Bazball for Brains: Why Equity Research Needs More Than Big Swings at AI

Modern Test cricket is being upended by a new philosophy: play positively, take the initiative, and back your instincts. The results at Headingley are occasionally explosive. The results in Australia tend to be less flattering.

In recent months, a wave of analysts have taken a similar approach to AI. Prompt engineering is the new slog sweep. Type something clever into a general-purpose model, get back a fluent answer, and move on. But as we explored in our analysis of Bazball prompting and the Ashes, not every ball is a half-volley.

This article considers a different angle: what happens when the instinct-first approach collides with the structural demands of professional research workflows.

Big Hits, Wild Misses

Prompting feels powerful. You ask a question, the machine gives you an answer, and you move on. But like reverse-scooping a bouncer, it is a high-risk shot. The same prompt on two different earnings calls might give you one summary focused on revenue growth and another that highlights margin pressure. Both are technically in play, but neither is consistent. You cannot build a research process on that.

For equity analysts, the issue is not whether AI can produce output. It is whether it can build conviction over time. Research is cumulative. Analysts compare quarters like overs, build conviction across sessions, and look for patterns rather than highlights.

The Need for a Good Line and Length

AI systems that rely solely on prompt input have no sense of rhythm. There is no line, no length. Just a new delivery every time. That means the analyst has to reconstruct structure on each use: manually tagging guidance changes, sorting through management commentary, and trying to spot what actually changed.

That is manageable if you are facing one company report. Try doing it for 15 companies reporting in the same week, across three time zones and five sectors.

Reading the Field and the Filing

Another challenge is data coverage. US-listed companies play on flat pitches with clear sightlines: structured filings via EDGAR, XBRL formats, and rich metadata. International names present a different surface entirely. Unstructured PDFs, inconsistent reporting standards, and patchy regulatory structures.

Generic AI tools do not adjust for the conditions. They treat every document the same. Analysts need systems built to handle varying levels of structure and data quality across jurisdictions.

This is where Automated Data Import becomes critical. A system that ingests primary documents from companies worldwide, regardless of format, and makes them AI-ready without manual preprocessing.

Picking Your Moments

AI adds the most value not in slogging through every delivery, but in helping analysts spot the short ball outside off. Tools like Guidance Tracking extract management's public commitments and check whether they delivered. It is the modern version of keeping score.

When used correctly, this lets analysts assess not just what the company said, but whether the management team plays with discipline. That distinction between consistent performers and serial over-promisers is where real research value lies.

Research Is a Team Sport

Perhaps the biggest gap in prompt-first AI is that it treats research like a solo pursuit. Equity research is shared, distributed, and collaborative. Outputs need to be repeatable, auditable, and usable by more than just the person who typed the question.

A single analyst's insight is only useful if someone else can build off it. That requires consistent formatting, traceable sources, and structured outputs that do not change based on how the question was phrased.

Bringing It All Together

AI has earned its place in the analyst toolkit. But it is not an all-rounder, and it should not be used unassisted. The more productive approach is a structured middle-order partnership: human judgment supported by AI structure, each contributing to its strengths.

Play your shots. But remember, the best Test teams win not by chasing every delivery, but by knowing which balls to leave.