What the agent does. What you approve.
A mandate on Maryah runs in five stages. At each one the agent produces named deliverables, the work is checked before you see it, and nothing moves to a counterparty without your sign-off. This page walks through all of it.
Produce. Critique. Approve.
Every deliverable on this page passes through the same three-step loop before it counts as done.
The agent produces
It does the work, not the summarising: it reads the room, drafts the model, writes the memo, assembles the list. Chat is how you direct it; the deliverable is what you get.
Draft producedA second model critiques
Before a draft reaches you, a separate quality-control pass attacks it: unsupported claims, weak logic, missing counterarguments. Anything the agent cannot verify is flagged as unverified, never asserted.
Unverified: flaggedYou approve the gate
The advisor holds the judgment: the valuation range, the pages a counterparty will see, the names on the list, every send. The agent proposes; it does not act on the outside world without you.
Approved: releasedThe room audits itself, on every ingestion.
The data room is the brain the agent reasons from, so its integrity is enforced by machine, not by hope.
Strict tie-out
A figure counts as verified only when it matches a verbatim quote in its source document. Close enough is not verified.
Flags, not silence
Unverified figures and conflicts between documents are flagged where you will see them, and the flag persists until a source resolves it.
The gap list
The room is scored against what a buyer's diligence will request, and the gap list tells you what is missing before a buyer does.
One brain, one answer
Because every downstream deliverable draws on the same verified room, the IM, the valuation, and the Q&A cannot quietly disagree with each other.
This is also why the agent's answers are worth having: when you ask it a question about the company, it answers from the verified room and shows the passage it stands on.
How a mandate runs, stage by stage.
What actually happens, what the agent hands you, and where your gate sits.
The data room becomes the brain.
Files arrive the way they always do: a founder's folder, an accountant's export, a scanned lease. The agent ingests whatever comes, names and indexes it into a governed data room structure, and extracts the figures.
Then it checks its own work. Every extracted figure must match a verbatim quote in its source document. Figures that tie out get a ledger entry pointing at the exact passage. Figures that do not are flagged, and the flag stays visible until a document resolves it.
From the same read, it writes the gap list: the documents and answers a buyer's diligence will ask for that the room cannot yet support, each gap with a note on why it matters and how to close it. That list becomes the chase, and the chase becomes the questions that go back to the founder.
Your gate: what goes back to the founder: the gaps worth chasing, the questions worth asking, in what order.
Format specimen. The gap list resolves itself as the room fills.
A valuation you can defend.
The agent drafts the valuation from the room's own numbers, in parallel methods: trading comparables, precedent transactions, and a cash-flow view. It does not invent inputs. Where the room supports a figure, the figure carries its source; where it does not, the assumption is named as an assumption.
The drivers come with evidence. Recurring revenue, customer concentration, margin quality, owner dependence: each driver the valuation leans on is documented with the passage in the room that supports it, so when a buyer pushes back you are holding the document, not an opinion.
It runs the sensitivities so you can see what moves the range and by how much, and the whole draft passes the critique loop before it reaches you. The range itself is yours: nothing anchors the mandate until you approve it.
Your gate: the range, before it anchors the mandate. You see every input's source before you defend the number.
Format specimen. Every figure ties to a verbatim quote, or it is flagged.
The materials, in your voice.
The agent frames the transaction first: full sale, majority, minority, or a raise, with the reasoning for the recommendation. That frame parameterises everything downstream, because a teaser for a minority raise is not a teaser for a full sale.
The blind teaser has to pass its own gate. Before a teaser can leave the system, the agent runs a re-identification check against it: could a knowledgeable reader in the sector name the company from what the page reveals? If yes, it does not go out as written.
Then the full information memorandum, drafted from the room so every figure in it inherits its tie-out, structured for your edit rather than your rewrite. It reads like the work of your firm because you shape the voice and approve every page.
Your gate: every page a counterparty will see, and the structure the process is built around.
- Re-identification risk: the revenue line narrows the sector to three companies
- Does not go out as written
- Re-identification check passed
- Every figure tied to its source
- Identity stays blind until the NDA
Format specimen. A teaser that could identify the company does not leave.
A buyer list that argues its case.
The agent builds the buyer list from identity, not from a financial screen. Who has bought in this space and at what cadence, what a group's own reports and strategy pages say it wants next, where the target closes a gap a buyer has publicly described. An EBITDA band is a filter; it is never an argument.
Every name arrives as a written case. The argument for the approach, the risk that could kill it, the recommendation, and the warm path in, each claim carrying its citation. You can read the format below, annotated part by part; that is exactly how the list arrives, name after name.
You then argue back. Strike names, add the buyer only you know about, re-rank. The agent absorbs the reasoning and the list gets sharper, on this mandate and the next one.
Your gate: who makes the first wave, who waits, and who never hears about this deal at all.
Format specimen. Placeholder names; every rank arrives with its written case.
The process, tracked end to end.
The agent stages the process: waves, sequencing, and the follow-ups that otherwise live in a spreadsheet and someone's memory. Identity stays behind the NDA; the blind teaser goes first, and the reveal happens only after the gate you set.
Room access is per recipient, never general. Each admitted counterparty gets its own watermarked view of the data room through a named, revocable link. Every open and download is logged, and the Q&A is captured as it happens so answers stay consistent across buyers.
Nothing sends itself. Every outbound message and every admission to the room is a proposal until you approve it. What the agent removes is the tracking burden, not your hand on the process.
Your gate: every send, every reveal, and every admission to the room.
Format specimen. Placeholder names, no client data.
Every figure knows where it came from.
This is what a line of the valuation looks like when you open it. Not a cell in a model: a claim with its evidence attached.
Input. FY2025 revenue: EUR 18.4m[1], of which 62% recurring under multi-year contracts[2].
Verification. Ties out. Source quote: “revenue for the financial year amounted to EUR 18,412 thousand”[1]. Recurring share computed from the contract register; the register and the ledger agree[2].
Where it lands. Drives the comparable-company revenue multiple and the base case of the cash-flow view. A 10% change in the recurring share moves the midpoint by EUR 1.1m: see the sensitivity table.
Open flag. One customer contract in the register lacks a signed renewal for FY2027. Marked unverified and added to the gap list; it will surface in diligence if it is not closed first.
Format specimen, placeholder figures. Trading comparables, precedent transactions, and the cash-flow view are drafted in parallel, and each input in each method carries an entry like this one.
Paragraphs, not scores.
A number out of a hundred tells you nothing you can act on. Every buyer on a Maryah list arrives as a written argument you can interrogate, edit, and overrule. Here is its anatomy.
The case. Four acquisitions in adjacent niches since 2023, all reported in the EUR 10 to 40m range[1]. The latest annual report names this category a platform priority[2], and the target closes a distribution gap the buyer's own strategy page describes[3].
The risk. Partial overlap with an existing subsidiary could read as consolidation, and this buyer has historically paid less for consolidation.
The warm path. A banker-to-banker route through the buyer's retained advisor, and a named corporate development lead who has taken this kind of call before.
Format specimen. Placeholder names and figures, no client data.
- The caseWhy this buyer, now: acquisition cadence, stated strategy, the gap the target closes. Each claim cited to a deal record, an annual report, or a public statement.
- The riskWhat could kill the approach, written down before you commit a wave to it: overlap, consolidation pricing, a rival process, a bad history.
- The recommendationFirst wave, second wave, or hold, with the reasoning. The agent takes a position; it does not hand you a shrug.
- The warm pathHow to get in the door: a banker-to-banker route, a shared advisor, a named corporate development lead.
- The evidence fileEvery citation resolvable to its source. When a claim cannot be verified, it is marked unverified, and you can see that before you rely on it.
What the agent argues from.
The buyer cases are only as good as the record behind them. Maryah maintains its own market layer, and it compounds with every mandate.
The tape
A record of transactions: who bought what, when, and in which niche, so a claim like “four acquisitions since 2023” is a lookup, not a guess.
The theses
What acquirers say they want, taken from their own annual reports, strategy pages, and announcements, held against what they actually buy.
The news
A continuous sweep of deal and market signals, so a buyer's list position reflects what it did last quarter, not what it did in 2021.
The graph
Buyers, funds, and the people around them, connected: who holds what, who advised whom, and where the warm paths run.
Every mandate that runs on Maryah feeds outcomes back into this layer. The record of what buyers actually did, and what advisors corrected, is the edge that compounds.
Maryah will not assert a figure it cannot source, will not contact a counterparty you have not approved, and will not close a deal without you. It removes the preparation, not the advisor.
Watch it run a real stage.
The demo is a live walkthrough of the system on a worked example: the room, the tie-out, the valuation, the buyer cases. Bring the hardest questions you have.