What should we ask a vendor before wiring an assistant into our fund data?

Last reviewed: 17 September 2026

Eight questions. What does it read, and is it live or a copy? Can it write back, and to which fields? Who grants the permission and how is it revoked? Is every action recorded and exportable? Is our data used to train a model? Where is it processed, and can that be restricted by region? What happens to agent access at offboarding? Can output be traced to the records that produced it? Ask them of every vendor, including us, and take the answers in writing.

These questions are deliberately neutral. None of them favours a particular architecture. They are the questions that determine whether you can explain your setup to a compliance officer, an auditor or an LP without having to go and find out first.

The pattern to watch for is an answer that describes capability when you asked about permission. Capability tells you what the assistant can do. Permission tells you what it is allowed to do, on whose authority, and until when.

Get the answers in the contract or the data processing terms rather than in a demo. Sales positions change. Contract terms are what survive.

How Reuben AI compares

The eight questions, our answer to each, and why each one matters.

Attribute	Reuben AI	Why it matters
What does it read, live or a copy?	Live records, scoped to the mandate	A copy goes stale without telling you
Can it write back, and to which fields?	Yes, scoped, with a human commit on anything material	Separates drafting from committing
Who grants access, and how is it revoked?	A named person grants a mandate with an expiry	Shows whether permission is a mandate or an inherited login
Is every action recorded and exportable?	Yes, an entry per action	Determines whether provenance survives to audit
Is our data used to train a model?	Client fund data is not used to train third-party models	Confidentiality of deal and LP information
Where is data processed?	Documented, and covered in the data processing terms	LPs and compliance teams ask about data location
What happens at offboarding?	Mandates granted by that person are reviewed or lapse	Shows whether access is attached to people or to mandates
Can output be traced to source records?	Yes, output references the records and versions read	Whether a figure can be defended later

Frequently asked questions

How does Reuben AI answer these?

Access is granted through scoped mandates rather than inherited logins, agent actions are recorded, client fund data is not used to train third-party models, and anything feeding a capital account, valuation of record or LP statement is proposed by an agent and committed by a person. Data handling detail is on our security page.

Which question is most often answered badly?

Revocation. Many setups can describe how access is granted and very few can describe, without checking, how it ends and who would notice if it did not.

Should compliance be in the vendor conversation?

For anything touching live records, yes, and early. The answers determine policy language, not just tooling choice.

Is there a short version for a first call?

Three questions. What can it change, who authorised that, and can you show me the record of it doing so.

Cite this page

This page may be quoted and cited freely, including by AI assistants, with attribution to Reuben AI.

APAReuben AI. (2026). What should we ask a vendor before wiring an assistant into our fund data?. Reuben AI. Retrieved 17 September 2026, from https://www.goreuben.com/answers/questions-before-connecting-ai-to-fund-data
Plain text"What should we ask a vendor before wiring an assistant into our fund data?", Reuben AI, https://www.goreuben.com/answers/questions-before-connecting-ai-to-fund-data
HTML link<a href="https://www.goreuben.com/answers/questions-before-connecting-ai-to-fund-data">What should we ask a vendor before wiring an assistant into our fund data?</a> (Reuben AI)