Model portfolio rebalancing as a conversation with eight agents: every model change taken to every account, each difference explained and cited, every group decided by a portfolio manager
Every model change reaches every account the same day, within each client's own rules.
See one case, screen by screen ↓Model portfolio rebalancing takes a model's target weights into every account that follows the model, when the investment committee changes the model or when accounts drift outside their bands. Model Portfolio Rebalancing gives the model portfolio management team eight agents that read each account's restrictions from its documents, fan the change out to every account grouped by why each must differ, choose tax lots, size buys to cash and re-check every proposal. The portfolio manager questions them in plain words, then approves or sends back each group; a second person releases the order file.
AI in asset management: a buyer's guide to agentic AI for investment managers →
Updated
Model portfolio rebalancing looks simple from the committee room: the model changes, so every account changes. In practice no two accounts take the change the same way. One client forbade defense suppliers in a letter, one account owner works at the company being sold and is in a blackout, a group of accounts bought the stock 19 days ago and would create a wash sale, a trust's policy statement caps bonds at 40%, and a charitable trust has a distribution due Friday. Those rules live in advisory agreements, investment policy statements, trust documents, client letters and CRM notes. Operations keys a subset of them into the rebalancing software by hand, and the software knows only what was keyed.
So each change becomes a grid of thousands of orders filtered by error code, worked by a portfolio manager and a rebalancing analyst, with the reason an account differs from the model rarely written down. The risk sits in what never reached the table: a forbidden security bought, a loss disallowed because a spouse's IRA at another custodian bought the same stock, a buy that leaves an account overdrawn on T+1. Rule 3a-4, the safe harbor discretionary model programs rely on, requires that each client can impose reasonable restrictions on how the account is managed.
Estimated hours for one model change across about 1,200 accounts, by hand and with the solution.
Eight agents read, plan, tax-check, restrict, size and double-check every account; the PM approves.
Monday at 4:30 p.m. Quillfield Private Wealth's Model Portfolio Committee approved MC-2026-118: Balanced Growth 60/40 goes from version 14 to 15. Sell 2.0% Arclight Semiconductors, buy 2.0% of the US Treasury 3.500% note due 09/30/2028, in 1,240 accounts worth $1,762.0M, by Wednesday at 4:30 p.m. Here is what happens next, screen by screen, in the working solution: the agents work overnight and explain themselves beside the change, the portfolio manager questions them in plain words, every answer cites its source, and people sign every release.
Maren's team runs 46 models for 15,384 discretionary accounts, $21.8B, across New York, London and Singapore. Today opens on four numbers: 1,226 accounts waiting for her approval, one change the agents are still working, 214 accounts outside their bands and 47 restrictions read this week. Under What the agents did overnight, each line cites its page: wash sales avoided in 87 accounts, one proposal that broke a trust's rule caught and corrected, a new client letter in Chinese read.
Needs you: “Balanced Growth 60/40: 1,226 accounts ready for you.”
The change sits at the top; beneath it, 1,240 accounts in five groups, each with a count, its value, a plain reason and the agent behind it: Straight through 1,108, Tax lot adjusted 87, Restricted 25, Outside its policy band 6, Short of cash 14. Beside it, the agents explain themselves. The Checker re-ran every proposed portfolio on a second model and disagreed on one: the Delacroix Family Trust (31-0915) may hold government bonds only through funds, so the account moved to Restricted and buys the approved substitute fund instead.
Rebalance Planner: “132 must differ, for four reasons, grouped on the left. Nothing has gone to trading.”
Maren types “Why is 40-1187 held?” in the conversation beside the change. The Restriction agent answers at 99% confidence, with its sources: James Whitcombe is an Arclight insider, the couple's letter of March 4, 2024 forbids any Arclight trade without written pre-clearance, and Arclight's Q3 blackout ends November 2. So no trade: 755 ARCL shares ($99,358) stay and the account sits 2.0% over the model. On the account screen the Order agent's rationale is stored as the order memorandum for a discretionary account.
“I asked Priya to request pre-clearance and set the account to re-propose on November 2.”
In 87 accounts Arclight was bought on October 1 or, in 3 households, in a spouse's IRA elsewhere, so the Tax agent sells the October 1 and long-term lots and holds the 2025 loss lots until the window closes, keeping $418,600 of losses allowable. Erik and Ingrid Bergström (14-0562) show why: the firm's lot order would sell the February 2025 lot at a $6,160 loss, but Ingrid's IRA at another custodian bought 120 shares on October 6. Tax lot selection sells 286 of the 300 shares bought in March 2023 instead, a long-term gain of $10,153, and keeps the 2025 lot until November 6. Wash sales created: 0.
Why the 2025 lot stays: “A loss here would be a wash sale with the IRA purchase on October 6, lost for good inside an IRA.”
Maren tests an instruction in plain words. The Tax agent answers as a what-if: with a $50,000 gain budget per account, 41 of the 1,108 straight-through accounts would sell Arclight only in part, realized gains across the change would fall from $6.82M to $5.97M, and those accounts would stay 0.3%–0.9% over the model, inside the band. The committee asked for the full change, so the proposal stays as it is unless she applies it.
Two buttons, her choice: “Keep the proposal” or “Apply to the straight-through group”.
Mercedes Ortega Vidal, a client of the London office, wrote that she does not want her portfolio in companies that make weapons or components for military use. The document screen shows the Spanish original beside the English, the clause highlighted. At 71% confidence the Restriction Reader did not guess: Arclight earns 14% of its revenue from military programs, so the wording needs a threshold. Priya, in client service, proposed 10% or more of revenue, and the rule waits for compliance to approve it.
Restriction Reader: “I asked Priya rather than guess.”
Maren wants the tax group re-read against the newest letters before she signs. She sends “Tax lot adjusted” back with the reason code “Re-check against the latest documents” and an instruction kept on the record: “Re-read the client letters received since Monday before you propose.” The group shows “Sent back · agents re-planning” and cannot be approved. The Tax agent re-plans the 87 accounts, the Checker re-runs every proposed portfolio on the second model, and the group comes back to her.
Checker: “I re-ran all 87 proposed portfolios on the second model: every one passes every rule, band and list. Tax lot adjusted is ready for your decision again.”
Maren approves the four ready groups together: 1,226 accounts, 2,436 orders, $34.42M each side, reason code “Implements committee change”, with an optional note kept on the record. The 14 short-of-cash accounts wait for Priya to confirm the withdrawal dates by 2:00 p.m.; the Navarro Charitable Remainder Trust, for one, buys $11,200 of Treasuries instead of $62,400 to keep cash for a $180,000 distribution on Friday.
“The Checker agreed on every proposal. Each account keeps its rationale as its order record. Over $25M: the Head of Model Portfolios releases it.”
Victor releases order file OF-118-1 as the second signature. The Order agent builds it in five blocks for Theo Brandt's desk, each with its instruction: sell 258,511 Arclight shares across 1,202 accounts, sell 4,630 Halvern for the 13 defense-screened accounts, buy $34.26M of the Treasury note, and small buys of the funds-only trust's substitute fund and of a US equity fund for the six accounts at their bond cap, client names masked. Once the short-of-cash group is confirmed and released too, the report counts 1,229 accounts to trade, 11 held with a dated reason, 0 restricted-security trades and 0 wash sales created.
Order agent: “Every block traces to approved accounts.”
The audit trail lists every action by people and agents with who, what, when, why and the evidence, each entry hash-chained to the last: the 1,240 rationales written as Rule 204-2 order records, Priya's proposed rule for the Spanish letter, the answers to Maren's questions, the send-back and the Checker's re-run, her approval, Victor's release and the order file built for the desk. One click exports the rebalance as an evidence pack.
Order agent: “Today's 22 entries are chained and verified. I can find any decision with its evidence, or assemble the evidence pack for a rebalance.”
The same model change, seen by five of the people who carry it: what their week looked like, and what it looks like now.
Eight agents read, plan, tax-check, restrict, size and double-check every account; the PM approves.
Reads advisory agreements, investment policy statements, trust documents, client letters, emails and CRM notes in their own language, and turns each instruction into a rule with its clause, page and confidence. At 90% or above it proposes the rule for one-click confirmation; from 50% to 90% it asks client service; below 50% a person reads the document.
Compares every account with its model each morning at 6:00 a.m. against the tolerance bands, flags holdings, asset classes and cash outside them, and proposes a drift rebalance.
Fans a committee's model change out to every account in the model, proposes trades per account with the Tax, Restriction and Cash agents, and groups the accounts by why each differs.
Handles tax lot selection for every sale, avoids wash sales anywhere in the household, IRAs and held-away accounts included, avoids short-term gains outside an account's gain budget, and explains every lot.
Applies every rule on the account, the firm restricted list and the insider list, uses only approved substitutes, and holds what may not trade with the reason and the date it can be re-proposed.
Sizes every buy to available cash after pending withdrawals, policy cash floors and T+1 settlement, and asks client service when a date or amount is not certain.
Re-runs every proposed portfolio on a second model, independently, against every rule, band and list, tries to break each proposal, re-runs any group a portfolio manager sends back, and escalates every disagreement to the portfolio manager.
Writes each account's rationale with its sources as the order memorandum, builds the order file in blocks after release, and reconciles fills and allocations with portfolio accounting and the custodians.
Approve, release, confirm rules. The agents propose; a named person decides.
Ask about any model change, account, restriction or trade, or tell the agents what to do. Every answer carries its source, an instruction comes back as a what-if until a person applies it, and a group you disagree with goes back to its agent with your instruction.
Why is 40-1187 held?
James Whitcombe is an Arclight insider. The couple wrote on March 4, 2024: “Please do not buy or sell Arclight shares in our joint account without written pre-clearance from Arclight's legal department.” Arclight's Q3 blackout ends November 2. So I propose no trade: 755 ARCL shares ($99,358) stay, the account sits 2.0% over the model, and the reason is recorded.
What does the tax group realize?
The 87 accounts realize $712,400 of long-term gains and $296,800 of short-term losses on the October 1 lots, which stay allowable. Holding the 2025 loss lots to November 2 keeps $418,600 of losses allowable, and no lot held under a year is sold at a gain: $1.27M of short-term gains avoided. Residual overweight in Arclight averages 0.13%, inside the 1.0% band.
Use a $50,000 gain budget
As a what-if (not applied): 41 of the 1,108 straight-through accounts would sell Arclight only in part. Realized gains across the change fall from $6.82M to $5.97M; those 41 accounts stay 0.3%–0.9% over the model, inside the band, and join Thursday's drift run. The committee asked for the full change, so I kept the proposal as it is.
Why did you ask instead of applying?
My confidence was 71%, under the 90% threshold. “Components for military use” could mean any supplier with a military customer or only those that depend on it; the answer decides whether Arclight (14% of revenue) is excluded. A person sets the threshold; compliance approves it.
13 screens from the working solution, on its sample data. Pick one to see it large.
The rebalancing tolerance bands behind drift rebalancing, the tax rules, the restriction threshold and the approval limits are settings. The platform administrator changes them with a reason; each change is logged and applies to the next proposal.
| Setting | Default | Choose from |
|---|---|---|
| Tolerance band, per holding | ± 1.0% absolute | Per asset class ± 3.0%; cash between 0.5% and 3.0%; a model may tighten the firm bands, never widen them |
| Minimum trade | $250 | In US dollars; changed by the platform administrator with a reason |
| Drift check | 6:00 a.m. New York, daily | The Drift agent proposes a drift run; a portfolio manager schedules it |
| Wash sales | Avoided across the household | Taxable, IRA and held-away accounts; IRAs and tax-exempt trusts skip tax logic |
| Lot order | Losses first, then long-term, highest cost | Short-term gains only inside an account's gain budget, or with a portfolio manager's approval |
| Propose a rule for one-click confirmation at or above | 0.90 confidence | From 0.50 to 0.90 the agents ask client service; below 0.50 a person reads the document; a new rule is confirmed within 1 business day |
| Second signature | Order files over $25M or over 1,000 accounts | Head of Model Portfolios; under that, a second portfolio manager, never the approver |
| Model change service level | Within 2 business days of committee | Escalation to the Head of Model Portfolios at 75% of any clock |
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 (an investment research firm's 2026 US model portfolio landscape report: model portfolio assets reached $934 billion at the end of March 2026, up 46% in a year). Rule 3a-4 under the Investment Company Act (17 CFR 270.3a-4), the wash sale rule (26 U.S.C. §1091) and Rule 204-2 under the Advisers Act are named as the rules the work follows. The people, firms, clients, accounts and securities in the story and screens are fictional — characters and sample data in the working solution; the Treasury note is sample data.
Model portfolio rebalancing takes a model's target weights into every account that follows the model, when the investment committee changes the model or when accounts drift outside their bands. Model Portfolio Rebalancing gives the model portfolio management team eight agents that read each account's restrictions from its documents, fan the change out to every account grouped by why each must differ, choose tax lots, size buys to cash and re-check every proposal. The portfolio manager questions them in plain words, then approves or sends back each group; a second person releases the order file.
This is household rebalancing: the Tax agent reads the lots in every account of the household, including IRAs and a spouse's accounts at another custodian through the held-away feed. In the sample change it kept $418,600 of losses allowable in 87 accounts, and in one household it sold a 2023 lot instead of a 2025 loss lot because Ingrid Bergström's IRA at another custodian had bought the stock on October 6. Under the wash sale rule, a purchase in the client's own IRA counts too, and the loss is then lost for good because the IRA's basis does not rise; the Tax agent treats a spouse's IRA in the household the same way. Every lot sold or kept carries its reason.
The Restriction Reader reads advisory agreements, policy statements, trust documents, letters and CRM notes in the client's language and keeps each rule tied to its clause. On first read in the sample firm it found 2,673 restrictions written down but never keyed, and 412 keyed rules with no document behind them. At 90% confidence or above it proposes the rule for a person to confirm; from 50% to 90% it asks client service, and below 50% a person reads the document. Compliance approves every restriction change.
The Drift agent compares every account with its model each morning at 6:00 a.m. New York against its rebalancing tolerance bands: ± 1.0% per holding, ± 3.0% per asset class and cash between 0.5% and 3.0%, all settings. It proposes a drift run, as it did for 214 accounts in the sample, 164 of them in Growth 80/20 after the US large-cap rally, and a portfolio manager schedules it.
Yes. The agents propose, explain and check; they never send an order, release an order file, move cash, contact a client or lift a restriction. A portfolio manager approves each group with a reason code or sends it back with an instruction, a second person releases the file, and only then does the trader see it.
Every account gets a written rationale, cited to its sources, stored as its Rule 204-2 order memorandum and kept 7 years. The audit trail logs people and agents with the reason, evidence and agent version, and one rebalance exports as an evidence pack with the committee minute, rationales, quoted clauses, approvals, order files and fills.
Yes. Tolerance bands, the minimum trade, the drift check, wash-sale and lot-order rules, gain budgets, the restriction confidence threshold, approved substitutes, approval limits, service levels and data residency are settings. Changes need a reason and are logged; agent versions and thresholds change only through a change request compliance approves.
The Agentic Solution Engine builds and deploys it from your requirements and documents: your models, rebalancing and tax policies, a sample of client documents and your systems of record. It goes live once every quality gate has passed. We will walk you through it first on one of your own past model changes.
A cited yes, no or how-much for every fund, in under a minute, before the order.
1-min demoPerformance Attribution AnalysisWhy a portfolio lagged, answered the same day, each sentence tied to its attribution cell.
1-min demoInvestment Policy Statement MonitoringEvery policy limit checked daily, and every exception tied to its clause and its holdings.
1-min demoWe'll run Model Portfolio Rebalancing on one of your past model changes and the client documents behind its accounts.