Pre-earnings leading indicators with Claude MCP

Consider a stock that has fallen approximately 40% year-to-date. The prevailing narrative: the company is losing customer demand. The turnaround is stalling. Short interest remains elevated.

Now consider that branded organic search volume — a real-time, independently sourced demand signal — has grown year-over-year in eight consecutive quarters. That the correlation between that search volume and the company's reported web sessions is r = 0.93. That the Q1 data, available weeks before earnings, shows +23.2% branded search growth, implying session growth of approximately +13.5% based on regression analysis.

The market says demand is falling. The data says otherwise.

This isn't a stock tip. It's a methodology — one that became possible because of how MCP connects data sources inside Claude. This article documents the approach, the statistics, and what it means for how analysts can work before earnings.

The timing problem in earnings research

Earnings research has always had a structural problem: the most important data arrives after the fact. The transcript, the filings, the KPI disclosures — all of it comes out when the call happens. By then, the market has already moved.

There's a category of signal that exists before earnings: consumer demand data, branded search volumes, app store rankings, web traffic trends. Alternative data practitioners have worked with these signals for years. But the workflow to combine them with company-reported fundamentals — to validate whether the signal actually correlates with what the company eventually reports — has historically required a data engineering team, a warehouse, and a significant amount of manual work.

Model Context Protocol changes that. When you can pull first-party IR data and third-party demand signals into the same Claude conversation, the correlation analysis that previously took days now takes minutes. The entire pipeline — data extraction, statistical testing, regression, interpretation — runs inside a single conversation, with no manual data handling.

That's what we tested.

The setup: two MCP data sources, one conversation

The analysis used two MCP-connected tools inside Claude:

Quartr Pro MCP — for company-reported operational KPIs extracted from quarterly interim reports. Specifically: website sessions (Nordic markets), extracted from the KPI table in each earnings report across 12 quarters. The data is structured and parsed directly from the source documents — no manual reading, no copy-pasting from PDFs.

A third-party search analytics MCP — for branded organic search volumes across five markets (SE, NO, DK, FI, DE), pulled via the organic keywords endpoint. The methodology explicitly excluded terms unrelated to the company's consumer business to ensure signal purity.

The question was simple: does branded search volume predict reported web sessions? If yes, how strongly? And what does the current quarter's branded search data imply about the unreported result?

No code was written. No data was exported. The full analysis — raw data extraction, correlation testing, regression modeling, and interpretation — ran in a single Claude conversation.

The methodology

Data collection

Twelve quarters of paired observations were assembled: branded organic search volume (sum of monthly volumes across five markets) and company-reported website sessions (Nordic markets only, as reported in interim reports). The time series runs from Q1 2023 through Q4 2025, with Q1 2026 branded search data available as a leading indicator ahead of the April earnings release.

Correlation analysis — three lenses

To avoid the trap of spurious correlation, the relationship was tested three ways:

Levels (n = 12 quarters): Pearson r = 0.927, p < 0.001 Spearman ρ = 0.916, p < 0.001 R² = 0.858

Year-over-year changes (n = 8): Pearson r = 0.927, p = 0.0009 Spearman ρ = 0.786, p = 0.021 Sign concordance: 7/8 quarters showed directional alignment (p = 0.035)

Quarter-over-quarter changes (n = 11): Pearson r = 0.949, p < 0.001 Spearman ρ = 0.809, p = 0.003 Sign concordance: 11/11 quarters showed directional alignment (p < 0.001)

The QoQ result is striking: every single quarter-over-quarter movement in branded search was mirrored in reported sessions. Pearson r = 0.949 with a perfect 11/11 directional match. The relationship is not an artifact of shared seasonality — it holds in differenced data. It is not driven by outliers — Spearman rank correlations confirm the pattern.

Branded search volume explains approximately 86% of the variance in reported website sessions (R² = 0.858). That's not directional similarity. That's a highly significant statistical relationship.

The pre-earnings signal

Using the YoY regression (Sessions YoY ≈ 0.70 × Branded YoY − 2.82), the Q1 2026 branded search growth of +23.2% implies approximately +13.5% session growth year-over-year — equivalent to roughly 24,500 sessions (000). The more conservative levels regression implies sessions of approximately 22,800, still representing approximately +6.0% YoY growth.

Both estimates contradict the narrative that has driven the stock down approximately 40% YTD.

The Q1 2026 earnings release was not yet published at the time of this analysis.

What this means for equity research

The methodology has three components that need to work together:

The IR data must be structured. Raw PDFs don't yield this kind of analysis at speed. The KPIs need to be extracted, parsed, and queryable. The Quartr Pro MCP integration provides this — company-reported metrics extracted directly from interim reports, structured and accessible in Claude without manual handling.

The correlation must be tested rigorously. Directional similarity is not enough. Shared seasonality can create the appearance of correlation without predictive validity. The three-lens approach — levels, YoY changes, QoQ changes — and the use of both Pearson and Spearman correlations is the minimum appropriate standard. A pattern that holds only in levels but disappears in differenced data is a warning sign, not a signal.

The workflow must run in one place. If you're exporting from three platforms and running regressions in Excel, the friction is too high to apply this across a portfolio. The value of MCP is that the entire analytical chain runs in a single conversation — which means the barrier to applying the methodology to a new company is low enough to do it regularly.

The practical implication: for any company with publicly disclosed operational KPIs and a measurable demand proxy, this framework can be applied. Consumer web sessions and branded search are one pairing. App downloads and app store rankings are another. Reported store counts and foot traffic indices. The pattern is consistent: find a leading indicator in third-party data, validate the statistical relationship historically, then check where the leading indicator sits today.

Limitations

Three things this methodology is not:

It is not a trading signal. Statistical correlation in historical data does not guarantee predictive validity going forward. The relationship between any third-party proxy and a company's reported metric needs to be re-validated for each application. The regression implies a range of outcomes, not a point estimate.

It is not a replacement for fundamental research. What this workflow does is surface counter-signals or confirming signals that inform a thesis — it does not generate one. The analyst still needs to understand why demand might be diverging from price, whether the company's cost structure supports a recovery, and what the market is pricing in.

It works best for specific company types. Companies with publicly disclosed consumer-facing operational KPIs and a clean demand proxy are the natural candidates. For companies that don't report web sessions, app downloads, or similar metrics, the methodology requires a different approach — or may not apply at all.

The broader point

The tools to do this kind of research have existed in separate places for years. What's changed is the infrastructure that connects them.

As MCP-connected data sources multiply — first-party IR data, search analytics, alternative data providers, internal datasets — the analyst with the right workflow will have access to signals that weren't practically available before. Not because the data didn't exist. Because the friction to combine it was too high.

That friction is now low enough to do it before earnings. In a single conversation. With statistically validated results.

The market said demand was falling. Before the earnings release, the data said otherwise.

This analysis was produced using Quartr Pro's MCP integration inside Claude. Quartr Pro includes native Claude MCP access at no extra cost. Learn more →