Pre-trade compliance checks that answer “may this fund buy it, and how much?” for every fund, with the clause, before the order
A cited yes, no or how-much for every fund, in under a minute, before the order.
See one case, screen by screen ↓Pre-Trade Compliance Checks answers, before an order is placed, whether each fund or account may buy a security and how much. A portfolio manager asks in plain words; eight agents read the prospectus, SAI, IMA and side letters for every fund the order might touch, compute headroom from live holdings, and return one verdict per fund — yes, yes up to a limit, no, or compliance to decide — with the deciding clause quoted. Every answer is read twice and kept as a pre-trade record sent with the order.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Every fund and account has written investment restrictions: the prospectus and SAI of a mutual fund, the supplement of a UCITS, the investment management agreement and investment policy statement of an institutional client, and the side letters that add exclusions — no issuer with more than 10% of revenue from thermal coal, no tobacco, no name on a client’s sanctions list. Before a portfolio manager places an order, someone has to answer “may this fund buy it, and how much?” for every fund the order might go into. The order management system’s rules engine answers only for the limits someone has already coded.
Everything else goes by email or chat to an investment compliance officer, who opens the side letter, the IMA, the exclusion list, the issuer’s 10-K and this morning’s holdings, answers by email, and rarely records the answer against the order. That is where the failures live: a trade placed on a misremembered limit, a side-letter exclusion that was never coded, a data vendor’s stale revenue split letting an excluded issuer through, a split rating read two ways, a soft override with no second signature — and an answer nobody can find when the examiner asks.
Estimated split for one question the rules engine does not hold, answered by hand and with the solution. The 20–40 minutes per question by hand is our assumption, to be measured in a pilot.
Eight agents read every mandate twice and compute headroom; the PM sends, compliance rules and signs.
Quillfield Capital, a global asset manager with 412 funds and accounts under written restrictions, on Tuesday, October 20, 2026. A portfolio manager wants $20M of a new, split-rated bond across five credit funds — each with different documents. Here is what happens next, screen by screen, in the working solution.
Today shows 64 questions answered so far, a median of 41 seconds to an answer, 4 with compliance now, and 37 orders stopped before entry this month with 0 breaches on answered trades. At the top of Needs you: Kestrel Mining 6.875% 2030, a new issue on her watch list. She planned $20.0M across her credit funds.
“Ask before you call Tom: five funds, five sets of documents.”
She types “Can we buy $20M of Kestrel Mining 2030 notes across our credit funds?” The Question agent resolves it to CUSIP 49238KAD7, rated BBB− / Ba1 / BB+, $20.0M, five Global Credit accounts, and plans 23 checks. The Restriction Reader reads 14 documents — 3 prospectuses, 1 SAI, 1 UCITS supplement, 2 IMAs, 1 side letter and 6 board and policy documents — and finds 31 clauses.
Every step is shown as it runs, with its plan, sources and confidence.
Core Bond Fund: Yes, $621.0M of headroom under its diversified-fund limit of 5% of total assets in one issuer (prospectus p. 12). Global High Yield Fund, an Irish UCITS: Yes, up to $6.5M. Harrowgate Mutual Life’s general account: Yes, 2.1% ($67.2M) left in its 8% below-investment-grade bucket. Allenmoor County ERS: No. Short Duration Income Fund: Compliance to decide.
“Second Reader agrees” on four tiles; on the fifth, “Readers disagree · to Rohan”.
Allenmoor’s Side Letter No. 2 §3(c) excludes any issuer with more than 10% of annual revenue from thermal coal. The data vendor still shows 8.9%, from fiscal 2023. The agent reads Kestrel’s fiscal 2025 10-K, page 31 — thermal coal $612M of $4,310M revenue, 14.2% — so the answer is No, and the vendor value is flagged to the data owner. The Second Reader confirms the passage word for word.
On the tile: “Client limit: cannot be overridden; only the client can waive it, in writing.”
Show working lays out every line with its source: NAV of $406.3M from the fund administrator at 6:00 a.m., the 5% issuer limit in supplement 4(a) on page 22 ($20.3M), $13.8M of Kestrel 5.25% 2028 notes already held, $0.0M of fills today from the order management system. Headroom: $6.5M. The UCITS 10/40 rule passes too: holdings above 5% total 31.2% of 40%.
Both readers, separately: “Yes, up to $6.5M” — the Restriction Reader on Model A, the Second Reader on Model B.
Under the tiles she asks “Why is Allenmoor a no?” and gets the clause, the filing figure and the IMA article (Article 12) that says only Allenmoor can waive it. The Allocation agent proposes $8.0M Core Bond, $6.5M Global High Yield and $5.5M Harrowgate — $20.0M — with nothing to Allenmoor, and nothing to Short Duration until it is ruled. She can edit the split; every change is checked against headroom.
“Answer valid until 4:00 p.m. New York, or until holdings move by more than 0.25% of NAV.”
The Restriction Reader read prospectus page 9 — rated investment grade by at least one nationally recognized statistical rating organization — and said Yes, $115.5M. The Second Reader read SAI page 47 — the fund applies the lowest rating — so Kestrel’s Ba1 is below investment grade, and the 10% bucket at 9.6% leaves $9.2M. The screen lays out the firm’s hierarchy for US funds and shows the same question was ruled the same way three times in 2026.
“Neither reading counts until you rule. The PM can send the other funds meanwhile.”
Rohan picks DH-2, “Document hierarchy: the more specific document governs”: the SAI’s split-rating policy was adopted by the Board on March 12, 2024 and supersedes the general description in the prospectus. Short Duration becomes Yes, up to $9.2M. The ruling, with his note, updates the tile, the record and the audit trail; it never edits either reading.
The Restriction Reader recommended DH-2, and said so plainly: “I cannot rule; you decide.”
Elena keeps Short Duration out of this order and sends $20.0M to Tom as block OMS-7781203, with pre-trade record PTR-261020-0064 attached as its compliance note. The rules engine re-checks every fund at order entry: pass for 3 funds. Tom sees the record from question to order, accepts the order against it as the second key, and cannot change a verdict.
“Two keys on every order: you send it; Tom checks it against this record and accepts it before working it.”
The audit trail holds the question at 9:12:04 a.m., the Question agent’s plan, the headroom, the 10-K reading, the Second Reader’s disagreement, the record written at 9:12:46, Rohan’s ruling and the order sent at 9:21. The hash chain was verified at 9:00 a.m., records are kept seven years, and an evidence pack can be built for an SEC examination under Rule 38a-1 and Rule 206(4)-7, or for a client’s annual compliance certification.
“Who did what, when, why and on which evidence: people and agents alike.”
The same pre-trade question, seen by four of the people who carry it — what their day looked like, and what it looks like now.
Eight agents read every mandate twice and compute headroom; the PM sends, compliance rules and signs.
Turns the question into a security, a size and the accounts it may go into — “our credit funds” resolved from the PM’s book — and plans the checks.
Finds every clause that applies in the prospectus, SAI, supplement, IMA, guidelines, IPS and side letters, in their original language, quotes it with document, section and page, and classifies it as regulatory, prospectus, client or house.
Reaches its own verdict for every account from the same documents, on a different model, and compares only afterwards.
Computes today’s exposure and headroom in dollars from the 6:00 a.m. holdings plus today’s fills and current prices, showing each step with its source.
Where an exclusion depends on a fact about the issuer — thermal coal share of revenue, weapons, a sanctioned owner — reads the issuer’s latest filing in its own language, cites the page and compares it with the data vendor.
Proposes how the order splits across the funds answered yes, within headroom, by the PM’s model weights.
When a clause is not in the rules engine or has changed, proposes the coded rule, back-tests it on today’s holdings and sends it for two compliance signatures.
Writes and hashes every answer as a pre-trade record, attaches it to the order in the order management system and builds the exports for examiners, auditors and clients.
Send, rule, countersign, execute. The agents propose; a named person decides.
Under every answer is a conversation with the agents. Ask why, ask what-if, or tell them what to do — every reply cites its clause or holdings line, and nothing changes without a person.
Why is Global High Yield capped?
Its supplement allows 5% of NAV per issuer (4(a), p. 22). NAV is $406.3M, so the limit is $20.3M. The fund already holds $13.8M of Kestrel 5.25% 2028 notes (3.4%), so $6.5M is left. UCITS would allow 10%, but the fund’s own 5% is stricter, so it governs.
Why is Allenmoor a no?
Side Letter No. 2 §3(c) excludes issuers with more than 10% of revenue from thermal coal, taken from the most recent annual report. Kestrel’s fiscal 2025 10-K shows $612M of $4,310M = 14.2% (p. 31). The data vendor still shows 8.9% from fiscal 2023, so I used the filing and flagged the vendor value. This is a client limit: it cannot be overridden here; only Allenmoor can waive it, in writing (IMA Article 12).
What if we buy $10M for High Yield?
Not saved, a what-if only: $10.0M would take Kestrel to $23.8M, 5.86% of NAV, over the 5% limit by $3.5M. The rules engine would block it at order entry as a hard limit. The most High Yield can take today is $6.5M.
How did you split the $20M?
From your model weights for this idea (Core Bond 32%, High Yield 36%, Harrowgate 22%, Allenmoor 10%). I capped High Yield at its $6.5M headroom and removed Allenmoor (No). The $2.7M left went to the two funds with room, in proportion to weight: Core Bond $8.0M, Harrowgate $5.5M. Edit it if you prefer another split; I will check every change against headroom.
13 screens from the working solution, on its sample data. Pick one to see it large.
The thresholds, the document hierarchy and who may do what are settings. A change is a new version, and it goes live only when a second administrator approves it.
| Setting | Default | Choose from |
|---|---|---|
| Confidence floor | 0.90 — below it, the question goes to compliance | Any value from 0.50 to 0.99, proposed as a new version |
| When the readers disagree | Always to compliance | Always on; never resolved by an agent |
| Document hierarchy · US funds | 1940 Act and SEC rules › prospectus and SAI, the more specific governing › board resolutions › house manual | Set per vehicle |
| Document hierarchy · separate accounts | Law › side letter › IMA and guidelines › IPS › house manual | Set per vehicle; UCITS: UCITS rules › prospectus and supplement, the stricter governing |
| What can be overridden | House rules only | Regulatory, prospectus and client limits never |
| Answer validity | Until 4:00 p.m. New York, or until holdings move by more than 0.25% of NAV | Open answers re-checked automatically when they expire |
| Data vendor against the issuer’s filing | 2 percentage points | Above it the filing value is used and the vendor value flagged to the data owner |
| Review times | Disagreement 30 min · override 60 min · new rules 3 days | Past the time, to the Head of Investment Compliance |
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 (about 1,280 questions a month × 20–40 minutes each, our assumption) · “typical” = published figures (Sidley, fiscal 2025 review of SEC enforcement against investment advisers — $4 million penalty, October 21, 2024; SEC OCIE risk alert of November 7, 2019, via Dechert, on compliance programs lacking policies to prevent violation of investment limitations and guidelines). People, firms, funds, clients, issuers, identifiers and record numbers named on this page are fictional — characters and sample data in the working solution.
Pre-Trade Compliance Checks answers, before an order is placed, whether each fund or account may buy a security and how much. A portfolio manager asks in plain words; eight agents read the prospectus, SAI, IMA and side letters for every fund the order might touch, compute headroom from live holdings, and return one verdict per fund — yes, yes up to a limit, no, or compliance to decide — with the deciding clause quoted. Every answer is read twice and kept as a pre-trade record sent with the order.
The rules engine in your order management system or investment compliance software keeps blocking what it already holds; the agents answer the questions it cannot, such as a side-letter exclusion, a split-rating policy in the SAI or a new client’s guidelines. The pre-trade record goes to the order management system as the order’s compliance note, and the rules engine re-checks every fund at order entry. When a question surfaces a clause the engine does not hold, the Rule Drafter proposes the coded rule, back-tests it and sends it for two compliance signatures before it syncs.
Pre-trade compliance tests a proposed trade against a fund’s investment restrictions before the order is placed; post-trade compliance, the investment guideline monitoring that usually runs in an overnight batch, finds breaches afterwards, when they must be cured. This solution works before the order: it answers for every fund the trade might go into, from the documents and this morning’s holdings, so the order is right the first time.
The Restriction Reader reads each document in its original language and quotes the clause with its page; the original sits beside the English. In the working solution a German special fund mandate’s 38 restrictions were read this way, 36 agreed by both readers and 2 sent for a ruling, each with a proposed coded rule and a back-test.
Where an exclusion depends on a fact about the issuer, the Issuer Evidence agent reads the issuer’s latest filing and compares it with the data vendor. In the sample case the vendor showed 8.9% of revenue from thermal coal for fiscal 2023; the fiscal 2025 10-K showed 14.2%, so the client’s 10% exclusion applied and the vendor value was flagged to the data owner.
Yes. The agents read, compute, propose and cite; they never decide, send, override or trade. When the Restriction Reader and the Second Reader disagree, or confidence is below 0.90, the question goes to an investment compliance officer with both readings and the document hierarchy, and only a ruling with a reason code settles it.
A portfolio manager can request an override, and only of a house limit, with a reason code, a note, an end date and evidence. A compliance officer gives the first key and the Head of Investment Compliance the second; the requester can never approve it. Client, prospectus and regulatory limits cannot be overridden at all.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your fund documents, IMAs and side letters, your compliance manual and a sample of past restriction questions — and it goes live once every quality gate has passed. We will walk you through it on your own mandates first.
Every model change reaches every account the same day, within each client's own rules.
1-min demoPerformance Attribution AnalysisWhy a portfolio lagged, answered the same day, each sentence tied to its attribution cell.
1-min demoInvestment Policy Statement MonitoringEvery policy limit checked daily, and every exception tied to its clause and its holdings.
1-min demoWe'll run Pre-Trade Compliance Checks on your own prospectuses, IMAs and side letters, and a sample of past restriction questions.