Bazball Prompting: Why England's Ashes Struggles Mirror Generic AI in Investment Research

When England's Test team took "Bazball" to Australia this winter, the promise was bold cricket played on instinct, fuelled by freedom and confidence. By the third Test, freedom had blurred into fragility. A lot of head-scratching followed.

The same pattern plays out in investment research. Analysts turn to generic AI tools with a few "power prompt slogs," hoping instinct will suffice. When precision matters, that approach quickly shows its limits.

This builds on our earlier analysis of generic AI prompting risks. It examines why, in equity research, gut feel and cut-and-paste prompts deliver results as disappointing as England's second-innings collapse in Adelaide.

Generic Prompting is the Reverse Sweep of Research

The idea sounds simple: give every analyst generic tools to craft their own prompts. But it's often counterproductive.

England's batters insist on playing the reverse sweep first ball, even when the situation demands caution. Analysts do the same with prompts like "Summarize this earnings call" or "What's the guidance?" without considering sector context, temporal relevance, or data quality.

Like those rash cricket shots, generic prompts can deliver quick runs but often misinterpret nuance, misread sentiment, or miss critical disclosures buried in documents.

The Specialist Advantage: Knowing the Pitch, Not Just the Shot

England's Bazball method draws criticism for sticking rigidly to style rather than adapting to scenario. Analysts who rely on prompt templates built for "general use" make the same mistake. They ignore that equity research is heavily contextual.

Specialist tools don't just "know the game." They understand the pitch: whether data is structured or unstructured, which parts of the filing actually matter, and how company-specific language influences interpretation.

It's not about flashy strokes but consistent runs. Turning unstructured filings, earnings calls, and forward guidance into accurate, structured insights requires discipline.

Consider the difference:

• Generic prompt: "Summarize the Q3 earnings call"
• Specialist system: Automatically identifies guidance changes vs prior quarter, flags tone shifts in management commentary, extracts margin drivers with exact source citations, compares to peer group patterns

The first gives you a summary. The second gives you material insights ready for investment decisions.

Bazball Prompting Misses the Macro Conditions

Australia's bowlers exploited England's one-pace approach by reading conditions better. They held back when needed, attacked when the time was right. Generic prompting assumes one-size-fits-all answers work across companies, quarters, and regulatory regimes. That's a dangerous assumption.

Sentiment in a US tech CEO's Q3 call isn't comparable to a European industrial company's annual report, even if both include "bullish" language. Specialist prompting systems adjust for region, format, and tone, just as a seasoned analyst would.

Example: Management tone analysis

A generic tool might flag this as positive: "We remain optimistic about enterprise demand."

A specialist system recognizes this is management hedging language when the prior quarter said: "Enterprise bookings accelerated 35% with particularly strong momentum in financial services and healthcare verticals."

The shift from specific data to vague optimism signals deceleration. Generic tools miss this. Specialist systems catch it because they understand financial communication patterns.

It's Not Just the Tools, It's the Discipline

England's Ashes campaign suffered from inconsistency, not lack of talent. The issue was a misplaced belief that all scenarios demand the same response.

In equity research, the danger isn't just bad outputs from LLMs. It's the illusion of productivity. A 1,000-word "summary" produced by ChatGPT may look impressive. But was the prompt rigorous? Was the data structured? Did it pull from the correct version of the latest earnings transcript?

Specialist platforms build discipline into the system, ensuring prompts are context-rich, outputs are scoped to research needs, and human review remains central.

Self-Prompting and the Aluminium Bat: Just Because You Can Doesn't Mean You Should

For a brief moment in the 1980s, Australian batsman Dennis Lillee walked out to the crease wielding an aluminium bat. Technically legal, practically disastrous. The sound was off, the bat damaged the ball, and within minutes, the umpires demanded its withdrawal.

DIY prompting shares the same flaw. Just because anyone can craft a prompt or spin up a local LLM doesn't mean the result is fit for purpose in a regulated, high-precision domain like investment research.

Aluminium bats and generic prompts ignore the environments in which they operate. In equity analysis, accuracy, auditability, and repeatability matter more than novelty or convenience. Poorly engineered prompts might technically function but often distort output or degrade interpretability in ways analysts only notice after decisions are made.

Specialist prompting systems, like well-made bats, are tested, balanced, and built for performance under pressure, not one-off slogging.

Play the Match, Not the Moment

Test cricket, like research, rewards those who combine preparation with adaptability. Prompts that work during earnings season may fail during M&A cycles or regulatory stress periods. That's why specialist prompting systems aren't optional—they're infrastructure.

You can't play every delivery like it's the last over of a T20. Nor can you prompt every dataset as if it were a chatbot query. Winning the long game in research means using AI with purpose, not quick slog sixes.

From Bazball to Baseline Discipline

There's a place for flair in both cricket and AI. But over-reliance on improvisation, whether on a green-top in Brisbane or in investment research, tends to end the same way: clean bowled or lobbing up a simple catch.

AI should support analysts, not distract them. Prompts should guide models, not confuse them. And like any good Test innings, research output should be measured, reliable, and built on sound understanding of the conditions.

When it comes to AI in research, the real win isn't a six into the stands. It's a steady 80 not out from a specialized system that knew when to play, when to leave, and when to dig in.

With a career batting average of 44 (in village cricket), I know a thing or two about building an innings.

Using dedicated AI tools in investment research means playing the conditions rather than trying for a flashy six and getting clean bowled.