Still asking clients for documents they already sent?

A client sends a document to the person managing their relationship. A few days later, someone else at the firm asks them for it again.

From the client’s side, that can feel careless. Inside the firm, there may be a fairly ordinary explanation. The document is in another inbox. It arrived under a different file name. Someone reviewed it, but the record was never updated.

These are useful places to look when a financial services firm wants to introduce AI. The work is familiar, the frustration is recognizable and the team can usually show you where things become difficult.

Before choosing a tool, it helps to understand how information moves through onboarding today.

Information arrives in pieces

An intake form might begin the process, but it rarely contains everything the team will eventually need.

A client explains something on a call. Supporting documents follow by email. A colleague asks for clarification. Compliance raises a question. A bank requests an additional item later in the process.

The person closest to the client can often hold those details together. They remember what was discussed, know which attachment matters and understand why a request is still outstanding.

That becomes harder when a colleague needs to take over.

They need to work out what the firm has received, which details have been checked and what still requires action. If those answers are spread across inboxes, folders and private notes, they may have to reconstruct the history before doing anything useful.

AI could help with parts of that work. First, the firm needs a clear place for the answers to live.

Start with the client record

A useful client record should let an authorized colleague understand the current position without asking the relationship manager to explain everything again.

That record may already exist in the firm’s CRM, onboarding platform or case-management system. An assessment should begin there, before considering another system for staff to maintain.

Take one recent onboarding file and ask:

• What information and documents have we received?

• Where did each item come from?

• Has someone checked it, or is it still awaiting review?

• Are there conflicting details or unresolved questions?

• What needs to happen next, and who owns that action?

The distinction between receiving something and accepting it matters. A document can be attached to a record while still requiring review. A detail extracted from an email can be useful without being confirmed.

Keeping those states visible helps the next person understand what they can rely on.

Where could AI help?

Once the team agrees how the record should work, there are several bounded tasks worth assessing.

Preparing a summary of correspondence. An AI assistant could summarize approved emails associated with a case, highlighting information received and questions that appear unresolved. Reviewers should be able to open the underlying messages and check the summary.

Drafting updates to the record. AI could suggest details to add from a document or email. A member of staff would confirm the source and accuracy before treating the entry as verified.

Preparing a focused follow-up. Using the reviewed record and the firm’s applicable checklist, AI could draft a request for outstanding information. The person responsible would check that the request is necessary, accurate and has not already been answered elsewhere.

These are possible uses to test, rather than capabilities to assume every tool will handle well. Their usefulness depends on the information available, the systems involved and how much checking the output requires.

Keep the review boundary clear

Onboarding information can be sensitive. Any proposed tool needs to fit the firm’s requirements for access, confidentiality, retention and permitted use before client material is introduced.

The implementation also needs clear limits.

An assistant preparing a follow-up may only need access to the relevant case and approved correspondence. It should not automatically gain access to every client folder or internal mailbox.

Missing or conflicting information should remain visible. If two documents contain different details, the system should bring that discrepancy to a reviewer rather than choose an answer without telling anyone.

Suitability assessments, acceptance decisions and compliance approvals remain with the people responsible for them. An uploaded document alone should never cause the system to mark a requirement as satisfied.

These boundaries are part of the workflow design. They need to be understood by the team using it, including what to do when an output looks plausible but cannot be supported by its source.

Test whether it makes the handoff easier

A first test could focus on one stage of onboarding, such as preparing the outstanding-information follow-up.

Start by understanding how the team does that work today. How long does it take to establish the current position? Where do people search? How often does the draft need correction before it can go to the client?

An early demonstration can use fictional cases. Any testing with real client information should happen only after the appropriate access and handling arrangements are in place.

Include awkward cases in the test: an attachment with an unclear name, a revised document, a question answered in a later email or two sources that disagree.

Then assess the whole task, including review. A draft produced quickly may offer little benefit if a colleague has to spend longer checking it than they would have spent preparing it themselves.

The practical question is whether an authorized colleague can open the record, understand the case and take the correct next step with less searching and fewer avoidable mistakes.