated figure check against the source before release
- Review and approval workflow for client facing summaries
- Entitlement and embargo filtering tested on every change
- Inventory entry with owners in research and distribution
Frequently asked questions
- Will the AI invent numbers or price targets?
- It can, if you let it write numbers freely. The safe design copies every figure from the source report, checks the output against it, and keeps citations visible. Deutsche Bank's DB Lumina, for example, grounds answers in internal research with inline citations and source viewers.
- How much time does it save?
- Deutsche Bank analysts report saving 30 to 45 minutes on earnings note templates and up to two hours on research reports and roadshow updates. On the distribution side, Morgan Stanley's global director of research told CNBC that a salesperson needs one tenth of the time to answer the average client inquiry with AskResearchGPT.
- Can summaries go straight to clients?
- Only through the same review as other research and marketing communications. Internal use with citations is the low risk starting point; client facing output needs templates and sign off.
How to cite this page
Blits.ai AI Use Case Library, "AI summaries of investment research and the house view", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/investment-research-summarization. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published