The broker vote as a conversation with agents: every ballot pre-filled from evidence, every high score explained, every research dollar checked twice before the COO signs
Every research dollar tied to its vote and evidence, approved in days, not weeks.
See one case, screen by screen ↓Broker Vote is broker vote software for asset managers. A broker vote is the twice-yearly review in which analysts and portfolio managers score the brokers and research providers they used, and with the desk's broker scorecard it sets the approved broker list and the research budget. In Broker Vote, seven agents pre-fill each ballot from logged research use, question scores that rest on thin evidence, score execution and counterparty credit separately and turn votes into dollars, while voters decide their own scores and the COO approves the list and the budget.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
The broker vote decides two things a buy-side firm cannot get wrong: which brokers it may trade with, and how its research budget is split. Twice a year analysts and portfolio managers score the brokers and research providers they used, the trading desk scores execution, and counterparty credit risk weighs in. In the UK, paying for research jointly with execution is allowed again only with a research budget, a written policy, ongoing research valuation, a fair allocation of costs across clients, written provider agreements and disclosure (FCA PS24/9, extended to fund managers by PS25/4). In the US, the Section 28(e) safe harbor covers research paid for with client commissions only when the adviser judges in good faith that the commissions are reasonable for the research received.
Yet at most large managers the broker review process is still built on memory. Voters score 60 brokers from what they remember, broker service statements go unreconciled, execution complaints leak into research scores, corporate access gets paid from client money, and a $14 million research budget is approved with no way to show why one broker got $1.46 million. A broker that should go stays on the approved broker list until someone writes an email.
Estimated split of Research Operations' working days in one half-year broker vote cycle, by hand and with the solution.
Seven agents pre-fill, question and re-check every vote; voters decide, the COO approves the list and budget.
Quillfield Capital's H2 2026 broker vote covers research use from April 1 to September 30, 2026: 124 voters in six teams, 18 traders on the desk scorecard and 60 brokers and research providers on the ballot. The voting window closed on October 21. It is Thursday, October 22, 8:40 a.m. in New York. Here is what happens next, screen by screen, as the people who run, cast and sign the vote work with seven agents.
Nadia's Today screen shows 121 of 124 ballots in, 10 of 41 policy questions open, and a provisional research budget of $14.2M across 31 brokers, down 3.1% on H1. Under Needs you, each item has one next step: Leah Brandt's ballot is due at 12:00 p.m., 112 interactions are unmatched, and 7 broker statements differ from the logs by more than 10%. Beside it the Ballot agent has posted the day's plan, each step with its owner and the check it must pass before it hands on, and a box where Nadia can ask the agents or tell them what to do, such as “What is left before the COO can approve?”
“Agents never score, approve, remove or pay.”
The Evidence agent read 21,486 interactions: 6,904 from the CRM, 2,618 from calendar and email, 10,622 reports opened in 31 provider portals and 1,342 model downloads. It matched 20,972 on the first pass and 402 more from aliases such as “PELLAM CO LDN”; 112 under 0.92 confidence wait for Nadia. It also reconciled 34 broker service statements: Northgate Securities claims 512 analyst calls, 375 are logged, and 109 of the claimed calls name voters who do not cover the stock. The query is drafted for Nadia to send.
Under 0.92 confidence the Evidence agent never matches on its own; duplicates are marked, never deleted.
Leah covers European industrials from London and was extended to noon after a site visit. Her ballot lists her 14 brokers down the side and the four services across the top, weighted 30% written research, 30% analyst calls, 20% models and 20% corporate access. The Ballot agent filled it from 118 logged interactions: Pellam & Co. shows 9 reports read, 6 calls, 3 models used and 4 meetings. She clicks “6 calls” and sees each one with Helena Voss, from April 17 to September 18, with its CRM record number.
The Ballot agent never enters or suggests a score; it shows the count and the records behind it.
Leah scores Aldgate Securities 9 for analyst calls, but only one call is logged. The Policy agent checks the score against Research Valuation Policy v3.1 as it is entered and asks why: §4.2 needs a reason for a score of 8 or more on fewer than 3 logged interactions. It has already found a candidate — her August 27 call with Tobias Wren on Varden Rail, which her September 2 note “Varden Rail: upgrade to Overweight” cites — and offers to keep the 9 with this reason, change the score, show the call, or give a different reason. She can also ask, in plain words, “What happens if I lower Aldgate to 6?”
“You scored Aldgate Securities 9 for analyst calls on 1 logged call. Policy §4.2 asks for a reason … I found one candidate.”
Leah keeps her score: “Keep 9: the Aug 27 call drove my Sep 2 upgrade.” The Policy agent records it with the reason code single high-value interaction, attaches the call and the note, and Priya Raman sees it in her review. Nothing else is open, so Leah submits: 122 of 124 ballots are in, with 2 voters excused. Against a 6, her 9 is worth about $8,530 to Aldgate.
Had she given a reason that names no interaction or decision, her score would stand unchanged and the question would go to Compliance with her words attached.
The Policy agent checked 5,206 scores on 121 ballots and raised 41 questions: 19 high scores on weak evidence, 8 declared conflicts, 6 scores far from the team, 4 research comments that mention execution, 3 brokers not on the research list and 1 low score on heavy use. Five answers with a reason and linked evidence wait for Priya, who can accept all five, take them one by one, return one to the voter or send it to Compliance. Declared conflicts go to Grace Liu in Compliance.
“I never accept my own questions; I escalate when an answer does not meet the rule.”
The Execution agent scored 24 counterparties from three quarters of cost data, settlement fails and 18 trader surveys. Calloway Markets is above expected cost three quarters running — +3.1, +3.9 and +4.6 bp — with 11 settlement fails, 1.9% against a desk average of 0.6%, and the lowest trader score, 3.1 of 5. Sam proposed its removal on October 21 with the reason code execution quality. Tidewater Securities and Larchfield FX go on watch, and their research money is untouched.
Desk and credit scores set the approved broker list and stay out of research budgets; the Policy agent flags any research vote that mentions execution (§2.3).
The Credit agent read Hanzei Financial Group's September 26 notice revising its earnings forecast, in Japanese, and set the English beside it: an expected interim loss of ¥18.4 billion and a consolidated total capital ratio of about 9.6% at the end of the second quarter, against 14.8% on March 31, 2026, both cited to page 2. The parent was downgraded to BB+ on September 29, below Quillfield's BBB− counterparty floor and the 10.5% capital minimum. Japan orders moved to Sendai Hoshino Securities and Vellmar Bank Markets on October 20, with $0 unsettled. Owen proposed the removal.
The Credit agent flags against the floor and builds the case; it never proposes a removal itself. The COO decides.
The Budget agent turned the ballots into $14,200,000 across 31 brokers: voter weights of 1.5 for portfolio managers and 1.0 for analysts, the service weights, then each broker's share of points. Brightwater Research is capped at its $385,000 agreement price, though its votes implied $412,000; Greyfriars Research and Penrose Analytics fall below the $25,000 floor and are not paid. Research of $12.35M is funded $4.6M from the client-funded UK research account and $7.75M from Quillfield's P&L; all $1.85M of corporate access comes from the P&L. The Policy agent then recomputed all 31 lines on its own: 31 of 31 match.
“Second check: the Policy agent recomputes every line independently; any difference over $1 stops the pack.”
The Pack & Notice agent assembled the approval pack: the research budget of $14,200,000 across 31 brokers, the removals of Calloway Markets and Hanzei Securities with their evidence, two brokers kept on watch, and the vote with all 41 policy questions resolved. Nadia prepared it on October 22, Priya reviewed it on October 26 and Grace countersigned on October 27. Daniel approves as proposed. Then people take the next steps: Priya releases letters to 58 brokers, Nadia sends the payment instructions for November 16, and Sam confirms approved list v2026.2 for the order management system.
“One person per step · agents never sign.” The preparer can never approve, and the COO approves but cannot edit.
The same H2 2026 broker vote, seen by five of the people who run it, cast it, check it and sign it — what the six weeks looked like, and what they look like now.
Seven agents pre-fill, question and re-check every vote; voters decide, the COO approves the list and budget.
Builds the usage ledger from the CRM, calendar and email, research portal readership and model download logs, with analyst notes from the research management system cited as reasons. It matches aliases to brokers and voters, marks duplicates and reconciles broker service statements against the logs.
Pre-fills each voter's ballot with evidence counts by service and the records behind each count, and sends reminders and escalations on the clock.
Checks every score and comment against the research valuation policy as it is entered, asks the voter one plain question, proposes the evidence it found and escalates; then recomputes the whole research budget independently, by a separate calculation from the Budget agent's.
Builds the desk's broker scorecard from transaction cost analysis, settlement fails, trade errors and trader scores, and assembles the evidence for a watch or a removal.
Watches each counterparty's ratings, credit spreads, capital and filings, including filings in the issuer's own language, and flags any that fall against the credit floor.
Turns votes into dollars with voter and service weights, price caps from provider agreements, the floor, the funding source and the allocation across funds.
Assembles the approval pack, broker feedback letters, client research-cost disclosures, Form ADV input and the examiner export.
Score, review, countersign, approve. The agents propose; a named person decides.
Every screen of the broker vote has a box to ask the agents or tell them what to do, and an assistant follows you across the solution. Answers cite their sources, and the agents can open the screen or start the next step for you — never a decision.
What happens if I lower Aldgate to 6?
Aldgate would get about $8,530 less: 3 points × $2,930, less its small pro-rata share of the money that moves back across all 31 brokers, and the Policy agent's question would close because §4.2 applies only to scores of 8 or more.
Why did Pellam & Co. go up $180,000?
Its votes rose: written research averaged 8.1 (from 7.4), two new models were used by 31 voters, and logged calls rose to 495 from 441. The allocation is $1,460,000, from $1,280,000.
Show the evidence for removing Hanzei Securities
Hanzei Financial Group's Japanese notice of Sep 26 expects a ¥18.4 billion interim loss and a 9.6% capital ratio, down from 14.8%; it was downgraded to BB+ on Sep 29, below the BBB− floor. Owen Hart proposed removal; the COO decides.
Prove corporate access stays off client money
The UK research account carries $4.6M of research only; all $1.85M of corporate access is funded from Quillfield P&L. I check it on every run.
13 screens from the working solution, on its sample data. Pick one to see it large.
The weights, rules and thresholds the agents apply are settings, versioned and changed only with a second person's approval: the Director of Research proposes, Compliance approves, and every agent change is evaluated on the last two cycles first.
| Setting | Default | Choose from |
|---|---|---|
| Service weights | Written research 30% · analyst calls 30% · models 20% · corporate access 20% | Any split; try new weights as a scenario before the Director proposes it |
| Voter weights | Portfolio manager 1.5 · analyst 1.0 | By role, with team weights by research-intensive assets |
| High score on weak evidence (§4.2) | 8 or more on fewer than 3 logged interactions needs a reason | Also §4.3 low score on heavy use · §4.5 2.5 points from the team median goes to the Director |
| Declared conflicts · execution in research (§5.1 · §2.3) | Conflicts go to Compliance · no execution in research scores or comments | Each rule switched on or off; changes proposed by the Director and approved by Compliance |
| Floor and caps | $25,000 floor · capped at the provider agreement price | Providers below the floor are not paid and get feedback; caps come from each provider agreement |
| Auto-match threshold | 0.92 | Raised from 0.90 after 1,200 hand-checked matches; false matches fell from 0.4% to 0.1% |
| Counterparty credit floor | Below BBB− or capital under 10.5% | The Credit agent flags; Counterparty Credit Risk proposes |
| Ballot clock | Due day 10, 5:00 p.m. ET · reminders T−2 and T−1 | Escalation to the desk head at the deadline; COO approval T+5, payments T+18 |
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 (a survey of research budgets at the 44 largest asset managers, reported by Hedge Fund Alpha, September 4, 2025 — top 10 brokers 55.1% of research budgets in H1 2025, and 87% of managers expecting at least half of European research budgets to be client-funded within two years; FCA multi-firm review of research unbundling reforms, 2019, updated August 2025 — budgets for external equity research fell about 20–30%; FCA PS24/9, July 2024, and PS25/4, May 2025, on payment optionality for investment research; SEC interpretive guidance on Section 28(e), 2006). People, firms, brokers, funds and issuers named on this page are fictional — characters and sample data in the working solution.
Broker Vote is broker vote software for asset managers. A broker vote is the twice-yearly review in which analysts and portfolio managers score the brokers and research providers they used, and with the desk's broker scorecard it sets the approved broker list and the research budget. In Broker Vote, seven agents pre-fill each ballot from logged research use, question scores that rest on thin evidence, score execution and counterparty credit separately and turn votes into dollars, while voters decide their own scores and the COO approves the list and the budget.
The Evidence agent matches CRM, calendar, research portal and model download logs to brokers and voters. The Ballot agent pre-fills every ballot with those counts, and the Policy agent checks each score against your research valuation policy as it is entered and asks the voter one question when a score needs a reason. Research Operations prepares the pack, the Director of Research reviews, Compliance countersigns and the COO approves.
The Budget agent applies voter weights and service weights, turns each broker's share of points into dollars, caps payments at the provider agreement price, leaves providers under the floor unpaid and assigns the funding source. The Policy agent then recomputes every line on its own; in the working solution 31 of 31 lines match to the dollar, and any difference over $1 stops the pack.
No. Desk and credit scores set the approved broker list and never enter research budgets, and the Policy agent flags any research comment that mentions execution. Corporate access is scored on the ballot but paid only from the firm's own P&L, never from a client-funded research account.
PS24/9 allows joint payment for research and execution with a research budget, a written policy, ongoing assessment of research value and price, a fair allocation of costs across clients, written provider agreements and disclosure to clients. Broker Vote keeps the evidence for each one, and the Pack & Notice agent drafts the annual assessment with the evidence beside each condition for Compliance to review. Whether your arrangement meets the rules stays Compliance's judgment.
Broker Vote decides how much each provider earns and from which funding source; payments still go out through your commission and payment system, which holds research account balances and payment instructions. The Pack & Notice agent drafts Form ADV Part 2A Item 12 input on research and other soft-dollar benefits from the approved vote. Whether an arrangement sits inside the Section 28(e) safe harbor stays Compliance's judgment.
No. The agents never score, accept, approve, remove, pay or send. Voters score only their own ballot, a removal is proposed by the Head of Trading or Counterparty Credit Risk and decided by the COO, and broker feedback letters go out only when a checker releases them — and never disclose an individual vote.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your research valuation policy, provider agreements, last cycle's ballots and the systems that hold your usage logs — and it goes live once every quality gate has passed. Each new cycle is then set up from the last one and back-tested before the Director approves it.
Committee pack in two hours, every outlier explained, every figure tied to its source.
1-min demoTransaction Cost AnalysisEvery trade's cost explained in plain words, by broker, algorithm and venue, by morning.
1-min demoCommission ManagementEvery research dollar matched to an eligible use and a signed split, before money moves.
1-min demoWe'll run Broker Vote on a sample of your own interaction logs, broker statements and last cycle's ballots.