ScreeningaportfolioforESGcontroversieswithanAIagent

Key takeaways

• Decide which event types matter to your mandate before you search. A broad negative-news screen buries the flags you can act on.

• Record when an event happened separately from when it was reported, or old controversies will keep coming back as new alerts.

• Keep a running event record across screens and check every new result against it before raising an alert.

• Use the screen alongside your ESG ratings, not as a replacement score, and agree in advance who acts on each type of flag.

At a small asset manager, ESG monitoring is often part of one person’s job. They have a ratings feed, a list of holdings and no realistic way to read every story about every company. So they lean on the rating, which is fine for what it is: an assessment built on a framework. It just isn’t designed to tell you that a recall was announced yesterday.

An AI agent can do the reading. Give it a verified company list, a date window and the specific kinds of event you care about, and ask for each event with its source and date. Running the search is the easy part. Most of the work is checking what comes back: is it actually new, is it about the right company, and who should see it?

What the news gives you that a rating doesn’t

A reported lawsuit, product recall or regulatory action is something concrete to look into. Bigdata.com’s Negative News Screening example shows how to sweep a portfolio for controversy-related news and then dig into a single issue.

Be careful how you describe what you find. Sustainalytics says it updates controversies daily, and MSCI describes ongoing monitoring with daily checks on controversies and governance events. Spotting a story in the news doesn’t mean your provider missed it or will be slow to react. If you ever want to say the news got there first, you need dated evidence from both records.

So for each flag, get the agent to answer four things: what was reported, when it happened, when it was reported, and whether your existing ESG research already covers it. If a source doesn’t establish when something happened, leave that date blank rather than guessing. And check whether you’re looking at a new incident or the next chapter of a case you already follow.

Decide what you’re looking for before you search

A broad negative-news search will bury you. Earnings misses, downgrades and boardroom gossip are all negative and mostly irrelevant to a stewardship mandate. One user described generic AI search across a portfolio as “spray and pray”. Start instead with events your team can act on. The Bigdata.com example uses lawsuits, product recalls and regulatory changes, which makes a sensible base to adapt.

Treat these as the questions a reviewer needs answered, not fields the agent will fill perfectly every time. Ask it to quote the passage that justifies each category and to mark anything it’s unsure of. Keep allegations, findings and completed enforcement clearly separate; they mean very different things to a portfolio manager.

Write your exclusions down too. An earnings miss stays out unless it’s tied to a controversy. That kind of news belongs in a portfolio sentiment report, which our portfolio monitoring guide covers.

The prompt we’d start from

Attach your company list and date range, then use something like this:

Screen the attached company list for the specified ESG controversy categories within the supplied date range. Return one record per distinct event, including the company, event type, event date where stated, publication date, source link and supporting passage. Explain what has changed since the previous screen. Distinguish allegations from established findings. Flag uncertain matches, missing information and coverage gaps. Exclude negative news that falls outside the requested categories.

Have an analyst read the event records before any of them become alerts. Save the date range and instructions with the results, so next week’s run can be compared like for like.

The date trap

This one catches almost everyone. Bigdata.com’s search filters on publication date, which is what you’d expect. But an article published inside your window can describe something that happened years ago: a retrospective, an anniversary piece, a court filing that recaps old allegations. If you alert on publication date alone, old controversies come back as new ones every few weeks, and your reviewer stops trusting the screen.

The fix is an event record that lives across runs. For each event, keep the company identifier, category, location or case reference, first known event date, latest development and supporting sources. On each run, compare new results against that record before raising anything, and send unclear matches to a person.

In practice: a recap of an old investigation gets attached to the existing event and goes no further. A newly reported enforcement decision in the same case is a real update and goes to a reviewer. Let the passage decide which it is, not the headline.

Don’t let the screen pass for a score

An event screen tells you what happened and what deserves a closer look. A rating is an assessment against a defined framework; Sustainalytics, for example, describes separate risk and impact event scores and a controversy rating scale. Label the agent’s summaries clearly so nobody mistakes them for that. If you want provider scores sitting next to the screen, ESG scores are part of the premium catalogue on the Pro plan, so the agent can pull both into the same note.

For each flag, note whether your provider already covers the event, what date its assessment reflects and whether the new information adds anything. Record the comparison, but don’t assign a revised score yourself.

A sentiment scan across the whole portfolio can point you at names worth screening more closely, and our ten-workflow breakdown includes one. Keep the two reports linked and say what each measures. Negative coverage on its own doesn’t tell you a controversy happened, or how serious it is.

Who acts on a flag

Agree on this before the pilot, not after the first awkward alert. A common split: the company analyst checks the facts and the business exposure, the ESG or stewardship lead weighs what it means for the mandate and any engagement, and the portfolio manager decides whether the investment view changes. In a small team one person may wear two of those hats. That’s fine, as long as everyone knows who decides what.

Different events deserve different responses. A fresh allegation, a confirmed decision and a remediation update shouldn’t all trigger the same thing. When you escalate, pass on the evidence, status, date and reason for referral, so the next person doesn’t have to redo the search.

Write down the outcome of every review, including “no action needed”. It takes a minute, and it’s the record you’ll want if a client or regulator asks later. Use wrongly matched companies and recurring junk results to tighten the instructions.