A research management system where every thesis keeps its reasons, each one cited to its source
Every thesis keeps its reasons when analysts leave, each one cited to its source.
See one case, screen by screen ↓A research management system (RMS) is the investment research management software where a buy-side research team keeps its notes, meetings, models, emails and expert calls so the firm's reasoning on each name can be found again. Research Management System files every item to the thesis version it supported, so a portfolio manager can ask why any call was made and get the original reasoning, every sentence cited, even after the analyst has left. Seven agents file, screen, track, answer and verify; compliance decides every held call with two people, and the Director of Research approves every rating change.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
A research team's reasoning is spread across analysts' notebooks, inboxes, Excel models revised week by week and expert-network call transcripts. Most research notes software tags a note to a ticker, but it does not know which note, call or model version changed the rating. So when a portfolio manager asks “why did we cut this to Hold?” before committee, the search returns a pile of documents, and the real answer sits with the analyst who made the call — until the analyst leaves.
Expert calls are a regulatory exposure as well as evidence. The SEC Division of Examinations' risk alert of April 26, 2022 lists not logging expert-network calls and not reviewing call notes among the deficiencies it saw, and many teams still sample transcripts by hand, so a call with possible MNPI can reach the shared library before anyone reads it. And research is an investment adviser's books and records under Rule 204-2, kept five years or more — usually in three systems, so an examiner's request for “all research supporting the May 2025 decision” means days of pulling files from each one.
Estimated split for one name and one rating change, collected by hand across notebooks, the email archive and shared drives, and with the solution. The solution's own number in the demo: the exam pack is built in seconds, after a person reviews the scope.
Seven agents file, screen, track, answer and verify; compliance decides every held call with two people, and the Director of Research approves every rating change.
Quillfield Capital, a global asset manager, holds $328M of Tessaro Health across three portfolios. Next Thursday, October 22, the committee hears an analyst's draft upgrade. The analyst who cut the stock to Hold in May 2025 left in August that year. Here is what happens, screen by screen, in the working solution.
Celia runs the Global Health Care Equity Fund, 62 names. Her home screen opens on an ask box over 538,214 items since 2018 and what the agents did overnight: the Archivist filed 1,142 of 1,186 new items, the Compliance agent screened all 41 new expert-call transcripts and held 6, and the Thesis Tracker blocked Tessaro Health v5. Top of “Needs you”: “Before Thursday's committee: why did we cut Tessaro?”
“Answers only from Quillfield's own records, within your entitlements.”
Celia asks “Why did we cut Tessaro Health to Hold in May 2025?” on the Tessaro river: five thesis versions, 214 items. The Historian plans, reads the river and finds version 3, Cut to Hold on May 14, 2025; the Verifier checks all six citations. The answer is four sentences, each cited — the memo, an expert call cleared with one passage redacted, the supplier's filing, model cell F44 and the next day's trim from 4.6% to 3.1%.
“4 of 4 sentences cited · 6 citations · Verifier 0.94 · read 214 items · answered in 38 s.”
One click opens Kaneda Medical Components' earnings summary on page 3 of 18. The Japanese original sits beside the English, and the same passage is highlighted in both: shipments of sensor film to its major US customer fell 31% year on year, owing to lower yields. The English is a machine translation the Verifier checked against the original, passage by passage.
“Figures are taken from the original.”
The Historian answers from Priya Raman's draft version 5, Upgrade to Buy at $82: film yields back above 90% and Kaneda's shipments up 18%. It also says the draft cites an expert call from October 13 that it was not allowed to read, because the call is on compliance hold. The Thesis Tracker has blocked version 5 from going to the Director of Research. Celia saves the answer, with its citations, to the committee pack.
“1 item not read: EC-26-1013 · on compliance hold H-26-0388 · escalated to Naomi Feld.”
From firm email (512), notebook sync (236), the model store (203), calendar and CRM (148), document management (46) and three expert-network platforms (41), the Archivist read and translated 57 items not in English, matched each to a name and a thesis version, and sent all 41 expert-call transcripts to compliance screening first. It filed 1,142 on its own at confidence 0.90 or above and set aside 9 duplicates. Platform B failed to sync at 05:40; the screen says what that means for the record.
“Until then, any call taken on platform B since 05:40 is not in the record.”
Priya types “Confirm everything at 0.85 or above.” The Archivist finds 7 waiting items at that confidence, and says what it left out and why: the Aldvik model v5.3 at 0.87 belongs to Marta Lindqvist, a leaver whose successor is not yet approved. Nothing is confirmed until Priya presses “Confirm them”, and she can still refile any item later with a reason.
The Archivist never edits an author's text, and files on its own only at confidence 0.90 or above.
In 12 months the Compliance agent screened 2,140 calls and held 63; 49 have since been released, 33 with redaction, 5 kept restricted, and 9 are open. H-26-0388, a call with a current regional sales manager at a US diabetes-supplies distributor, is high risk at 0.91, blocks Tessaro v5 and is 1 hour past its two-business-day SLA, so it went to the CCO.
“Held expert calls are visible to compliance only.”
The agent flagged unannounced Q3 order volumes “up about 40%” (R-03, 0.94), a current employee of the issuer's distributor (R-07, 0.82) and unannounced contract pricing (R-05, 0.77), and recommended keeping the call restricted. Owen proposes keeping the call restricted and putting TSRO on the watch list in the order management system until the October 29 results (R3), and reporting the expert to the network (R4). When he tries to approve his own proposal, the screen refuses.
“Not allowed. You proposed this decision; a different person must approve it (segregation of duties). The attempt is logged.”
Naomi reads the same transcript, flags and recommendation, and approves. The decision is recorded with both names and the reason code, TSRO goes on the watch list until October 29, and the Thesis Tracker tells Priya to rebuild version 5 without the call.
“Decided: keep restricted. Proposed by Owen Pryce, approved by Naomi Feld at 09:29 (R3).”
Marta Lindqvist leaves on October 30: 27 names, 4,812 items. The Handover agent built her pack from the HR notice, linked 4,771 items to a thesis version and lists the 41 it could not — handwritten notebook pages, emails naming more than one issuer, meetings with no notes yet — for Marta to confirm. It proposes Tobias Engel for 19 names and Hannah Cole for 8; nothing moves until Robert Hale approves, and on her last day her access ends while her records stay.
“4,771 of 4,812 items linked to a thesis version · 99.1%.”
The same thesis river, seen by five of the people who carry it — what their week looked like, and what it looks like now.
Seven agents file, screen, track, answer and verify; compliance decides every held call with two people, and the Director of Research approves every rating change.
Captures notes, meetings, emails, model versions, filings and expert calls, and files each one to its ticker, the thesis version it supported and its retention class. Reads documents in their original language and keeps page anchors.
Screens every expert-call transcript for possible MNPI against the expert's disclosed employment, the restricted and watch lists and the issuer's results calendar, holds risky calls and recommends a decision with a reason code and the cited passages.
Detects thesis versions from memos and the ratings feed, keeps each name's river current, flags theses past their 120-day review date and blocks a version that cites a held or restricted item.
Answers “why did we…” questions over the firm's own research history: it plans, reads the river and lights the stretch that answers.
Re-reads every cited passage, checks numbers against model cells and filings, checks translations against the original and removes any sentence it cannot support.
When the HR system reports a resignation, builds a handover pack per name, imports the leaver's notebook sections, lists the items it could not link for the leaver to confirm and proposes successors by coverage overlap.
Applies retention classes, legal holds and data residency, proposes disposal batches and builds exam packs with every item, version, access event and decision, hashed.
Approves every release. The agents propose; a named person decides.
Ask on the thesis river or on any working screen — coverage, capture, screening, records, handover, audit. Every answer cites its sources, shows the agent's plan and the Verifier's check, and any action waits for your yes.
Why did we cut Tessaro Health to Hold in May 2025?
We cut Tessaro Health from Buy to Hold on May 14, 2025 and lowered the target from $96 to $71, because yields on its main supplier's second-generation sensor-film line had fallen to about 70%. An expert call on May 9 described yields “in the low seventies most of the first quarter”, and the supplier, Kaneda Medical Components, reported shipments to its major US customer down 31% year on year. Model v3.2 cut FY2026 gross margin from 63.5% to 58.9% in cell F44. Daniel Okafor wrote the memo and left on August 22, 2025; his handover pack passed Tessaro to Priya Raman, who kept Hold on November 4, 2025.
What would take us back to Buy?
Priya Raman's draft version 5 (Oct 15, 2026) proposes Buy, $82: management said film yields were back above 90% for the full quarter, and Kaneda's shipments of sensor film to its major US customer rose 18% year on year. The draft also cites an expert call from October 13 that I was not allowed to read: it is on compliance hold, so I did not use it, and version 5 cannot be published until compliance decides.
Which holds should I decide first?
H-26-0388 first: it is 1 h past its two-day SLA, escalated to Naomi Feld, and it blocks Tessaro v5. Then H-26-0385 (Brightwell, due at 15:40 today) and H-26-0393 (Lianhe Circuit, high risk 0.87: a current employee of a customer).
Who has read the held Tessaro call?
Nobody outside compliance. EC-26-1013 was opened four times: twice by Owen Pryce, once by Naomi Feld, and once by the Compliance agent when it screened the call. The Historian skipped it twice and said so in its answers.
13 screens from the working solution, on its sample data. Pick one to see it large.
The thresholds the agents work to are settings: a research operations admin proposes a change, it is saved as a new version, and changes to compliance rules need the CCO.
| Setting | Default | Choose from |
|---|---|---|
| Auto-file confidence | 0.90 | Below this, the Archivist asks the analyst |
| Hold an expert call at any flag above | 0.50 | Compliance agent, rule set v2.4 · changes need the CCO |
| Thesis review date | 120 days | The Thesis Tracker flags names with no review since |
| Hold SLA | 2 business days | Then the hold escalates to the CCO |
| Results window | 15 days | Calls this close to an issuer's results are flagged |
| Verifier minimum support | 0.80 | Sentences below this are removed or sent back |
| Leaver items linked before the last day | 100% | The Handover agent reminds the leaver 3 days and 1 day before |
| Research records retention | 7 years | Rule 204-2 · MiFID II · firm policy; legal holds override every class |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (Advisers Act Rule 204-2 — research records kept five years, the first two in an appropriate office; Substantive Research, September 2025 — research budgets flat, −0.2%, at the 44 largest managers in H1 2025; SEC Division of Examinations risk alert of April 26, 2022 on MNPI and expert-network calls; MiFID II Article 16(6)–(7)). People, firms, issuers and funds named on this page are fictional — characters and sample data in the working solution.
A research management system (RMS) is the investment research management software where a buy-side research team keeps its notes, meetings, models, emails and expert calls so the firm's reasoning on each name can be found again. Research Management System files every item to the thesis version it supported, so a portfolio manager can ask why any call was made and get the original reasoning, every sentence cited, even after the analyst has left. Seven agents file, screen, track, answer and verify; compliance decides every held call with two people, and the Director of Research approves every rating change.
It links each note, meeting, email, model version and expert call to the thesis version it supported — the memo that cut the rating, the model cell that moved, the call that started it. That is what lets the thesis river answer “why did we…” with the stretch that answers lit, instead of returning a pile of documents.
The Compliance agent reads every transcript before it reaches the shared archive, against the expert's disclosed employment, the restricted and watch lists and the issuer's results calendar. Any flag at 0.50 or above holds the call. A compliance analyst proposes a decision with a reason code, the CCO approves, and a thesis version that cites a held call cannot be published.
Rule 204-2 requires five years, the first two in an appropriate office; MiFID II asks for five, or seven when the authority asks. In the working solution, firm policy keeps research records 7 years and the audit trail 10. The Records Keeper applies the classes, honors legal holds, never deletes anything and proposes disposals that need the records manager and the CCO.
When the HR system reports a resignation, the Handover agent builds a pack per name, links the leaver's items to thesis versions and lists the ones it could not link for the leaver to confirm before the last day. It proposes successors, the Director of Research approves, and on the last day access ends while the records stay.
No. The agents propose and explain; people decide. The Compliance agent never releases a held call, the Thesis Tracker never publishes a rating change, the Handover agent never reassigns coverage and the Records Keeper never deletes. Hold decisions, thesis versions, handovers, disposals and legal holds each need two different people, and any action an agent suggests waits for a yes.
Yes. The Archivist reads documents in their original language and keeps page anchors — the demo's overnight run includes Japanese, German, Swedish, Norwegian and Chinese items. The source view shows the original page beside a machine translation the Verifier checked passage by passage, and figures are taken from the original.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your research and compliance policies, retention schedule and a sample of past notes, models and expert-call transcripts — and it goes live once every quality gate has passed. The working solution's own first-time setup runs seven steps, from connecting identity and capture sources to backfilling the history and building each name's river.
Every call read the day it reports, every claim tied to its transcript line.
1-min demoEquity Research AutomationEvery result read, modeled and noted before the open, every figure tied to its page.
1-min demoFinancial Data ExtractionEvery covered name's model refreshed the day it files, every number tied to its page.
1-min demoWe'll run Research Management System on a sample of your own notes, models and expert-call transcripts.