Every treaty broker knows the room. It is the last two weeks before January 1, dozens of reinsurers are sending authorizations by email, portal message and scanned slip, and every written line arrives carrying its own conditions — a minimum signing, a collateral trust, a subjectivity or two. The broker re-keys each line into a spreadsheet by hand, works out the signing, and starts again every time a new line lands.
A vendor called Botminds has built its reinsurance placement product for exactly that room, and its problem statement deserves quoting because it is the correct one: “Brokers re-key each written line by hand, and a condition buried in an email can slip into the signing.”
Those are two different problems wearing one coat. The signing down is arithmetic. The buried condition is judgment. This piece takes them in that order — because the product does, and because confusing the two is how placements go wrong.
The arithmetic first: what signing down actually is
When reinsurers write more than 100% of a layer, the layer is over-placed, and the written lines have to come down to signed lines that total exactly 100%. Under the most common method — pro rata, with the lines that stand keeping their full share — every other line is multiplied by the same factor: what’s left after the standing lines, divided by what was written above them. The results are rounded so the layer totals exactly 100.00%. Then each signed line is checked against that reinsurer’s own minimum, and against the client’s buying instructions.
By hand, Botminds estimates, working one property catastrophe program’s signing takes about 17 hours — re-keying authorizations, checking conditions, checking security, signing down, writing the report. Their estimated split with the product is 4.5 hours. Hold both numbers at arm’s length: the page labels them estimates, built from its own by-hand rows, and we have not measured either one. What is not an estimate is the shape of the work. One new or withdrawn line changes every number in the signing, and by hand every change means the whole calculation starts over.
Their worked example — the page labels it sample data from the demo, and so do we — is a four-layer catastrophe tower, $330 million excess of $20 million, with 23 reinsurers on 39 lines. Overnight the layers stand written at 112.5%, 96.0%, 118.0% and 104.5%. One switch takes the tower from written to signed: Layer 1 comes down by a factor of ×0.9467, Layer 2 by ×0.8163, Layer 4 by ×0.8718, the lead’s line stands at full width, and each over-placed layer sits at exactly 100.00%. Layer 3 stays visibly short — which is its own story, below.
Here is the detail in that arithmetic that bites in practice, and the example shows it plainly. On Layer 2 the factor is worked in the open: with the lead’s 20% standing, the other lines share 80% of what was written above them — 80.0 ÷ 98.0 = ×0.8163 — rounded to 0.01% by largest remainder so the layer still totals exactly 100.00%. Then the minimums: one reinsurer wrote 12.5% subject to a minimum signed line of 10%, and signs at 10.21%. Another wrote 4.5% subject to signing not below 3.5%, and signs at 3.67%. Both hold — barely. A uniform factor does not know about contractual minimums. A layer can read a perfect 100.00% with a line sitting underneath the minimum its reinsurer agreed to, and that error is not discovered until the signed-line advice goes out — after which, as the page puts it, a signing error “has to be corrected with each reinsurer,” one by one, in the worst fortnight of the year.
The judgment: the condition in the third paragraph
Now the other problem. In the same worked example, one line differs from the firm order, and it is the most instructive five percent in the tower.
A reinsurer writes 5.0% on Layer 2 “subject to an hours clause of 144 consecutive hours for any one loss occurrence.” The firm order says 168 hours. If you place reinsurance you already know why that sentence is money; if you don’t, the page explains it in plain words: a shorter hours clause splits one storm into more losses, each with its own retention — less cover for the cedant. Twenty-four hours quietly removed from the clause is not a clerical variation. It is a different product than the one the client ordered, arriving inside an authorization that otherwise looks routine.
In the product’s design, that line gets caught by a terms comparer that lays each authorization against the firm order clause by clause — reinstatements, hours clause, exclusions, wording, premium, security — and marks every condition as a difference, an open subjectivity, or cleared. Four subjectivities sit open on this tower, each with a due date before inception. And here the design draws the line that matters: the comparer “accepts no difference; that is the broker’s and the client’s call.” It never clears a subjectivity on its own. The broker opens the line and sees the reinsurer’s own email, the 144-hour condition highlighted in its passage, with a one-button ask for the reinsurer to confirm 168 hours. When the signing is proposed to the client, the line is not hidden and not folded in silently: it is shown at its signed share of 4.08%, subject to that reinsurer confirming the firm order hours — an open point, stated, for the client to accept or refuse.
That is the whole ballgame in one screen. A line written on different terms does not belong in the signing until the difference is settled — and “settled” is a human decision, recorded, not a default.
Six agents, and the things they are not allowed to do
The product runs six agents, and the honest way to describe them is by their prohibitions, because the prohibitions are the design.
The line reader takes each authorization from email, portal or scan and records the written line, the layer and every condition with its source passage. It never records a line without the reinsurer’s written authorization, and a newer authorization supersedes an older one rather than editing it. The terms comparer does the clause-by-clause work above and shows the reinsurer’s exact words. The quote analyst builds the comparison by layer against the firm order and the expiring terms, and finds the reinsurers who quoted but have not written — it never sets the firm order price, and it drafts requests that the broker sends. The security checker tests every reinsurer against the client’s buying instructions — rating floor, US status, collateral for the unrated, maximum share, sanctions — and re-checks on every change to the signing; it never waives a requirement, and an exception goes to the head of placement. The sign-down calculator works the signing by the client’s chosen method and checks every minimum line; it proposes only, and it never sends a signed-line advice. The report writer drafts the placement report and the advices from the live line sheet in the broker’s own template, marked draft until approved, and sends nothing without the broker’s release.
Above the six sits the rule the whole page keeps returning to: the broker proposes, the client approves. The signing goes to the cedant’s head of ceded reinsurance as a proposal — each layer written and signed with its factor, open points stated — and signed-line advices go to reinsurers only after the client approves. The approval is recorded with name, time and meaning. Every line read, clause compared, recalculation and human decision sits on the program’s activity trail, and reinsurers see only their own lines.
The buying instructions themselves are settings, agreed with the client and saved once: signing method, which lines stand, a 2.5% default minimum signed line (a reinsurer’s own higher minimum wins), the 0.01% largest-remainder rounding, an A− rating floor, collateral at 100% of share of limit for unrated reinsurers, a 20% maximum share per layer and per program, reminders on open subjectivities at 10, 5 and 2 days before inception. The security checker and the calculator apply those settings again on every change — a rule set once is not a rule checked once.
And the short layer gets its honest ending. Layer 3 sits 4.0% short — $4.0 million of limit, $290,000 of premium. The quote analyst finds, among the eleven quotes on the layer, a reinsurer that quoted 7.00% against a 7.25% firm order with capacity up to 5% but never wrote a line. It checks that reinsurer against the buying instructions — AM Best A+, reciprocal jurisdiction, the line would take it to 5.3% of program limit against a 20% maximum — and drafts the request. The broker sends it. The authorization comes back for 4.0% at firm order terms, the line reader records it against its source, and the tower stands placed: $330.0 million of $330.0 million.
The fine print, read the way you read a slip
Botminds publishes its terms — that is worth saying first, because not every vendor in this space does. They sit under “Terms and Conditions” on the company’s own site, and they are real SaaS terms for the platform, not website-only boilerplate. The contracting entity outside India is Botminds Inc, a Washington corporation; the “Agentic Solution Engine” that builds each solution is the company’s own product, and this placement product is one of more than a thousand solutions the company says it and its partners have built on that engine across nine industries.
Now the broker’s questions, because they are the same ones you would ask of any system that will hold your clients’ firm orders and every reinsurer’s authorization.
The terms define Customer Data narrowly: “information pertaining to your clients that you submit or collect via the Subscription Services” — and then, pointedly, “Additional information we collate and provide for your use is not Customer Data.” For a treaty broker, that definition matters more than it might look. Everything sensitive in a placement — the firm order terms, the written lines, the conditions, the collateral positions — is information pertaining to your clients. The definition covers it. But the second sentence carves the vendor’s own collated layer out of it, and the operative clauses a broker actually needs — who owns the data, whether anything you submit can be used to train or improve the vendor’s models, how long it is kept, and what exactly happens to it when the contract ends — appear nowhere in the published terms at all: no ownership clause, no licence grant, no termination or data-return provision. By the terms’ own definitions, the operative contract is the order form and a separately executed SaaS Agreement. The company’s security page does state the right policies in plain words — “Your inputs and outputs are not used to train models. Your data works for you, not for the platform”; data “kept only as long as the workflow needs it”, with deletion requests that remove it “from the platform, verifiably”; and documents, extracted data and decisions all exportable — but policy is not contract. Read the full agreement before a single live authorization goes in, and get the training, retention and exit questions answered there, not in a demo. The website’s privacy policy will not help you: it governs the website — accounts, forms, payments — and says nothing about data processed through the platform. (It does disclose one thing worth knowing before you ask for a demo: when you request one, an AI agent reads public sources about you and your company to prepare for the conversation. Work-related only, kept to their team, and they will discuss it on request — but you should know the demo starts before the meeting does.)
What is public is the stated governance posture, and it points the right way: the company’s homepage describes one governance model across its solutions — a person signs every decision that matters, every value cites its source, every agent step is on record — with each customer’s solution isolated, access by role, encryption in transit and at rest, and — on the security page — an immutable audit log: “Every action on the platform — human or agent — lands in an audit log that cannot be altered. Full traceability for every AI-assisted decision.” As with the trust pages we have covered here before, that is policy and product design, not contract. The first question is whether the SaaS Agreement says what the pages say.
Two more silences, stated plainly. No pricing is published anywhere. Setup is bespoke: the vendor builds and deploys the solution from your buying instructions, firm order templates, report template and a sample of past authorizations, and it goes live when the vendor’s quality gates pass — they will run the product on a sample of your own authorizations and firm orders as a trial. That is a real implementation project with a sales process, not a subscription you start on a Tuesday. And there is no named broker, and no production result, behind this product’s numbers yet. Every worked figure on the page is demo data or an estimate, and the page says so with unusual honesty, labelling each figure a demo, a target, an estimate or a published market figure — the “same day” signing proposal is a target, the 17 hours is their estimate, and the −12% fall in property cat rates at January 1, 2026 is a published figure they attribute to Guy Carpenter via Artemis. The engine’s production proof on the homepage — validation protocols live at a global medtech manufacturer, hours saved on regulatory monitoring — belongs to a different industry on the same platform. Believe the labelling discipline; it is exactly what you want from a vendor. Just notice that it means the placement product itself is, on its own public record, unproven in production — and price that into how you pilot it.
What it means at your desk
If you place treaty business — or if your cedants’ programs are placed by brokers who do — three things are worth taking from this, whether or not this particular product ever touches your book.
First, signing down is the cleanest test case there is for agent work in placement. It is deterministic: given the written lines, the standing lines and the client’s method, there is one right answer, it can be checked to two decimal places, and every input can be cited to its source passage. Work that passes that test — arithmetic with citations — is work an agent can be trusted with. Work that fails it has no business near the judgment calls. When any vendor demos placement automation for you, ask to see the signing arithmetic worked in the open, factor and rounding and minimums, the way this page works it. If they will not show the arithmetic, they are selling you the other kind.
Second, the buried condition is the real risk, and it is not an arithmetic problem. The expensive sentence in your inbox this December will not be a number. It will be a clause — an hours clause, a reinstatement difference, an exclusion that is not in the firm order — sitting inside an otherwise routine authorization. The control that matters is not a faster spreadsheet; it is a clause-by-clause comparison against the firm order, with every difference surfaced in the reinsurer’s own words and held out of the signing until a named person accepts it. Whatever stack you run, that is the checkpoint to build and to audit.
Third, the trail is the product. A signing that can be reconstructed — every line cited to its authorization, every comparison, every recalculation, every approval with a name and a time — is a signing you can defend: to the client, to the reinsurer correcting an advice, and to your own E&O carrier if it ever comes to that. The fortnight produces errors under any system, human or agent. The question is whether the file shows what happened. Ask for the activity trail in any demo, and ask who else can see it — in this design, reinsurers see only their own lines, which is the correct answer.
The practice
On your next treaty placement — or the next one placed for your cedant — run the condition audit before the signing goes out, by hand if you have to. Take every authorization and lay it against the firm order terms, clause by clause: reinstatements, hours clause, exclusions, wording, premium, security. Mark each condition in one of three words — same, different, or open — and refuse to let anything marked different into the signing until the client has accepted it in so many words. Then check every signed line against its reinsurer’s minimum after the factor is applied, not before. It is an hour of unglamorous work on a live placement. It is also the exact hour where the money is — because the condition you catch in December is a conversation, and the condition you miss is a correction, a dispute, or a gap in the cover your client thought it bought.