A security master you work with in plain words: agents draft each new bond from its prospectus, cite every term and price every vendor disagreement; two stewards decide
Target zero wrong coupons, calls or day counts reaching the funds, each term cited.
See one case, screen by screen ↓Security Master Data Management is security master software for buy-side reference data teams: it sets up and checks instrument terms from prospectuses and term sheets. An analyst works with seven AI agents in plain words — they draft each new instrument record from its offering document in any language with every term cited to its page, compare it with two vendors, price each disagreement on the firm's position and show the funds and rules it touches. One data steward prepares the record and a different one approves it before it publishes to the golden copy.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Every trade, price, compliance check, NAV and client report reads the security master, yet a reference data team still sets up each new bond by hand from two vendor feeds that often disagree, opening the 300-to-500-page prospectus only when they do. For private placements, local-market bonds and new issues documented in French, German, Italian or Japanese, no vendor has the terms yet, and an analyst keys every field — often with the desk on the phone and a booking cut-off a few hours away.
The errors surface downstream, late, and in someone else's numbers. A wrong day count misstates every coupon and accrual in every fund that holds the bond and turns up as an accrued-interest break with the custodian. A missing call schedule overstates yield and duration, and a wrong minimum denomination lets a model portfolio buy a lot that cannot settle. Under T+1, an instrument that is not set up the day it trades cannot be booked, affirmed or priced.
Estimated split for setting up one complex bond from its offering document, by hand and with the agents; the by-hand total follows practitioner accounts of one to three hours.
Seven agents draft, check, price and publish every instrument; one steward prepares and a second approves.
Friday, October 16, 2026. Quillfield Capital's credit desk has been allocated $220 million of Calderwick Holdings S.A.'s new 6.125% Senior Notes due 2034 across six funds. Allocations must book by 15:00 ET, and the only full document is a 412-page prospectus in French. Here is what happens next, screen by screen, in the working solution.
Lena opens Today in London: 8 setups book today out of 31 open requests, 61 term changes wait for a steward, 8 items wait for a checker, and 0 were published without a second steward this month. The agents' briefing reads like a conversation — 23 documents arrived since midnight, each read in its own language; one memorandum is password protected; 412 term changes came in and 351 were accepted by golden-source rules.
Validator agent: “Calderwick 6.125% 2034 (French prospectus): 4 terms disagree with the vendors. The day count matters most: Vendor A’s 30/360 would overstate the first coupon by $18,458.90 on our $220 million.”
She asks the agents in plain words on the Setups screen. The answer names the two setups at risk at 16:30 London, 48 minutes away — Hanseatic Wind waits only for a checker, Ostrava Energie has two disagreements open on pages 41 and 42 — and says Calderwick books at 15:00 ET with 4 disagreements priced. Then the agents filter the list to today's bookings. The new instrument setup queue holds documents in Czech, French, German, Italian, Japanese, Korean, Norwegian, Portuguese, Spanish and English.
Every box for asking the agents carries the same line: “answers cite their source · the agents propose, a person decides.”
The Intake agent took the French prospectus and Ashbury Lane's two-page English pricing term sheet from the syndicate inbox, checked the ISIN check digit and linked both to Eli Brandt's request. The Setup agent drafted all 142 fields — 131 cited to prospectus pages, 6 to the term sheet, 5 derived (CFI DBFUGR, FISN, sector, country of risk, liquidity bucket) — and drew the bond's whole life on one timeline: 16 coupons, the make-whole window, calls at 103.063, 101.531 and par, the 101 change-of-control put and a sustainability step-up. Her next step is on one panel: settle 7 items, then submit.
Three fields sit below 0.90 confidence — the step-up observation date, the clean-up call threshold and the guarantor LEI. Each starts unticked with a “confirm” chip and its page; the record cannot be submitted until Lena checks the page and ticks it.
The Validator agent compared 31 economic terms with Vendor A and Vendor B: 27 agree, 4 disagree. Vendor A has a 30/360 day count, a 2,000 minimum piece and no step-up; Vendor B rounds the 2030 call price to 101.500. Each row says what it would cost on Quillfield's $220 million: $18,459 on the first coupon, unsettleable lots, up to $2.2 million of income if the step-up is triggered, $68,200 if the bond is called in 2030. Beside it, the agents' work on the record reads as a thread, each step with its plan and sources.
The Validator agent never overwrites a value and never resolves a disagreement itself.
One click on “p. 47” opens the page itself: Article 5.2, “Base de calcul”, the exact number of days elapsed divided by 365, “Exact/365”, with the English beside it and the phrase highlighted. The drawer lists everything else cited on that page — day count, business days, payment calendars — and the source is kept unchanged with its hash. No translator, no hunting through 412 pages.
“Interest will be calculated on the basis of the actual number of days elapsed in the relevant Interest Period, divided by 365 (“Actual/365”).”
The Validator generated every coupon under the prospectus basis and under Vendor A's 30/360. The first coupon on April 22, 2027 is $3,054,109.59 per $100 million under Actual/365 (Fixed) and $3,062,500.00 under 30/360 — $8,390.41 too much, or $18,458.90 on Quillfield's $220 million — and 30/360 misstates the 16 coupons by $73,835.62 on the position over the life of the bond. An accrued-interest check recalculates any settlement date and notional for a trade in the secondary market.
The Downstream agent shows what the record will touch: six funds buying $220 million, from Quillfield Global High Yield Fund at $72 million to the Northgate Teachers' Pension separate account at $15 million, each against its own issuer limit, rating floor or client guidelines; 2 open orders; 14 compliance rules, 13 passing; 3 pricing models; and 7 systems to publish to. The warning: the USD 200,000 minimum piece blocks three wealth model portfolios that trade smaller lots.
The Downstream agent never publishes before a second steward approves, and blocks publishing while a rule breaks.
Lena ticks the three proposals after checking their pages, accepts the four prospectus values with the reason “Offering document governs — new issue” and submits; she cannot approve her own record. At 11:14 Priya Raman, senior data steward, approves version 1.0: “Checked against the source pages”. At 11:15 the Downstream agent publishes to the security master, the order management system, portfolio accounting, the compliance engine, the pricing engine, the risk system and the fund administrator — every acknowledgment back within 40 seconds, hours before the 15:00 ET cut-off.
“Maker: Lena Fischer · Checker: Priya Raman · recorded on the audit trail with the evidence.”
The Challenge agent had drafted CH-2291 to Vendor A (day count, minimum piece and step-up, pages 3, 47 and 52) and CH-2292 to Vendor B (the 2030 call price, page 59). Priya approves and both go out at 11:16 with the pages attached. Every challenge feeds the scorecard: this month Vendor A disagreed with the document on 1.5% of 7,657 compared terms and corrected 19 of 26 challenges in a median 3.2 days; Vendor B on 0.9%, 10 of 12 corrected in 1.8 days.
The Challenge agent never sends without a steward's approval, and never shares one vendor's licensed data with another.
On the audit trail she asks, and the agents answer: Priya Raman approved version 1.0 at 11:14 ET after Lena Fischer prepared it at 11:06; the evidence is the prospectus page behind each value, the test cash flows and the impact. The trail is filtered to SR-26-10418 — every step by the agents and by people, with its reason, source and hash. One click exports an evidence pack for the auditors.
Hash-chained and kept 7 years after an instrument is inactive.
The same new bond, seen by five of the people who carry it — what their day looked like, and what it looks like now.
Seven agents draft, check, price and publish every instrument; one steward prepares and a second approves.
Receives prospectuses, offering circulars, final terms, term sheets and call, redemption and step-up notices from email, the document store and uploads; classifies them, detects the language, reads identifiers, checks ISIN check digits and links each to a request or an instrument.
Drafts the full instrument record from the offering document — 142 fields for a corporate bond — each cited to page and phrase with the original-language clause beside the English, and builds the instrument's life timeline.
Compares every economic term with Vendor A, Vendor B and the listing venue, generates test cash flows and accrued interest under each source's terms, and prices each disagreement on the firm's own position.
Shows the funds, open orders, compliance rules, pricing models and systems a record touches, runs a dry-run publish, and publishes once a second steward approves.
Watches vendor deltas and notices for term changes — about 5,000 a month: coupon resets, calls, step-up triggers, sinking-fund draws, consents and identifier changes — and accepts them by golden-source rule or routes them to a steward.
Drafts vendor challenges with the cited pages, sends them once a steward approves, tracks the reply and the feed, closes each on correction and keeps the vendor scorecard.
Checks the whole security master every night: bonds matured but still active, missing mandatory fields, duplicate identifiers, stale vendor data and records with no source document.
Prepare, approve, publish. The agents propose; a named person decides.
Every work screen has a box to ask the agents about what is in front of you. Answers cite the page they came from and take only safe steps — filter, open, show. Anything that changes the security master stays a person's decision.
Why does the day count differ?
The prospectus fixes the day count on page 47: interest is the actual days divided by 365, “Exact/365” in the French original. Vendor B agrees; Vendor A carries 30/360. On Quillfield’s $220 million, Vendor A’s basis would overstate the first coupon by $18,458.90 and misstate all 16 coupons. Under the golden-source rule for new issues the document governs.
What happens if it is called in 2030?
If Calderwick calls on October 22, 2030, it pays 101.531: $223,368,200 on Quillfield’s $220 million, plus the coupon due that day. Vendor B’s 101.500 would understate that by $68,200. Before October 22, 2029 only the make-whole applies, at Treasury + 50 bp. I have marked the 2030 call on the timeline.
Which funds and rules does it touch?
The record touches 6 funds buying $220 million, 2 open orders, 14 compliance rules and 3 pricing models, and publishes to 7 systems. One rule needs attention: the USD 200,000 minimum piece blocks three wealth model portfolios that trade smaller lots. Every other rule passes after the trade, e.g. Global High Yield at 1.05% of its 3% issuer limit.
Which vendor is right more often?
This month Vendor A disagreed with the document on 1.5% of 7,657 compared terms (118), mostly day counts and call schedules; Vendor B on 0.9% (68), mostly call-price rounding. Vendor B also corrects faster: a median 1.8 days against 3.2. I opened the scorecard.
13 screens from the working solution, on its sample data. Pick one to see it large.
Which source wins, per asset class and field, and every threshold the agents work to are settings. Each change needs a second approver and is on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| New issues — which source wins | Offering document → Vendor B → Vendor A, for 30 days after issue | Precedence of document, Vendor A, Vendor B and listing venue, per asset class and field |
| Auto-fill a field at confidence | 0.95 or higher | Your threshold; a change waits for a second administrator |
| Leave a field for a person below | 0.90 | Your threshold; a change waits for a second administrator |
| Always needs a steward | Coupon, day count, calls, puts, maturity, denomination, identifiers | The economic terms your control policy names |
| Bulk accept | Non-economic only · at most 50 at a time | Your cap; economic changes never go in bulk |
| Escalate a disagreement to the Head of Reference Data | Above $50,000 on Quillfield's position | Your dollar threshold on your own position |
| Setup SLA | Before the booking cut-off · escalate 60 minutes before | Your cut-offs and escalation window |
| Term changes and vendor challenges | Term changes within 4 business hours · chase a challenge after 2 business days | Your service levels |
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 a pilot · “estimated” = our estimate · “typical” = published figures (IHS Markit and Waters Technology study via The TRADE, August 2, 2016 — 12% of tier 2 and 3 asset managers say three-quarters of their data is mastered; almost 60% rank data quality control as their first data priority). SEC T+1 (May 28, 2024) and SEC Rules 204-2 and 31a-2 are named as context. Quillfield Capital, its funds, the issuers, banks and vendors, and the people named on this page are fictional — characters and sample data in the working solution.
Security Master Data Management is security master software for buy-side reference data teams: it sets up and checks instrument terms from prospectuses and term sheets. An analyst works with seven AI agents in plain words — they draft each new instrument record from its offering document in any language with every term cited to its page, compare it with two vendors, price each disagreement on the firm's position and show the funds and rules it touches. One data steward prepares the record and a different one approves it before it publishes to the golden copy.
The security master, sometimes called the security master file, is the golden copy of instrument reference data for everything a firm holds or trades: identifiers, issuer, coupon, day count convention, calls, puts, step-ups, sinking funds, maturity and classifications. The order management system, portfolio accounting, compliance, pricing, risk and the fund administrator all read it, which is why reference data management is a control, not a clerical task: one wrong term reaches every fund that holds the instrument.
It is bond prospectus data extraction with a source behind every value. The Intake agent reads and links the offering document, and the Setup agent drafts the full record from it — 142 fields for a corporate bond — each with its page, the original clause and the English translation side by side. Anything below 0.90 confidence is only proposed: a person checks its page and confirms it before the record can be submitted. In the demo it works from a French prospectus; the setup queue also holds Czech, German, Italian, Japanese, Korean, Norwegian, Portuguese and Spanish documents.
The Validator agent compares every economic term with both vendors, generates test cash flows and accrued interest under each source, and says what the difference costs on your position — $18,458.90 on the first coupon in the demo. A steward accepts the document value, the vendor value or types one, always with a reason, and the Challenge agent sends the vendor the page once a steward approves.
The Change agent reads vendor deltas and notices — about 5,000 a month — and applies your golden-source rules: on the demo day 351 of 412 changes were accepted by rule and 61 went to a steward. Calls, coupon changes, maturity changes and identifier changes always need a steward, and a notice in German or Spanish is shown with its English translation.
Yes. The agents draft, compare and propose; they never publish, approve, overwrite a value or send a vendor challenge on their own, and a field below 0.90 confidence stays a proposal until a person confirms it. The person who prepares a record or an economic change can never approve it, golden-source rule and agent changes need a second approver, and every decision is on a hash-chained audit trail.
Since T+1 settlement in the US on May 28, 2024, an instrument that is not set up the day it trades cannot be booked, affirmed or priced. Setups are ranked by booking cut-off, a setup is escalated 60 minutes before it would miss its cut-off, and in the demo the Calderwick bond publishes at 11:15 ET for a 15:00 ET cut-off.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your golden-source rules, field dictionary, vendor files and a sample of recent offering documents — and it goes live once every quality gate has passed. A guided first-time setup connects the sources, maps the fields, imports the golden-source rules from your current reference data management system, invites the team and runs the agents on last month's 50 new issues to compare with what was approved.
Target 20–30% fewer fails, each fix cited and checked before the market’s cut-off.
1-min demoStandard Settlement Instruction ManagementEvery SSI change proven at the source and called back before a dollar moves.
1-min demoPosition and Cash ReconciliationEvery break explained from its source before the NAV is struck, every fix proved first.
1-min demoWe'll run Security Master Data Management on last month's new issues from your own offering documents and vendor files.