Client inquiry management with AI agents that assemble every reply from cited sources, answer your questions about it, and wait for a person to approve
Client requests answered from the source systems, every figure cited and cleared.
See one case, screen by screen ↓Client inquiry management is how an asset manager receives, answers and records the requests its institutional clients, consultants and intermediaries send: holdings, performance, fees, statements, fund data, due-diligence and AI-use questions. Client Inquiry Management adds seven AI agents for asset management client service that split each request into its questions, fetch every answer from its system of record with an as-of date, check it against the disclosure policy and the contact's entitlements, and draft a cited reply that a person checks or approves before it goes out.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Client inquiry management at a global asset manager can mean thousands of requests a month from pension plans, insurers, endowments, sovereign funds, their investment consultants and intermediaries: holdings, performance, fees, statements, tax documents, fund data and, more and more, how the manager uses AI. Each one arrives in a shared mailbox, the client portal, a phone note or an intermediary platform, and client inquiry handling starts with an associate working out who is asking, which client it concerns and what that contact may receive, then looking up the answer in portfolio accounting, the performance system, billing, the fund administrator and the document store. A simple lookup takes 1 to 3 minutes; a request that crosses systems can take an hour.
The risk is in what goes out. A mid-month holdings list sent to one investor before all the others is selective disclosure. A preliminary return sent as final, a consultant whose authorization lapsed still receiving client data, benchmark constituents passed on against the data license, or a wire change from a look-alike domain acted on are not service slips. The disclosure rules live in policy PDFs and in senior people's heads, compliance is asked case by case by email, and response clocks are reported after the client has already waited.
Estimated split for a sensitive request with holdings, fees and performance, by hand and with the solution. What is left is mostly the wait for the associate's check and the director's approval; the agents' own work on the sample request took minutes.
Seven agents split each request, fetch every answer from its source, check what may go out and re-check every figure; people approve.
Thursday, October 15, 2026, at Quillfield Capital, the mid-October peak after quarter end: Q3 fees were billed on October 7, official Q3 returns were final on October 12, and the September 30 top 10 went to every investor at 8:00 a.m. At 10:04 a.m. a consultant writes on behalf of Calder Valley PERS, a Tier 1 public pension with $1,842.6 million in the Global Equity Focus trust. Here is how the client service team works with the agents on it, screen by screen, in the working solution.
At 10:09 a.m. Nadia's Today screen shows 23 replies waiting for her approval, the closest due at 11:18 a.m., 9 requests close to the clock (3 of them Tier 1), 1,404 answered this month at 97.6% on time, and 14 disclosure holds open for compliance to decide. Beside it, what the agents did since 7:00 a.m.: 64 requests read from four channels, 2 unknown senders held for a person, not guessed, and 151 blocks pulled from 9 systems. She asks the agents what could miss the clock; the Triage agent names the three closest, Shirakawa Mutual Life with 14 minutes left first, and filters the queue to them.
Tie-out agent: “Caught a transposed fee before it reached Merrow County: the draft said $412,600, the invoice $421,600.”
Marisol Vance, Senior Consultant at Harrowgate Consulting Partners, wants four things for Calder Valley's board pre-read on Monday: the trust's top 10 holdings as of Friday, October 9; the Q3 management fee and effective rate; the Q3 net return against the benchmark; and a paragraph on how Quillfield uses AI. The Intake agent finds her a known, entitled contact, authorized for reporting, performance and fees to January 31, 2027, with no duplicate. The Triage agent highlights the four questions in the message and splits them out.
The Triage agent sets a Tier 1 clock of 4 business hours, due 2:04 p.m., which escalates at 50% to the owner and at 75% to the director. The Fetch agents pull the trust's holdings as of October 9 and September 30 and the official Q3 returns from portfolio accounting and performance measurement, and the Q3 invoice and an approved answer on AI use from billing and the answer library. Each answer arrives as a block stamped with its system, record and as-of date. Replay the agents shows the run step by step.
Triage agent: “Split into 4 questions; set a Tier 1 clock: 4 business hours, due 2:04 p.m.”
Invoice QF-INV-2026Q3-04418, billed October 7: on Calder Valley's average net assets of $1,839.3 million, the tiered schedule charges 0.45% on the first $250 million, 0.35% on the next $750 million and 0.28% on the $839.3 million above $1 billion, $6,100,040 a year or 33.2 basis points. One quarter is $1,525,010, which is the invoice. The Disclosure agent clears it under rule HD-7: a client's own fees may go to the client and to a consultant authorized for fees.
The Fetch agents are read-only, and compute only stated arithmetic like an effective rate, with the working shown.
Rule HD-2 releases a commingled trust's month-end top 10 to all investors at once, 15 days after month end; interim lists go only to a recipient with a signed confidentiality agreement, which Harrowgate does not have, and the interim list includes a position still being built. In the reply instead: the September 30 top 10, released to every investor at 8:00 a.m. today, 33.01% of the trust. The Q3 return is cleared with conditions under PF-2: 4.37% net of fees (4.45% gross) against 3.91% for the benchmark, period and benchmark named.
The Disclosure agent can hold a block; it never releases one. Only a compliance reviewer releases.
Leo Marchetti prepared the reply and left his note in the conversation: he agrees with the September 30 list and kept one sentence on the policy so Marisol can tell the board why. Nadia asks who else received the September 30 list; the Fetch agent answers that it went out at 8:00 a.m. on the client portal to all 214 investors in the trust, 37 of whom have opened it so far, the same file attached to this reply, and cites the release. Then she types “Tell Ethan we held it”, and the Drafting agent sends Ethan Brooks, the relationship manager, an internal note, not to the client, on the audit trail.
Drafting agent: “Sent Ethan Brooks an internal note (not to the client): ‘We held the Oct 9 list under HD-2 and sent the Sep 30 top 10; Marisol may ask on Monday.’ It is on the audit trail.”
The Drafting agent assembled the reply from the four blocks, each paragraph carrying its block number, with the September 30 list attached and the AI paragraph taken verbatim from approved answer AI-USE-03. The Tie-out agent re-checked 11 of 11 figures and dates against their blocks. Performance and a hold make the reply sensitive, so it needs a director who did not prepare it: Nadia approves, signed with her name, role and the time through single sign-on. The reply goes to Marisol Vance with Ethan Brooks copied, and the response clock stops at 6 minutes against a 4-hour Tier 1 target.
“4 blocks · 1 attachment · tie-out 11 of 11 · held block replaced by its alternative. Prepared by Leo Marchetti; you are the second person.”
The audit trail shows each step of REQ-104882 with who, what, when, the reason and the evidence: the Intake agent's read against authorization AUTH-CV-2026-01, the Triage agent's split under clock rules v2026.4, the hold under HD-2, the Tie-out agent's 11 of 11, Leo as maker and Nadia's approval as maker-checker. Each entry carries a hash of the one before, the chain can be verified on screen, and an evidence pack can be exported for an examination.
The same request queue, seen by five of the people who carry it: what the day looked like, and what it looks like now.
Seven agents split each request, fetch every answer from its source, check what may go out and re-check every figure; people approve.
Reads the shared mailboxes, client portal requests, phone-call notes and intermediary platform messages; identifies the sender, the client, the account and the contact's entitlements; detects the language; finds duplicates; and logs the request in the CRM.
Classifies the request, splits it into its questions, sets the response clock by the client's tier and the hardest question, spots complaints, security risks and ambiguity, and routes questions that need investment judgment to a specialist.
Pulls holdings, transactions, returns, attribution and capital accounts from portfolio accounting, performance measurement, the fund administrator and the transfer agent, and stamps every block with its system, record and as-of date.
Pulls fees and invoices, statements, tax documents, client reports, prospectuses and approved answers, in any language.
Checks every block against the holdings and performance disclosure policy, the contact's entitlements, data licenses and preliminary-number rules; holds what is early or not permitted and proposes what may go out now.
Assembles the reply in the client's language and the firm's tone, each sentence tied to its block, with required wording and attachments, and proposes new approved answers from specialists' replies.
Re-reads every figure, date and attachment in the draft against its block, on a different model family and prompt from the Drafting agent, so the check is independent.
Check, approve, answer, release. The agents propose; a named person decides.
On every screen, ask about a request, a hold or the queue in plain words; answers name their source. An instruction to change a draft comes back as a proposed change to apply or keep, re-checked before it applies.
Why can't we send October 9?
Rule HD-2 releases a commingled trust's month-end top 10 to all investors at once, 15 days after month end. An October 9 list would reach Calder Valley before the trust's 213 other investors, and Harrowgate has no confidentiality agreement for this trust. The interim list also shows a position still being built.
Show the fee working
Tiers on average net assets of $1,839.3 million: $250M × 0.45% = $1,125,000; $750M × 0.35% = $2,625,000; $839.3M × 0.28% = $2,350,040. Total $6,100,040 a year, 33.2 bps; one quarter is $1,525,010, which is the invoice.
What could miss the clock today?
9 requests have under a quarter of their time left, 3 of them Tier 1. Closest: Shirakawa Mutual Life (14 min left), Thamesbridge LAPF (17 min left), Shirakawa Mutual Life (23 min left). I filtered the queue to them.
How many requests were about AI?
58 requests this month were about AI use (3.6% of 1,618), up from 31 in September; 52 were answered with approved answers in under an hour.
13 screens from the working solution, on its sample data. Pick one to see it large.
Clocks, thresholds and routing are settings, versioned and logged. The administrator proposes a change; the Head of Client Service approves it.
| Setting | Default | Choose from |
|---|---|---|
| Tier 1 clock | Acknowledge in 15 min · routine and sensitive 4 business hours · specialist 1 business day | Set per tier: Tier 1, Tier 2, Tier 3 and intermediaries; business hours 8:00 a.m.–6:00 p.m. New York |
| Tier 2 and Tier 3 clocks | Tier 2: acknowledge in 1 hour · reply in 1 business day · specialist 2 business days | Tier 3: 4 hours · 2 business days · 3 business days; intermediaries: 1 hour · 1 business day · 3 business days (acknowledge · reply · specialist) |
| Escalation ladder | 50% to the owner · 75% to the director · breach to the Head and the relationship manager | A clock pauses only for a documented wait on the client |
| Routine reply | Fetch and Tie-out agents at 95% confidence or more | Below it, the block goes to the associate unconfirmed |
| Classification | Below 85%: a person classifies | Thresholds are set per agent and versioned; each change is approved by the Head of Client Service |
| What makes a reply sensitive | Performance, holdings, a complaint, a hold, an edited figure or non-standard text | Sensitive replies need a director, never the preparer |
| Bulk check and send | 5% of bulk sends go to a director | Routine replies only |
| Data and retention | Replies, blocks and evidence kept 7 years | US, EU, UK and Singapore regions; EU clients' data never leaves the EU region; legal hold |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (KPMG, Gen AI in asset management, 2025 — 1–3 minutes per simple client inquiry, up to an hour for complex ones across systems, 80–85% response accuracy before AI; SEC press release 2024-98, August 14, 2024 — $392.75 million against 26 firms for off-channel recordkeeping failures; SEC Release 33-8408 on selective disclosure of portfolio holdings). The 15-day and 30-day holdings lags are the demo firm's own policy, in line with common industry practice, not a rule. People, firms, clients, funds and figures named on this page are fictional — characters and sample data in the working solution.
Client inquiry management is how an asset manager receives, answers and records the requests its institutional clients, consultants and intermediaries send: holdings, performance, fees, statements, fund data, due-diligence and AI-use questions. Client Inquiry Management adds seven AI agents for asset management client service that split each request into its questions, fetch every answer from its system of record with an as-of date, check it against the disclosure policy and the contact's entitlements, and draft a cited reply that a person checks or approves before it goes out.
The Intake agent identifies the sender and what that contact may receive; the Triage agent splits the request and sets a clock by the client's tier; two Fetch agents pull holdings, returns, fees, statements and documents; the Disclosure agent clears or holds each answer; the Drafting agent writes the reply; and the Tie-out agent re-checks every figure. On the sample request, four questions became an approved reply in 6 minutes.
The Disclosure agent checks every block against the firm's holdings and performance disclosure policy. An interim holdings list is held unless the recipient has a confidentiality agreement and every investor may have it on the same terms, and the agent proposes an alternative that may go out now, such as the month-end list already released to all investors. Only a compliance reviewer can release a hold.
No. This is client service automation with people in charge: the agents draft and check, and a person sends every reply. You can ask the agents about any block or tell them what to change, and an instruction to change a draft comes back as a proposed change to apply or keep. An associate checks routine replies, a director who did not prepare it approves every sensitive reply, and the agents never release a hold, act on an instruction change, close a complaint or write a figure without a source block.
Yes. There are no off-channel replies: answers go only from the shared mailboxes and the client portal, and every reply, block, approval and hold is on a hash-chained audit trail kept for 7 years, longer than the periods in SEA Rule 17a-4 and Advisers Act Rule 204-2. Internal audit can verify the chain and export an evidence pack.
Yes. Requests in German, French and Japanese are read and answered in the client's language, with the English beside it for the associate to check. A figure taken from a fund document, such as the SFDR classification in a German prospectus, is cited to its page.
An account-instruction change is never handled in a reply: it is held at intake and routed to Client Operations for a call-back to an authorized signer on the number on file, confirmed by a second person, and the agents draft nothing to the sender. A complaint inside a request is split out and logged as its own complaint for compliance, with its own clock, and the reply acknowledges it.
The Agentic Solution Engine builds and deploys it from your requirements and documents: your disclosure policy, client tiers, approved answers and read-only connections to your CRM and systems of record. It goes live once every quality gate has passed. In the working solution's go-live plan, the agents first draft last month's 3,212 requests in shadow mode and nothing is sent; 2,849 drafts (88.7%) matched what the team sent. Then come two weeks of shadow running, routine replies with an associate checking each one, and sensitive replies with a director approving each one.
Every consultant profile current each quarter, every figure tied to its source.
1-min demoPitch Book CreationFinals decks the same day, with every number stamped to its source.
1-min demoInstitutional Client OnboardingWon mandates funded in weeks, not months, with every signer and bank detail proven.
1-min demoWe'll run Client Inquiry Management on a sample of your own past requests, your disclosure policy and your approved answers.