Clossara reads the email threads behind every mandate, facility and renewal and tells the relationship manager why a deal is stalling, whether it will close, and who has gone quiet, with the message behind every finding. It runs in a dedicated EU instance or on a box inside the bank. No third-party AI. No training on client data.
Corporate and commercial banking, private banking, transaction banking and treasury sales.
A relationship manager carries dozens of relationships and hundreds of open threads. The renewal that went to another bank, the treasury cross-sell that was never followed up, the CFO who stopped replying after the term sheet: each one was visible in email weeks earlier. Nobody had time to re-read it.
A facility renews in 60 days. The CRM says "on track" because the last call note says so. The thread says the CFO asked two questions three weeks ago and got one answer. The other bank got both.
The treasurer mentioned FX exposure in passing. It sits in message 14 of 31. Nobody opened a cash-management conversation, because nobody saw it.
One champion in finance, nobody in treasury, nobody on the board. When the champion moves on, so does the mandate. The pipeline review never asked who else was on the thread.
Every conversation-intelligence product your team evaluated sends client correspondence to a shared cloud model. For a bank that is an outsourcing of client data to an AI vendor, and the answer was no. So the reps went back to guessing.
Clossara reads each mandate's full email history against the CRM and writes a short diagnosis. Every line points at the message it came from. What the model cannot support, it does not show.
Why is this mandate stalling? The root cause and the signals behind it: an unanswered question, a slipped date, a single thread, a counterparty who went quiet, each cited to the message.
Will it close, and how confident is that, from the conversation rather than the pipeline stage. A thread with too little signal says so instead of guessing.
A reply that answers what the client actually asked, in the bank's tone. Any rate, fee or date that is not in the thread or your documents is flagged before it becomes a promise. The RM reviews and sends from their own mailbox.
These are the questions bank security, outsourcing and model-risk teams have asked us. The answers are what is in place, not a roadmap.
| Your requirement | How Clossara meets it |
|---|---|
| Client data stays in our jurisdiction | A dedicated instance hosted in the EU on Enterprise, or the Clossara Box on your own network, where nothing leaves the building. Your choice. |
| No client correspondence to a third-party AI | Clossara's own model, no third-party LLM at runtime. On the box, the model runs on your hardware; in the cloud it runs on servers Clossara operates, and sees redacted text only. |
| Client confidentiality | Names, addresses and identifiers are replaced by labels before any text reaches the model. The Portal puts the names back for the RM; the model never holds them. |
| No use of our data for training | Never. Not in the cloud, not on the box. The model is trained on Clossara's own development data and ships as a gated release; it reads your threads, it does not learn from them. |
| Model risk: can we check what it says? | Every finding cites the message it came from; a finding without a citation is not shown. Each release is measured on a fixed held-out set and the report is available to your reviewers. |
| Least privilege on mailboxes | Read-only mailbox permissions (Gmail gmail.readonly, Microsoft 365 Mail.Read). Clossara cannot send, delete or modify mail. Drafts are copied and sent by a person. |
| Access control and audit | Enterprise adds audit logging and an SLA; SSO and SCIM are coming soon. Every stored analysis is traceable to the thread and the release that produced it. |
| Outsourcing and DORA questionnaires | We complete your security questionnaire as part of the review and provide the DPA and subprocessor list. On the box there is no processing outside the bank to register. |
| Independent certification | in progress. We say so rather than imply it. Ask for the current status. |
Details, data flows and retention are on the security page. Nothing here is specific to banking law; it is what the product does.
Every release is judged on 176 held-out sales threads it has never seen, finding by finding, against the full conversation. The current release:
of the findings an RM sees were judged true against the thread. What the model cannot support, it hides.
of citations point at a real message. The reviewer can open the message behind every claim.
invented numbers in the gated release. A rate, fee or date not found in the thread or your documents is flagged for review.
November 2026 release. The four-page evaluation report is sent with every briefing.
Enterprise: your own application and database, isolated from every other customer, hosted in the EU. The model sees redacted text only, over an encrypted link, and stores nothing.
The whole product, model included, on a desktop-sized AI box on the bank's network. No outbound connection. Updates arrive as packages your IT installs. Switch it off at the wall.
On the Clossara Box, no: there is no outbound connection. On a dedicated EU instance, redacted text goes over an encrypted link to Clossara's own model server, which stores nothing and is operated by Clossara, never a third party. The security page lists exactly what is read and where it is kept.
No. Clossara's model is trained on Clossara's own development data and shipped as a gated release. It reads your threads; it never trains on them, in the cloud or on the box. That is in the terms, not only on this page.
Email threads from connected mailboxes with read-only permissions, CRM records including logged calls where connected, and documents you attach such as pricing and term sheets. Chat platforms and voice recordings are not read today.
Yes. Threads can be imported as .eml files from a mailbox export. Import a set of mandates whose outcome you already know and compare what Clossara says with what happened. That is the fastest honest pilot.
Each person sees the threads their mailbox and CRM access already give them. Clossara adds analysis on top of existing access; it does not widen it. Team dashboards show the deals, not the private correspondence of other RMs.
A small team that builds and trains its own model. We would rather tell you that than pretend otherwise, and it is why the box exists: a bank should not have to trust a vendor's operations when it can keep the whole thing inside.
Thirty minutes with the people who built it, with your security, outsourcing and model-risk reviewers in the room. Tick the box below and the evaluation report comes with our reply.
Prefer email? Write to [email protected]