Each of the five things they asked for, shown as a chain — evidence, inference, decision — then the artefact, then how to break it. If you disagree with the words that's cheap. If you can break the chain, the words were never the problem.
⚠ Pre-human
No human has read a word of this. Machine output, published raw so the reasoning can be attacked before anyone polishes it. Judge the chain, not the copy — if a numbered step is wrong, say which.
How to judge a pre-human strawman
Attack the chain, not the copy. "I don't like the headline" cannot be acted on. "Step 3 is wrong, our buyer isn't personally liable" rebuilds everything downstream.
Ask what would falsify it. Every chain ends with a stated break-condition. If that condition is already true and we didn't know, say so first.
Judge the evidence grade, not the confidence of the prose. Machine copy sounds equally certain whether it rests on a measured fact or a benchmark guess. The grades are marked.
Assume the voice is wrong. Meaney's register was derived from nine public posts by something that has never met him.
1 · Short-term messaging — workshop Thursday
1
The intended position is already taken. Four of six competitors run the supervised-autonomy argument.Duco stand: "Agentic reasoning. Human judgement. Rules-based processes." · AutoRek ARIA: "explainable, auditable… enhances — not replaces" · Xceptor: "how can firms adopt AI safely without introducing new operational and regulatory risks?" measured
2
All of them argue it to the trading floor and middle office. None argues it to the accountable party.No competitor content addresses depositaries, trustees or ManCo directors. measured
3
Fund Recs' buyer carries personal regulatory liability. Theirs does not.RE holds the licence; depositary is strictly liable for loss of assets under AIFMD/UCITS; ManCo directors are named. A bank's head of middle office is not personally on the hook for a NAV. domain
4
Their content states the control and almost never the consequence.The two hardest consequence figures they own — GlobeOp US$43.5m and the SS&C US$5.9m fraud — are on the blog, from 2015 and 2020, and on no product page. measured
→
Therefore the differentiator is not the technology stance. It is who is answerable. The move is changing the subject of the existing argument from the software to the signatory.
The strawman
Fund Recs is oversight for the people who sign.
Everyone in this market is now selling autonomous AI to operations teams. Fund Recs sells to the person who signs off a NAV they did not calculate — the depositary, the trustee, the ManCo director — for whom an agent that cannot show its working is not efficiency, it is an audit finding.
Today
Strawman
"Automates stale price detection by comparing day-over-day price movements"
"A stale price looks exactly like a stable price. That's why it's the one that gets you."
"The doing platform for funds"
"Oversight for the people who sign"
"Access a growing library of AI-enabled workflows"
"98 controls that already know what a capital call is"
Workshop shape: don't open by choosing a position. Open by picking the three numbers they will repeat for twelve months and retiring the rest — "up to 90%" in particular, because five competitors claim the same range and it therefore carries no information.
Breaks if: their actual buyer is a middle-office operations lead rather than an accountable officer; or if win/loss shows they lose on features, not on trust.
2 · Product press release — was blocked, now isn't
1
They already have a published, reconciliation-side AI agent.Capital Call Data Extraction Agent — 737 words, live, categorised "Agent:", built on confidence scoring with user-set acceptance thresholds. measured
2
It carries no number and no client.Zero quantified claims on the page. measured
3
Rex proves they know how to write this well — in the CEO's personal feed.95% of conversations, 42% independently resolved, 94.1% CSAT, n=699 / 4,317 replies / 99 ratings. Support, not product. measured
4
The category benchmark is soft and beatable on method rather than magnitude.SmartStream label the same before/after "70%" in March and "97%" in June — no named client, no sample size, no methodology. measured
→
Do not announce AI. Announce one named agent doing one narrow job, with the denominators shown. A modest number with method beats a large number without it.
The strawman — headline and lede
Fund Recs publishes the accuracy of its capital call extraction agent — including where it gets things wrong
Dublin, [date] — Fund Recs today published measured performance for the Capital Call Data Extraction Agent, together with the method used to measure it.
Across [N] capital call notices processed between [date] and [date], the agent extracted [N] fields at [X]% confidence or above, with [Y]% requiring human review. Fund Recs is publishing the review rate alongside the extraction rate because, in an oversight function, the number that matters is not how much the machine did — it is how reliably it knew when to stop.
The one blocker, precisely stated: we need N, X and Y from product. Any real figures work — a 71% extraction rate published with its method is a stronger release than 97% published without one, because the second is indistinguishable from SmartStream's and the first cannot be argued with.
The real decision: is publishing a review rate — effectively an error rate — commercially acceptable? The argument for it is that their buyer's entire job is knowing when the machine is wrong.
3 · LinkedIn article — ship before 15 Sept
1
Meaney out-reaches his own company.8,915 followers vs 6,126. Same InvestOps post: 27 reactions on his profile, 5 on the company page. measured
2
Engagement is inverted — product content dies, human content travels.Rugby 102 · golf 79 · client videos 64–69 with 10–12 reposts · product posts 11–15 · the AI panel 5. measured, n=24
3
He is moderating an AI governance panel on 15–17 September.Themes: accountability, auditability, human-in-the-loop, measuring beyond headcount. With Man Group, Royal London AM, Partners Capital, S&P Global. measured
4
His register is specific and reproducible. Short declarative opener, arithmetic shown, caveat volunteered, one compressed closing line."The plumbing is the moat." "That's not AI replacing people." inferred, 9 posts
→
Ship from his profile, before the panel, to set the question he asks from the chair. A moderator who has published the question owns the room's frame.
The strawman — opening
Next week I'm moderating a panel on AI in investment operations. Here's the question I actually want answered.
Most of the agentic AI conversation in our industry is aimed at the middle office, where the cost of an error is time and money. That's a real problem and the technology is genuinely good at it.
But a large part of this industry isn't the middle office. It's depositaries, trustees, ManCo directors and fund boards — people whose job is not to process the work but to be answerable for it.
If an agent resolves 42% of a workload autonomously, the operations question is: what did I save? The oversight question is: can I evidence what it did, six months from now, to someone who is investigating me?
The caveat: I sell software that does this, so treat my framing accordingly. But the panel isn't mine — it's Man Group, Royal London, Partners Capital and S&P Global, and I'd rather hear their answer than give mine.
Automation that can't be evidenced isn't a control. It's a hope with a dashboard.
Highest-risk item on the site. This is a ghostwritten draft of a man's own opinion, built from his public posts by something that has never met him. Treat it as a prompt for him to rewrite. If he sends it unedited, we've failed.
4 · Homepage carousel — HubSpot CMS
1
They introduce themselves two different ways.Website: "the #1 Library for Data Automation". LinkedIn bio: "Reconciliation and data processing software for the global Funds Industry." measured
2
Their best proof exists and is buried.Channel Capital 79% straight-through. EMIR 5%→20% and 57%→97%. 41.6bn data points across 3,000+ funds — on a DTCC integration page. None on the homepage. measured
3
Client faces travel; product copy doesn't.Northern Trust video 69 reactions / 12 reposts, Apex 64 / 10 — against 11–15 for product posts. measured
→
The carousel is not a feature tour. It is three numbers and a face — assets they already own and don't use.
Slide
Content
1 · the claim
Oversight for the people who sign. 98 controls. Built for funds.
2 · the operator
79% of position reconciliations completed straight through, no human touch — at Channel Capital, across A$18bn.
3 · the regulator
41.6bn data points checked across 3,000+ funds since EMIR Refit.
4 · the person
Clive Bellows, Northern Trust, on camera: "It will reduce risk for us… a really good product and a really good team. Put the two together and that's really powerful."
Slides 2, 3 and 4 need no product input and could ship this week. Only slide 1 depends on the messaging decision. Free fix the same day: the LinkedIn bio still says "reconciliation and data processing software" — change it to match slide 1.
Breaks if: client consent is refused. Channel Capital merged with Fidante in June 2026 to form Channel Group, A$150bn — re-consent is required, not assumed, and the figures may now be stale.
5 · Pull-up banner — printer Friday 4th, the only hard deadline
850 × 2000mm roller, read from four metres by someone walking to coffee. One idea, no logo-and-strapline.
Route A — recommended
"It is extraordinary to see some trustees comfortable with limited, almost entirely manual indicators to monitor potential harm."
Simone Constant · Commissioner, ASIC
We automate the indicators.
Why: it isn't Fund Recs' opinion, it's a regulator's — quoted verbatim, already published on their own blog. Nobody argues with a regulator at a coffee stand, and it sets up the panel Meaney moderates two days later. Risk: ASIC is Australian and the room is European, mitigated by the fact that every EU regulator is saying the same thing.
Routes B and C
B · the number.$43.5m — what one miscalculated NAV cost one administrator. GlobeOp to Regents Park, 2009. Stoppable, but fear-led. Never run without the attribution line.
C · the question. "How many portals does someone open before they can finish one reconciliation?" One client's answer, before us: eight. Best for footfall, weakest strategically. Use as the tabletop card at the coffee competition.
Run A on the banner, C on the table. A does the strategic job, C starts conversations. They don't compete.
Verify before print: check the ASIC quote against ASIC's original statement, not Fund Recs' blog — never print a regulator's words from a secondary source, and confirm the ellipsis is honest. Stand number unknown. "Oversight for the people who sign" is a new strapline and needs Meaney's sign-off, not marketing's. This is a layout, not print-ready artwork.