Farsight deploys on-premise across Deutsche Bank IB
Farsight says Deutsche Bank deployed it on-premise across investment banking and capital markets teams, keeping data and workflow activity inside the bank.

Farsight's October 13, 2026 announcement is easy to file as one more bank artificial intelligence (AI) logo. Having watched banks license general-purpose assistants and then argue about what may leave the building, the fact that matters is where the system runs. Farsight said Deutsche Bank has deployed Farsight on-premise across Investment Banking and Capital Markets teams in Europe and the Americas, keeping sensitive data, proprietary knowledge and workflow activity inside the bank's own technology environment.
The GlobeNewswire text, linked from that page and timestamped 09:00 New York, draws a narrower map. In the copy reprinted by Markets Insider, the bank "has deployed Farsight at present" across those teams in Europe and North America. Neither version gives a contract value, a seat count or a date beyond the announcement. FinTech Global's same-day account matches the on-premise claim and adds no price.
Teams can apply the system to internal data while the bank keeps greater control of access and use. Farsight says it can fold prior work, internal standards and proprietary data into workflows built around a group's formats, processes and standards. Bankers then draft and revise in an automated chat that carries context and feedback forward. In the product description, not as a bank metric, teams also use a web interface plus PowerPoint, Excel, Word and email, and the platform can run partly or wholly inside the firm's own environment.
Farsight's security page lists on-premise beside a single-tenant Amazon Web Services-hosted virtual private cloud and a customer-cloud option, and states that no customer data is used to train models. House style, on that account, has to come from precedent, templates and desk rules rather than from training on the bank's files.
Standards still sit with the desk
"Our teams have very specific standards for what goes in front of a client, and those standards vary from group to group," said George Tierney, managing director and global chief operating officer for mergers and acquisitions at Deutsche Bank, in the October 13 announcement. "Farsight gives our teams a way to apply AI to their own workflows and materials while operating within Deutsche Bank's technology environment. That combination of control and customization is critical to making AI work in a global, regulated enterprise like Deutsche Bank."
Samir Dutta, co-founder and chief executive of Farsight, said: "Running Farsight within a bank's own infrastructure allows that knowledge to remain inside the institution while becoming part of how teams use AI. By incorporating banker feedback, Farsight can help AI reflect how a bank approaches client work rather than relying solely on general-purpose outputs." The note does not say a draft reaches a client without a person.
No rival vendor and no peer bank is quoted. Hebbia said on October 13, 2026 that an Octus integration would give credit teams research, deal terms and private-company data inside its workspace. That pulls more outside material in. It is not an on-premise deployment at an investment bank, and it is not a reply to Farsight.
Bloomberg's April 29, 2026 feature described Rogo, built by former bankers for deal-document work, as a $2 billion AI tool. Farsight's note states no price and instead fixes where firm knowledge may sit. Silence from Rogo or Hebbia is not a market-wide switch.
JPMorgan's arrangement with SambaNova is on-premise inference at the chip layer, alongside SambaNova's $1 billion raise, not a tool for pitch books. Saris's $28.8 million raise targets agentic AI in bank back offices, which is operations rather than client materials. Supervisors are pulling frontier-model risk into technology rules, including a convergence across the European Union, the United Kingdom and Singapore.
On June 26, 2025 Farsight said it had raised a $16 million Series A. On October 7, 2026, six days before this note, it announced a multi-year partnership with TresVista covering research and client deliverables for more than 400 active clients. TresVista is a managed-services firm, not an investment bank, and that release describes templates and deliverables rather than an on-premise bank install.
Farsight's October 24, 2024 survey, which it said covered more than 50 investment-banking firms, reported that 36% had a pilot that year and that one-third had a production use case. Sanctioned ChatGPT workflows counted, and about 70% had a pilot or subscription, including what it called Microsoft CoPilot or a firm-level ChatGPT license. A named on-premise install is still a different object from a chatbot seat.
The next test is adaptation without training. Feedback is supposed to shape outputs to how a bank approaches client work, while customer data is not used to train models, so any change has to sit in templates, retrieved precedent and desk standards. If a coverage group will send the result, peers will copy the residency pattern and keep the data inside.
Reporting by Rick Steves. Filed 14 October 2026, 13:53 GMT.




