Portfolio commentary drafted by agents from attribution and the PM's notes, every sentence tied to its figure, for the PM to review and sign in minutes
Every sentence tied to its attribution figure, and PMs review in minutes, not hours.
See one case, screen by screen ↓Portfolio Commentary Generation is AI portfolio commentary software for asset managers. It drafts each strategy's quarterly fund commentary from signed-final performance attribution and the portfolio manager's own notes, and checks every sentence before a person reads it. Seven agents write every version, tie each figure to its attribution cell, send predictions and gross-only figures to compliance and keep the factsheet, mandate letters and translations saying the same thing, while the writer reviews, the PM signs, compliance clears and a fourth person releases.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Every quarter the investment writers and product specialists owe a portfolio commentary for each strategy: what the fund returned against its benchmark, what drove it, what the manager did and how the portfolio is positioned. Then the same investment commentary is cut into a 900-character factsheet paragraph, letters to institutional clients whose accounts returned slightly different numbers, a plainer version for wealth advisers and translations for every market the fund is sold in. It is assembled by hand from the performance attribution report, the PM's emails and the trade blotter, then chased through PM approval and a compliance check, all in email.
The risk sits in the sentences nobody had time to prove. Basis points written as percent, a total effect described as an allocation effect, a gross figure with no net beside it, a PM's "we expect the rally to continue", a prediction FINRA Rule 2210 does not allow, a Japanese translation that says 0.83% where the master says 0.38%. When the PM edits the master on day nine, the factsheet, the mandate letters and the translations were cut from day seven's text. Each miss is a reissue to clients and a books-and-records problem.
Estimated split of one strategy's quarterly commentary, by hand and with the solution, in writer hours. Waiting time for PM approval and compliance is not included.
Seven agents draft every version, check each other sentence by sentence, and send what they cannot prove to a person; the PM signs and a fourth person releases.
Quillfield Capital's Q3 2026 cycle has 214 strategies and a publish deadline of Wednesday, October 21. Here is how its people work with the agents on one of them — Global Core Equity, a $6.8 billion strategy run from London against the Halden Global Developed Index — screen by screen in the working solution: what they ask, what the agents show as evidence, and what each person decides and signs.
Maren has 9 drafts in writer review, 5 due today. Overnight the Drafting agent wrote 207 of 214 strategies within an hour of their attribution being signed final; the other 7 wait for final numbers. The Fact-check agent checked 18,940 sentences and corrected 286, settling 241 with the Drafting agent and sending 45 to writers. Global Core Equity is first on her list: 1 line for the PM, 1 translation issue.
She can ask the agents anything about the cycle from the same screen: “Why did Global Core Equity outperform in Q3?”
Maren types her questions in plain words. “Why did Global Core Equity outperform in Q3?” The Fact-check agent answers: 3.93% after fees (4.12% before) against 3.47%, stock selection +54 basis points, the largest effect Industrials selection at +38, led by Vossberg after its August 6 guidance raise — each figure cited to the returns, the attribution or Vossberg's release. “Which PM approvals are past their SLA?” The Publishing agent names three, says desk heads were copied, and offers reminders it has drafted: nothing is sent until she presses Send.
“Answers cite their sources; anything that changes data waits for your approval.”
The fund commentary reads like the printed page: 10 sentences, 16 figures. Class I returned 3.93% after fees (4.12% before) against 3.47% for the index; stock selection added 54 basis points. Beside it sits the Brinson attribution report, ATT-GCE04-26Q3 v3 (Brinson-Fachler, by sector), signed final by Ana Ruiz in Performance on Oct 8 at 17:42 ET. Every figure sentence is drawn to the cell it comes from; green is verified, blue was corrected by the agents, violet comes from the PM's notes or trades, amber needs a person.
“Hover a sentence to light its source; click it to see the check.”
“Stock selection in Industrials added 38 basis points, led by Vossberg Antriebstechnik, which rose 21.4% after raising its full-year guidance on August 6.” The check shows the cell (Industrials selection +0.38%, page 2), the unit converted to basis points, the period, the final status, and Vossberg's 21.4% return from portfolio accounting. The reason comes from Vossberg's ad hoc release, shown in German beside an English reading. Confidence 99%.
“… the Management Board raises its 2026 revenue forecast to EUR 4.1 to 4.3 billion” — Vossberg release, page 1.
The run took 11 minutes and read 12 sources, among them the attribution, official returns, 92 holdings, 19 orders, Hugo's two notes, the German filing, market data, the house view and the style guide. The Fact-check agent, with different instructions and no access to the Drafting agent's reasoning, sent two sentences back: the yen rose 4.1%, not 6.1% (USD/JPY 158.40 to 152.15), and the Financials overweight added 9 basis points, not 12, which is the sector's total effect. Both were corrected before Maren opened the draft.
Disagreements the agents cannot settle in two rounds go to the writer, with both positions shown.
The claims grid compares the master with the factsheet, the UCITS fund, three mandate letters (Halsey County, Brandvold Pension, Kestrel Hill), the adviser version and the German and Japanese versions. Different numbers that are right are explained: the UCITS class returned 3.95% net, Halsey County's account 4.08%. The Japanese monthly report, read in Japanese, says 0.83% for Industrials selection where the master says 0.38%. Maren accepts the fix and the translation vendor is notified.
“The digits look transposed in translation.”
Sentence 10 reads “We expect the rally in Japanese equities to continue well into 2027 and see significant further upside.” The Tone checker ties it, dashed amber, to the Investment Committee's house view (Japan a modest overweight, no return expectations) and to style guide §4.2 on predictions, and proposes wording — but leaves the line to Hugo. With the Japanese fix in, Maren sends it to him: “Hugo — sentence 10 reads as a prediction and is stronger than the house view on Japan.”
“It is Hugo's view, so I won't rewrite it myself.” — Tone checker
Hugo opens a change summary: 2 corrections the agents settled, 1 line for him, 1 translation fix accepted. He rewrites sentence 10 in his own words: “We remain modestly overweight Japan, and we are watching the path of interest rates closely.” Before it is saved, four agents re-check it: no figures, and the positioning matches the holdings (Japan 9.8% against 6.1%); in line with the house view; a views statement for compliance; 6 versions carry the old Japan line, so aligned text is proposed.
“Your edit is re-checked by the agents before it is saved; the first draft stays on the record.”
Hugo's sign-off shows every check passing — 16 figures tethered and verified, his edit in line with the house view, 6 versions aligned — and he signs “Approved as edited”. The Disclosure agent sends one statement to Priyanka Rao in compliance, with the proposed disclosure: “Views are those of the portfolio manager as of September 30, 2026 and are subject to change.” She clears it with the standard disclosure. The same queue holds the cycle's other flags, such as “poised to outperform over the next 12 months” in International Small Cap.
The Disclosure agent never clears a statement. Only compliance does.
The release check names four different people: Maren wrote it, Hugo approved it, Priyanka cleared it, and Lena releases it. Nine versions go out together: the factsheet paragraph to factsheet production, the fund pages to the website, three mandate letters to the client portal, the record copy to document management, an activity in the CRM, and a retail communication onto the FINRA filing calendar. The German and Japanese versions go to their distributors.
“Four different people — the releaser neither wrote nor approved this commentary.”
The same Q3 cycle, seen by five of the people who carry it — what quarter-end commentary looked like, and what it looks like now.
Seven agents draft every version, check each other sentence by sentence, and send what they cannot prove to a person; the PM signs and a fourth person releases.
Watches the performance system for attribution signed final, then gathers the official returns by share class, holdings and contribution from portfolio accounting, filled orders from the order management system, market data and the PM's notes for the period.
Plans the story — performance against the benchmark, what drove it, what the manager did, how the portfolio is positioned — and writes the fund commentary, the factsheet paragraph of at most 900 characters, mandate letters with each account's own returns, the adviser version and translation briefs, in house style.
Checks every figure against its cell: value, basis points or percent, sign, period, final status and the gross-and-net pairing. Ties each sentence to its cell, note or passage, and sends failures back to the Drafting agent with the reason.
Compares every sentence with the current house view and the style guide, marks sentences that stray, predict performance or use words the firm does not use, and proposes wording in line with the house view.
Compares each version with the master — figures, message and disclosures — reads German and Japanese versions in their own language, explains right-but-different numbers for share classes and mandate accounts, and proposes aligned text when the master changes.
Screens every sentence for predictions (FINRA Rule 2210(d)(1)(F)), gross attribution shown without the fund's total gross and net return for the same period (SEC Marketing Rule), and the disclosures each audience and jurisdiction requires, and sends each flag to compliance with the rule, the reason and proposed wording.
Keeps each strategy's SLA clock and drafts reminders for a person to send. After release it sends each version to its channel, confirms receipt, logs the CRM activity, adds retail communications to the FINRA filing calendar and files the record copy.
Drafts, approves, clears, releases. The agents propose; a named person decides.
Ask about any strategy, sentence or version in the cycle, or tell the agents what to do. Answers cite the attribution, note or filing they come from, and anything that changes data waits for your approval.
Why did Global Core Equity outperform in Q3?
Global Core Equity returned 3.93% after fees (4.12% before) against 3.47% for the Halden Global Developed Index. Stock selection added 54 bps; the largest effect was Industrials selection, +38 bps, led by Vossberg after its August 6 guidance raise. Health Care selection cost 24 bps (Orrelle) and the Energy underweight 11 bps; currency added 9.
Which PM approvals are past their SLA?
Three PM approvals are past their 2-day SLA: US Large Cap Growth (Celeste Marchetti, 1 day), Core Plus Bond (Dana Whitfield, 1 day) and International Core Equity (Hugo Ferrante, 4 hours). Each has a change summary ready; desk heads were copied at the breach. I drafted reminders — send them?
Which commentaries still predict performance?
57 statements are open with compliance across 24 strategies; 41 have proposed wording. The ones still predicting performance: International Small Cap (“poised to outperform”), Japan Equity (“15% upside for the index”, from a note written in Japanese) and Global Multi-Asset Income (“income is assured”). FINRA Rule 2210(d)(1)(F) does not allow predictions.
Why is PM approval the slowest step?
PM approval is the longest step: a median of 1.3 business days of the 3.0 from final attribution to publication, against a 2-day SLA. 88% of PM approvals were within SLA. Writer review takes 0.6 days because the agents settle 241 of 286 corrections before the writer opens the draft.
13 screens from the working solution, on its sample data. Pick one to see it large.
When the agents may act and when they ask are settings, versioned and approved by a second administrator.
| Setting | Default | Choose from |
|---|---|---|
| Tether required for every sentence with a figure | On | On · Off |
| Auto-accept agent corrections at or above | 97% confidence | Any percentage you set · only when the Fact-check and Drafting agents agree |
| Escalate a disagreement after | 2 rounds | 1 round · 2 rounds · 3 rounds |
| Tone checker sensitivity | By audience | By audience (strict for retail and adviser versions) · Strict everywhere |
| Hold a translated version on any figure mismatch | On | On · Off |
| Net return first; gross beside it, same period, equal prominence | On | On · Off |
| No predictions or projections of performance | On | On · Off |
| Keep commentary records | 7 years | 5 years · 7 years · 10 years |
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 vendor-published estimate — 2–3 hours to write the commentary for one strategy; a client reporting survey of asset managers and asset owners, September 2025 — 12% say their reporting meets all client needs and 60% cite a lack of automation; FINRA Rule 2210(d)(1)(F) — no predictions or projections of performance; FINRA Rule 2210(c)(3) — retail communications promoting a registered fund filed within 10 business days of first use; SEC Marketing Rule, Advisers Act Rule 206(4)-1, and staff FAQ of March 19, 2025 — attribution shown gross needs the total gross and net return for the same period; SEC Rule 204-2, books and records). Quillfield Capital and every person, fund, index, issuer, client and account named on this page are fictional — characters and sample data in the working solution.
Portfolio Commentary Generation is AI portfolio commentary software for asset managers. It drafts each strategy's quarterly fund commentary from signed-final performance attribution and the portfolio manager's own notes, and checks every sentence before a person reads it. Seven agents write every version, tie each figure to its attribution cell, send predictions and gross-only figures to compliance and keep the factsheet, mandate letters and translations saying the same thing, while the writer reviews, the PM signs, compliance clears and a fourth person releases.
The Intake agent waits until a strategy's attribution is signed final, then reads it with the official returns, holdings, trades, market data and the PM's notes. The Drafting agent plans the story — performance, what drove it, what the manager did, positioning — and writes the performance commentary and each version in house style; every reason must come from a source and every view from the PM's notes. In the working solution it drafted Global Core Equity's fund commentary and eight versions in 41 seconds.
A separate Fact-check agent recomputes every figure against its cell — value, basis points or percent, sign, period, final status and gross-and-net pairing — and ties the sentence to it. Failures go back to the Drafting agent with the reason; after two rounds the writer decides, with both positions shown. In the demo it caught a Financials overweight credited with 12 basis points, which was the total effect, not the 9 basis points of allocation.
The Disclosure agent screens every sentence for predictions or projections of performance under FINRA Rule 2210(d)(1)(F), for attribution shown gross without the fund's total gross and net return for the same period under the SEC Marketing Rule, and for the disclosures each audience and jurisdiction requires. Each flag goes to compliance with the rule, the reason and proposed wording. The agent never clears a statement, and retail communications go onto the FINRA filing calendar when released.
Yes. The Consistency agent compares every version with the master — in the demo, nine of them — and explains numbers that are rightly different, such as a UCITS share class or a mandate account's own return. It reads German and Japanese versions in their own language, and when the PM edits the master it proposes aligned text for each version and waits for its owner to approve.
No. Fund manager commentary stays the manager's own. The Tone checker marks a line that strays from the house view or predicts performance and proposes wording, but it never rewrites a PM's view. Only the PM changes it — even the writer cannot edit that line, and can only send it to the PM with the proposal.
Yes. The writer reviews and routes, the portfolio manager of record approves and signs, compliance clears each flagged statement and a fourth person releases, each with a reason code. These duties are enforced, not advised, and every decision is on an audit trail that exports as an evidence pack.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your templates, style guide, house view, disclosure library and a past quarter's attribution and commentary — and it goes live once every quality gate has passed. First-time setup connects the systems of record, maps strategies to templates, sets roles and runs a trial on three strategies from a past quarter, compared with what was actually published.
Quarterly packs out by business day 3, not day 10, each figure tied to the book of record.
1-min demoFactsheet AutomationFactsheets out on business day 3, not day 10, with every figure and disclaimer checked.
1-min demoGIPS Compliance ReportingComposite errors found before the verifier finds them, with every move proven twice.
1-min demoWe'll run Portfolio Commentary Generation on a quarter you already published: your attribution, your PMs' notes, your house view and your style guide.