USE CASE

Portfolio exposure and risk analytics for AI agents

2026-11-06 · 5 min read

"What am I actually exposed to?" Abacus Labs' analyze tool answers with sectors, factors, single-name risk, overlap between funds, concentration and unrealized gains, in numbers an agent can relay in two sentences and act on in the next call.

The problem

Funds hide holdings. A person with an S&P 500 fund, a Nasdaq fund and a "growth" fund may hold the same ten stocks three times. Risk lives in factors and correlations, not in ticker symbols, so a portfolio that looks diversified by name can move as one position. Answering the question takes holdings look-through for every fund and a covariance model for every name.

People already ask AI for this. In a March 2026 survey of 938 US retail investors, 62% said they use AI tools for investment decisions, and the top worry, named by 38.9%, was incorrect or misleading recommendations (Investing.com). A language model asked "how much tech do I own" will produce a confident number with nothing behind it.

Point in time matters too. A risk model that knows today's index members when it looks at last year is wrong about last year. Our model is built point in time, covers US equities and ETFs, and is evaluated out of sample from 1993 to 2026.

What the agent asks

USER    What am I actually exposed to?

        agent -> analyze(account)

AGENT   Technology is 48% of your money and 71% of your risk,
        mostly through NVDA plus the overlap between your
        two index funds. You have no healthcare or energy.
        Your account moves about 31% a year; the broad
        market about 17%.

USER    How much do my two funds overlap?

AGENT   62% by weight. They share 9 of their top 10 names.

USER    And what's sitting at a loss?

AGENT   $11,960, in Intel and an energy fund. Want me to
        look at harvesting it?

        agent -> propose(account, "harvest losses, keep exposure", options=3)

What Abacus Labs returns

{
  "status": "ok",
  "as_of": "2026-11-06",

  "value": 412000.00, "cash": 6200.00,

  "risk": { "volatility": 0.31, "beta": 1.42, "benchmark_volatility": 0.17 },

  "sectors": {
    "technology": { "weight": 0.48, "risk_share": 0.71, "benchmark_weight": 0.32 },
    "financials": { "weight": 0.11, "risk_share": 0.09, "benchmark_weight": 0.13 },
    "healthcare": { "weight": 0.00, "risk_share": 0.00, "benchmark_weight": 0.10 }
  },

  "factors": { "momentum": 0.34, "size": -0.21, "value": -0.28, "volatility": 0.41 },

  "concentration": [
    { "symbol": "NVDA", "weight": 0.52, "risk_share": 0.71, "lots": 6 }
  ],

  "overlap": [
    { "funds": ["VOO", "QQQ"], "weight_overlap": 0.62, "shared_top10": 9 }
  ],

  "unrealized": { "gains": 148200.00, "losses": -11960.00,
                  "harvestable": [ "INTC", "XLE" ] },

  "flags": [
    "NVDA is 52% of value and 71% of risk",
    "no healthcare or energy exposure",
    "2 positions with harvestable losses"
  ]
}

How it works

Abacus Labs looks through every fund to its holdings, then maps each name to our point-in-time risk model: factor exposures plus specific risk. From that it computes portfolio volatility, beta, and the share of total risk that each sector, factor and position contributes. Weight and risk share are reported side by side, because the gap between them is usually the finding. Overlap is computed from holdings, not from fund names.

Tax lots are read directly, so unrealized gains and losses are per lot and split into short and long term. Every number sits next to its benchmark, and a short list of flags gives the agent sentences to relay. Analyze is free and answers in tens of milliseconds.

Analytics are the start of the loop, not the end. Each flag maps to a propose call, and a watch on the account re-runs analyze on a schedule and posts the result to the agent's webhook when a flag changes. Abacus Labs reports exposures; it does not say which exposures are good.

Share of portfolio risk by sector, example account


Technology    ########################################  71%

Financials    #####                                      9%

Consumer      ###                                        6%

Industrials   ###                                        5%

Other         #####                                      9%

Example account, illustrative numbers. Risk share is each sector's contribution to total variance from the risk model, so it sums to 100% while weights sum to the invested value.

Who uses this

General assistants via MCP
The first call on a newly linked account. The MCP server returns the numbers; the assistant tells the story.
Broker agents
A portfolio page that answers questions instead of showing a pie chart, and knows what to propose next.
Fintech app agents
Account review across thousands of users, with flags feeding proactive messages through webhooks.

The Personal plan is free: 3 portfolios and 300 credits a month, no card. Paid plans add portfolios, credits and client use.

Related