AI automation · Financial services AI for banks and financial advisers: automating documents, complaints and reporting under regulation

Typical workflows in financial services

These are typical processes, not client projects. In each, the AI prepares, and an adviser, underwriter or compliance officer decides.

1. Application packs and document checks

Applications arrive as bundles of forms, statements, payslips, accounts and identity documents, often incomplete.

• Input: the application and its attachments.
• AI prepares: recognises each document, reads the values, checks the pack against the checklist for the product, lists missing or unreadable items and drafts the request to the applicant.
• Staff check: completeness decisions and every deviation. Identity verification, sanctions screening and the credit decision are done by the regulated systems and the people responsible for them.
• Output: a complete file in the case system and a draft request for missing documents.

2. Correspondence and complaints

Customers write by email and letter. Some of it is routine, some is a complaint that starts a regulated process.

• Input: incoming messages and letters, the customer record.
• AI prepares: sorts by topic and urgency, flags wording that signals a complaint or vulnerability, drafts replies from your approved wording and, for complaints, assembles the facts and the timeline from the record.
• Staff check: every complaint and every reply that involves a decision or money. Complaints must be investigated competently, diligently and impartially, and people do that.
• Output: sorted correspondence, reply drafts and complaint files with a clear timeline.

3. Adviser file notes and suitability reports

After a client meeting, the adviser must record the client’s circumstances, objectives and the reasons for the recommendation.

• Input: notes or a recording of the meeting, the fact find and the research output.
• AI prepares: a draft file note and a draft of the report that follows your template, with the client’s own words where they matter and gaps marked.
• Adviser checks: every statement and the recommendation, because the advice and its record are the adviser’s responsibility.
• Output: the note and the report as drafts in your system.

4. Management and regulatory reporting

Figures sit in several systems, and the commentary is written under time pressure.

• Input: extracts from core, risk and finance systems, last period’s report.
• AI prepares: a report with numbers calculated by the systems and not by a language model, commentary on the largest movements and a list of data points that look wrong.
• Finance and compliance check: the figures, the commentary and every statement that goes to a regulator or the board.
• Output: a report draft with sources for each figure.

What stays with people, and what we do not do

• Credit decisions: systems that evaluate the creditworthiness of natural persons or establish their credit score are high-risk under Annex III of the AI Act, with an exception for fraud detection. We do not build them, see risk categories.
• Advice and recommendations: the regulated activity of the adviser.
• Anti-money-laundering judgements: screening, monitoring and suspicious activity reports remain with compliance and the specialised systems. The AI never decides that a transaction is suspicious.
• Marketing statements: anything that counts as a financial promotion must be fair, clear and not misleading and goes through your approval process, however it was drafted.
• Decisions about staff: we do not build systems that rate or select individual employees or applicants.

Rules that matter

The table lists the main supervisory sources as an orientation. Which rules apply depends on your licence, your products and your markets.

On top of this, Art. 22 GDPR limits decisions based solely on automated processing that significantly affect a person, and Art. 28 requires a data processing agreement with every provider. Art. 4 of the AI Act asks for sufficient AI literacy among staff, and Art. 50 requires AI chatbots to identify themselves. This overview is general information, not legal advice. Your compliance officer and your supervisor’s current texts decide what applies to your firm.

Systems you probably use

Banks and lenders work with core banking and loan platforms such as Temenos or Finastra, with CRM systems such as Salesforce and with document and case management. Advice firms typically use back-office platforms and CRM tools for the fact find, suitability and client files. We connect through the interfaces these systems offer (API, export, email or a watched folder) and work alongside them where none exists. The names are for orientation only and do not imply any partnership.

How AI assistants find financial firms

Customers ask ChatGPT, Gemini or Google’s AI Overviews for “an independent financial adviser for retirement planning in Edinburgh” or “a lender for small business equipment finance”. Assistants tend to point to registers, directories and regulators, because regulated status matters.

• Visible regulated status: the firm’s authorisation and reference number, matching the entry in the FCA Register or the equivalent register.
• A page per service: what you offer, who it is for, how you are paid and how the first conversation works.
• Consistent details: the same name, address and services on the website, the register, the Google Business Profile and directories such as Unbiased or VouchedFor.
• Fair content: guides that explain without promising outcomes, in line with the rules on financial promotions.

Is it worth automating? Eight questions

1. Do application packs regularly arrive incomplete, and does chasing take time?
2. Does someone read and sort incoming mail by hand?
3. Do complaints take long to start because the facts have to be gathered first?
4. Are file notes and reports written from memory after the meeting?
5. Does assembling a report mean copying figures from several systems?
6. Do staff retype the same details into several systems?
7. Is there a written rule on what may go into AI tools, and has compliance agreed it?
8. Can you name one task where a draft that needs checking would already save time?

Three or more yes answers usually point to a workflow worth testing. Question 7 is the prerequisite.