WEALTH MANAGEMENT

Advisor copilots that hold up to a fiduciary review.

Pre-meeting briefings, suitability checks, email compliance and household reporting — built around the relationship manager and the fiduciary duty, not the chatbot UX.

Customer-hosted on Kubernetes, in your cloud or on premises. For enterprise teams managing AI across business units.

Book a demo

Or email sales@kosmoy.com.

Wealth management's productivity ceiling is the advisor-to-client ratio. Most senior advisors run 100–250 client households; the bottom quartile of clients gets a phone call once a year. Generative AI breaks the ceiling — but only if every output meets the firm's suitability and best-interest obligations.

Regulators have noticed. MiFID II suitability and the SEC's Regulation Best Interest both require documentation of why a recommendation fits a specific client at a specific time. FINRA Rule 2111 mandates the same in the US broker context. DORA layers operational resilience obligations on top. The AI Act will pull personalised investment AI into a high-risk classification once the consumer-credit rules are extended.

Kosmoy is the operating layer that lets the advisor run faster without the firm running blind. Every AI interaction is attributable to a specific advisor, client and recommendation moment — and is reviewable by compliance after the fact, not as an audit fire drill.

What this industry runs into.

Personalisation without losing the fiduciary thread

Generic LLM output has no concept of a specific client's risk profile, tax situation, IPS or restricted holdings. Without grounding in the client's record, AI suggestions drift toward best-effort marketing language.

Compliance review of outbound communications

Email, chat and meeting notes from the advisor need pre-send review for suitability claims, performance projections and cross-border solicitation. Surveillance teams can't keep up at scale; AI can pre-screen, but only if its judgement is auditable.

Data fragmentation

Client data lives in the CRM, the custodian, the planning system, the document store and the news feed. An advisor-facing agent needs read access across all of them — through controlled tools, not a flat API key.

Cross-border distribution rules

A US-resident client of a Swiss bank, a UK retiree with an Italian portfolio — every cross-border thread carries its own marketing and licensing rules that AI can either help enforce or quietly violate.

Regulatory landscape.

The regulations that shape AI in wealth management — and where each one bites on AI deployment.

DORADigital Operational Resilience Act· EU

Applies to MiFID II investment firms, including private banks and wealth managers. ICT third-party register, exit plans, incident reporting.

MiFID IIMarkets in Financial Instruments Directive II· EU

Suitability rules require documenting fit between recommendation and client profile. AI-drafted advice is in scope. Telephone and electronic recording rules cover advisor-AI interactions tied to order flow.

FINRA Rule 2111FINRA Suitability Rule· US

Reasonable-basis, customer-specific and quantitative suitability all apply to AI-generated recommendations.

SEC Reg BIRegulation Best Interest· US

Broker-dealers must act in the retail customer's best interest. AI-drafted recommendations must be evidence-grounded and disclosed conflicts of interest must surface in the AI output.

Investment Advisers ActInvestment Advisers Act of 1940· US

RIAs owe fiduciary duty. AI-generated content cannot reduce that duty; the human advisor remains responsible.

GDPR + national data lawsEU General Data Protection Regulation· EU

Wealth data is highly sensitive. Cross-border transfers, profiling under Art. 22, and data minimisation all constrain how AI sees client data.

Use cases, and what to measure.

Pre-meeting briefing

Advisor's calendar shows a 90-minute review with a client in two days. The agent reads the client's IPS, recent portfolio activity, life events captured in CRM notes, market context for their major holdings, and any open service requests, and drafts a 1-page briefing: agenda, recent changes, talking points, suggested next actions. The advisor walks into the meeting prepared, with consistent quality across the book.

What to measure

Advisor hours on meeting prep per week, and how often the advisor corrects the briefing before the meeting.

Powered byAgent Builder MCP Gateway RAG-in-a-Box

Suitability check on proposed allocation

Advisor proposes a portfolio rebalance for a client. The agent reads the client's IPS, risk tolerance, time horizon, tax situation, restricted holdings and investment policy, and returns: green/amber/red on suitability, the specific points where the proposal pushes against policy, and a rationale tied to the client record. If amber/red, the proposal cannot be sent without a documented override.

What to measure

Suitability exceptions caught when the proposal is drafted versus at compliance review, and audit prep hours per suitability case.

Powered byGuardrails Agent Registry Action Capsule

Email compliance pre-flight

Advisor drafts an email to a client recommending a specific fund. The agent reads it before send: checks for unauthorised performance projections, inappropriate certainty language, missing risk disclosures, cross-border issues, and prohibited terms. Returns a clean draft or specific edit suggestions. Surveillance reviews exceptions, not all messages.

What to measure

The share of outbound messages the pre-send review flags, how many of those flags surveillance confirms, and how many issues still reach clients.

Powered byGuardrails AI Gateway Insights Dashboard

Annual review pack

Annual review prep used to mean an analyst building a 20-slide deck for each household. The agent reads holdings across the household, performance attribution, fees paid, life events, IPS adherence, and drafts the full pack. Advisor reviews and adjusts. Every pack follows the same structure.

What to measure

Analyst hours per household pack, the share of households that get a full annual review, and how much the advisor changes in each draft.

Powered byAgent Builder RAG-in-a-Box AI Gateway

Tax loss harvesting suggestions

End of fiscal year, the agent runs across the household, identifies positions in loss territory, cross-references the client's restricted list and wash-sale window, and suggests harvest pairs. The advisor reviews; decisions and trade authorisations stay human.

What to measure

Harvestable losses identified versus harvested, wash-sale conflicts caught before the trade, and whether each decision has a documented rationale on file.

Powered byAction Capsule MCP Gateway

Agent governance

Where wealth management agents need extra discipline.

Wealth management agents are advisor-attached: they read sensitive client data and draft client-facing material. The risk profile is suitability and fiduciary duty, not throughput. Kosmoy treats every advisor copilot as an agent in the Agent Registry — owner, model, retrieval scope, allowed actions, supervision regime. The Action Capsule prevents an advisor copilot from reaching beyond its scope; if it tries to query restricted-list trading instead of just reading it, the call fails at the gateway.

Supervision is built into the data the platform produces. The Insights Dashboard surfaces patterns: which advisors override the suitability agent, which agent versions are correlated with later complaints, which client segments produce the most flagged outbound messages. Compliance moves from spot checks on transcripts to pattern-based oversight on operational data.

Chatbot use cases

Chatbots, by surface and risk class.

Wealth client-facing chatbots are usually narrow — account access, holdings questions, statement download — because anything advice-adjacent triggers regulatory exposure. Internal chatbots, on the other hand, multiply quickly: planning, ops, compliance, marketing, training.

Client portal chatbot

Self-service for balance, transactions, statement download, secure messaging. Strict guardrails prevent any phrase that could be construed as advice; advice triggers a hand-off to the advisor.

Advisor onboarding agent

New hires ask 'how do I open a discretionary mandate for an Italian-resident client trading European equities?' and get a citation-grounded answer from policy and procedure docs.

Compliance helpdesk

'Does this gift count as inducement under MiFID II?' — agent reads the firm's inducement policy, summarises, defers to compliance for an actual decision.

Planning team agent

Planners ask the agent for tax-aware projections, retirement scenarios and trust structuring options. Outputs cite source policy documents and never substitute for the planner's professional judgement.

How Kosmoy fits.

Kosmoy fits the wealth firm because the platform's centre of gravity matches the firm's: a small number of high-value relationships per advisor, where every interaction matters and every record needs to survive a regulatory or client review years later. The AI Gateway records every model call by advisor and client. The Agent Registry tracks every advisor copilot. Insights surface patterns over time.

Deployment is single-tenant in the firm's Kubernetes cluster — including air-gapped modes for the most sensitive private-banking client books. No client identifying data leaves the perimeter. No model provider sees client portfolios. The firm controls the credentials, the retrieval set and the agents.

Module questions, answered straight.

Will the suitability agent replace our advisors?

No. The agent prepares, drafts and pre-checks; the advisor decides. Every recommendation is attributable to a human, every override is logged. The platform is designed around augmenting fiduciary capacity, not substituting it.

How is client data isolated between advisors?

Per-advisor identity flows through the AI Gateway. The retrieval scope is bounded to the books the advisor is authorised to view. An advisor cannot query another advisor's clients via the agent any more than they can query the underlying CRM directly.

Can the agent write trades directly?

By default, no. Agents draft, suggest and pre-check. Where the firm chooses to let an agent place trades (model portfolios, rebalancing within IPS bands), the agent runs in an Action Capsule with explicit allowed-action scope and pre-flight authorisation on every action.

How do we handle cross-border restrictions?

Each agent's scope is bound to the client's residency and the advisor's licensing. Cross-border solicitation rules (FINRA, FCA, FINMA, AMF) are encoded as policy in the Guardrails layer. Out-of-scope outputs are blocked at composition, not on the way out the door.

Enterprise proof

Built for environments where AI cannot become another unmanaged SaaS silo.

Kosmoy works with regulated and operationally critical organizations, including Banca d’Italia and Leonardo.

Runs in your Kubernetes
Azure · AWS · GCP · on-prem
LLM · MCP · A2A
No mandatory vendor-hosted control plane

Scale every advisor without losing the fiduciary thread.

See how the AI Gateway, Agent Registry and Action Capsule keep wealth-management AI inside MiFID II, FINRA and DORA boundaries.

Book a demo

Or email sales@kosmoy.com.