We replaced end-of-day batch rebalancing for a high-net-worth asset manager with AI that watches live markets, rebalances each portfolio when it drifts and records every trade. Tax value captured rose 18%.
Industry: FinTech & Financial ServicesService: AI & AutomationUpdated
Overview
The challenge
The client used older end-of-day batch processing to spot portfolio drift. In volatile markets, portfolios could sit outside their target allocation for up to 18 hours, causing tracking error and missed chances to save tax.
Tax-loss harvesting happened only once a quarter, and the old system could not cope with live monitoring for a growing number of high-net-worth clients.
Rebalancing after the fact caused 1.2% annual slippage against benchmark targets.
Tax-loss harvesting ran only quarterly, missing intraday price swings that could have offset gains.
The system could not handle the computing load of live monitoring for the expanding client base.
The solution
We built a fast data system that processes live prices and volatility signals, and starts automated services when drift passes set thresholds.
Each portfolio has its own AI agent, which works out the best trade lots to avoid wash-sale violations and capture the most tax value.
Every automated trade is saved with a snapshot of market conditions and the rules behind the decision, to meet SEC and FINRA audit requirements.
Before and after
Before
End-of-day batch
Portfolios were corrected once every 24 hours, so they drifted during the day.
After
Continuous monitoring
Agents watch drift live and correct it within moments of a threshold being crossed.
Before
Manual and quarterly
Tax-loss harvesting was a manual admin task done 4 times a year.
After
Automated harvesting
Agents harvest losses whenever the timing is right, which gives high-net-worth clients more tax protection.
Before
Slower as it grew
The system slowed down as client numbers rose, which delayed rebalancing.
After
Parallel scaling
Cloud nodes start up automatically to handle market surges across millions of portfolios.
Key features
Direct links to custodian APIs, kept open in advance, remove execution delays during volatile periods.
Business impact
Sub-50ms execution
Every automated decision can be traced back to the exact market signal behind it, which gives regulators 100% clear evidence.
Business impact
Zero audit findings
The AI works through complex wash-sale scenarios across accounts held with several custodians, so no tax credits are lost.
Business impact
18% tax value lift
Results
18% Tax Value Increase
Automated harvesting caught intraday price swings that the old batch system missed.
Before: BaselineAfter: 18% Higher
92% Less Rebalancing Lag
Drift is now corrected in under 5 minutes instead of up to 18 hours, and portfolios spend 92% less time outside their target allocation.
Before: 18 HoursAfter: < 5 Mins
50ms p95 Execution
Fast, automatic execution places trades before market conditions move against the client.
Before: < 100msAfter: 50ms
Faster delivery
These modules saved about 10 weeks of setup, which went into the custodian integrations and the compliance records.
Set the limits each portfolio agent works within, including the automatic pause during sudden market events.
Gave us a tested base for wash-sale checks and tax-lot selection, which we extended across custodians.
Handled the live feed of prices and volatility signals at 1M+ signals a second.
Kept computing costs in check as nodes scaled up and down with market activity.
Trust and control
AI logic is version-controlled and can be reproduced exactly for regulatory review.
Compliance readyEach client's data is encrypted and kept separate from other clients' data during processing.
SOC 2-readySafeguards pause the agents automatically during sudden market events.
Fail-safeThe client owns 100% of the AI logic and investment automation components.
100% ownedA live data system processes 1M+ price and volatility signals a second. When a portfolio drifts past a set threshold, its own AI agent works out the best trade lots, avoiding wash-sale violations, and places the trades through direct custodian links in 50ms at p95. Agents pause automatically during sudden market events.
Tax value captured rose by 18%, because agents harvest losses during intraday price swings instead of once a quarter. Drift is now corrected in under 5 minutes rather than up to 18 hours, and portfolios spend 92% less time outside their target allocation. Drift accuracy is 99.9%.
Every automated trade is saved with a snapshot of market conditions and the decision rules used, so each decision traces back to the exact market signal. The AI logic is version-controlled and can be reproduced for regulatory review. There have been zero audit findings, and client data is encrypted and kept isolated.
If portfolios drift for hours before batch jobs catch up, we can help you build agents that rebalance in minutes and record every decision.