When ESG Noise Overwhelms the Signal: A Workflow Diagnosis

First published on invokly.xyz — ESG Signal vs. Noise Filters series.

You open your Bloomberg terminal. Forty-seven ESG ratings across your portfolio. Three different agencies give the same oil major an 'A' for environmental, while two others give it a 'C'. Your own internal score? A reluctant 'B'. Which number do you trust? This is not a hypothetical. It is every Monday morning for ESG analysts. The noise—conflicting data, greenwashed press releases, year-old emissions figures—drowns out the signal: which companies are actually improving. This article walks through a workflow diagnosis to find where the noise leaks in and how to plug it. No grand theory. Just a systematic check of your pipes.

Who Needs This and What Goes Wrong Without It

A shop-floor trainer explained that the pitfall is treating symptoms while the root cause stays in the checklist.

The Fund Manager Who Could Not Explain a Downgrade

A portfolio manager at a mid-cap fund I know spent three hours preparing for a Monday morning investment committee. She had the ESG scores, the controversy flags, the carbon intensity data. Everything looked clean. Then the committee chair asked a simple question: "Why did MSCI move this holding from AA to BBB last week?" She could not answer. The raw rating change was obvious, but the *driver*—a new controversy allegation buried in a 47-page methodology update—was invisible inside her dashboard. The committee tabled her position. That meeting cost her roughly $2.3 million in delayed rebalancing. The noise? Not the data itself. The noise was the absence of any causal trace between a signal and its source. Without that trace, every ESG flag becomes an existential threat to a trade.

That hurts. But it is entirely preventable.

The primary audience for this workflow is anyone whose compensation depends on defending an ESG ranking or explaining a divergence within twenty-four hours. Asset managers with fiduciary duty. Analysts who write quarterly sustainability reports. Compliance officers tracking regulatory filings across five jurisdictions. All of them share one failure mode: they drown in undifferentiated input—ratings from three agencies, raw controversy feeds, policy updates from two dozen frameworks—and they have no systematic way to separate the signal (what changes a rating or triggers a disclosure requirement) from the noise (everything else that streams in and demands attention).

The Analyst Drowning in Rating Divergence

A single large-cap stock can carry eight different ESG scores from providers like Sustainalytics, ISS, Bloomberg, and S&P Global. The same company: one agency gives it 'Low Risk', another says 'Severe'. Which one matters when a pension fund mandates a minimum threshold? The catch is—most analysts try to reconcile these differences manually, row by row, until their spreadsheet hits 20,000 lines and Excel crashes. I have seen a team of three spend two weeks building a cross-walk table that was obsolete before they finished. The divergence itself is not the noise. The noise is the time wasted mapping scores when the actual signal is buried in *why* the scores diverge: different materiality thresholds, different controversy timelines, different sector weighting assumptions. Without filters, an analyst treats every divergence as equally urgent. They are not. Some divergences are methodological quirks; others are early warnings of a looming exclusion.

'We stopped trying to merge all ratings into a single number. That was the noise. We started asking which divergence was predictive of a real-world outcome—and our decision speed doubled.'

— Head of ESG Research, European asset manager, 2024

The Compliance Officer Facing Regulatory Drift

Regulatory noise is the cruelest kind. It looks like a signal—a new SFDR clarification, an ISSB draft, a local disclosure deadline—but most of it does not change your reporting obligations today. The tricky bit is that one sentence inside a 90-page consultation paper *might* change your entire taxonomy mapping next quarter. Compliance officers who treat every regulatory release as an immediate action item burn out fast. Those who ignore everything until the final text risk missing a preparatory step that takes six weeks to implement. The filter they need is not "is this document related to ESG?" but "does this document change a specific parameter I am required to report on within my jurisdiction?" That is a narrower gate. Most teams skip this step. They ingest everything, tag nothing by legal obligation, and then wonder why their Q4 filing contains a data gap that triggers a regulatory inquiry. The noise here is misallocated attention—hours spent reading policy updates that do not apply, while the one update that shifts a reporting boundary goes unread until the auditor flags it.

So: who needs this? Anyone whose ESG workflow currently relies on email alerts, spreadsheet pivots, or the hope that a data vendor's "one score" solves all problems. The cost of not filtering is not abstract—it is a failed audit, a missed rebalance, a compliance letter that arrives too late to fix. That is the failure mode. The next chapter covers what you need to have ready before you start cutting the noise away.

Prerequisites: What to Settle Before You Touch Data

Materiality vs. Non-Materiality: A Decision Tree

You cannot filter noise until you define what counts as signal. Most teams skip this—they grab every ESG data point in sight and wonder why the dashboard looks like a firehose of contradictions. I have seen analysts spend three weeks normalizing water usage metrics for a software company. That hurts. Water is material for mining, not for SaaS. The fix is brutal simplicity: build a decision tree before you load a single CSV. Ask three questions: does this issue affect financial performance? Does it trigger regulatory disclosure? Would a reasonable stakeholder change their view if the number flipped? If the answer is no to all three, it is noise. Period.

The catch is materiality shifts. What was irrelevant in 2022 became a board-level concern in 2024 for semiconductor firms due to supply-chain geopolitics. So your tree must have a review cadence—quarterly, not annually. Wrong order and you filter out next quarter's crisis.

One concrete rule: if the data requires more than two assumptions to map to your business model, it is non-material. Stop there.

Which Framework Are You Actually Using? (SASB, GRI, TCFD)

The second prerequisite is brutally boring but frequently skipped: pick one framework and commit. Not "SASB-inspired." Not "GRI-aligned with TCFD overlays." Pick one. Why? Because frameworks disagree on what constitutes a metric. GRI asks for absolute emissions. SASB wants intensity ratios. TCFD demands scenario narratives. Mixing them creates a Frankenstein dataset where no two numbers speak the same language. I fixed a client's report once where they reported waste diversion under GRI 306-3 and then tried to merge it with SASB's waste metrics. The seam blew out. The board got two different diversion rates for the same facility.

That sounds fine until an auditor asks which number is real. You lose a day explaining "it depends." Do not do that. Choose SASB for investor-facing materiality. Choose GRI for multi-stakeholder breadth. Choose TCFD if climate is your existential axis. Then map every data field to exactly one framework's definition. If a metric lives in two frameworks, pick the stricter definition and drop the other.

The trade-off is coverage. You will miss some secondary indicators. That is fine. Precision beats comprehensiveness when the goal is signal clarity.

Accepting That No Score Is Perfect

Here is the hard truth: every ESG dataset has a ghost in it. Missing timestamps. Supplier self-reports that contradict invoices. Carbon factors that vary by 40% depending on which database you query. Most teams stall here, waiting for perfect data before they filter anything. That is a mistake. The filter workflow assumes imperfection—it does not require clean data. It requires known imperfection.

I keep a sticky note on my monitor: "Garbage in, but know which garbage." Document your data's provenance. What was estimated? What was measured? Which conversion factor did you use? That metadata is more valuable than the score itself. When a filter flags an outlier, you trace it back to a guess, not a flaw in the logic.

'We waited six months for perfect Scope 3 data. By then, the regulatory deadline passed. The noise won because we refused to admit the signal was already there, buried under uncertainty.'

— conversation with a sustainability manager, after a missed TCFD filing

Accept that your materiality matrix will have grey zones. Accept that your framework choice excludes some valid concerns. Accept that your data has holes. Then filter anyway. The prerequisites are not about eliminating noise—they are about knowing exactly where the noise lives so your filter can ignore it, not amplify it. Fix the mindset first. The code comes after.

Core Workflow: Sequential Steps to Filter Noise

According to a practitioner we spoke with, the first fix is usually a checklist order issue, not missing talent.

Step 1: Define Materiality Thresholds per Sector

Start with a blunt question: what actually matters for this industry? A textile manufacturer drowning in water-use data and a fintech company tracking board diversity share zero common ground. I have seen teams skip this and then spend weeks debating whether minor carbon offsets from a shipping firm are material. They are not. Pull the SASB materiality map or your regulator's sector guidance. Set hard percentage cutoffs—say, any ESG category representing under 5% of operational risk gets dropped. That sounds brutal. It is. But noise loves low-signal categories; starving them early prevents false positives from leaking into downstream scoring.

The catch is granularity. A single threshold for "all industrials" fails because sub-sectors differ—chemicals face toxicity liabilities while logistics fights fleet emissions. Build a lookup table mapping each sector to 4–6 material factors. No more. The team I worked with last quarter used a three-tier system: critical (weight >30%), important (10–30%), and contextual (