The Institutional Equity Research Workflow, End to End

Institutional equity research is a repeatable lifecycle that turns public information into a decision you can defend and then keeps testing that decision for as long as you hold the stock. The full workflow runs through seven stages: generate an idea, map the business and where its revenue and profit actually sit, build a model from primary sources, forecast the drivers, run forensic risk checks, take it to an investment committee, and then monitor it continuously after purchase.

This piece is the map. Each stage below gets one tight paragraph on what happens and how AI is changing it, with links out to the deep dives. It is the AI-era, monitoring-inclusive companion to our older walkthrough of the equity research process. That piece covers the classic steps in detail; this one goes broader, treats monitoring as a real stage, and is honest about where machines now help and where they do not. None of it is advice, and every company named is a neutral, dated illustration.

The stages at a glance

Stage	What it produces	How AI helps
Idea generation	A short list of names worth the time	Screens a wide universe on many criteria at once, surfaces candidates from filings and news
Business and revenue map	A clear picture of how the company earns	Pulls segment disclosure together, tracks where revenue and margin actually sit
Build the model	Clean, reconciled history from primary sources	Gathers and aligns statements across years and formats, the slowest manual step
Forecast the drivers	Projected statements tied to real drivers	Structures the drivers, checks internal consistency, flags where a forecast breaks
Forensic risk checks	A list of red and yellow flags	Reads cash flows, related-party notes, and accounting choices at scale
Investment committee	A tested, written decision	Prepares the evidence pack, but the judgement stays human
Continuous monitoring	An up-to-date view versus the thesis	Watches every new quarter, guidance, and disclosure against what you assumed

Stage one: idea generation

The universe of listed companies is far larger than any desk can cover, so the first job is to narrow it to names worth the hours. Analysts screen on size, growth, returns on capital, leverage, and valuation, and let the list fall out, while other ideas arrive from a sector shift, a competitor’s results, or a name that keeps appearing at the edge of other work. The output is a short list, not a decision. AI widens this funnel: it can filter a large universe on many conditions at once and pull candidates out of filings and news that a manual screen would miss, which means the human hours go to judging ideas rather than finding them.

Stage two: map the business and its revenue

Before a single forecast, you have to understand what the company actually does and, crucially, where inside it the money is made. A single consolidated topline hides almost everything that matters. Take Reliance Industries in the quarter ended September 2025: Oil to Chemicals brought in around ₹1,60,600 crore of revenue at roughly a 9% segment margin, while Jio’s digital business did about ₹42,700 crore at roughly 52%. The biggest revenue segment is one of the thinnest, and the fattest margins sit in much smaller lines. Mapping revenue and profit by segment is the whole point of this stage, and AI helps by pulling scattered segment disclosure together and keeping it current. We go deeper in revenue mapping explained.

Stage three: build the model from primary sources

This is where the hours actually go, and it is the least glamorous part of the craft. Financial statements arrive across many years and many filings, in formats that drift over time; companies restate, reclassify, and redefine segments, so a line in one year does not automatically mean the same as the line above it. Before anything can be trusted it has to be pulled together, aligned, and reconciled from the primary documents themselves, not from a tidy summary someone else made. This is the stage AI changes most, because gathering and standardising history is mechanical, high-volume work that machines do faster and more consistently than a person with a spreadsheet. The mechanics of the model itself are covered in how to build a three-statement model.

Stage four: forecast the drivers

Now the work turns forward. A model translates a view of the business into projected statements, and the discipline is to forecast the drivers, not just the totals. Revenue is broken into the things that move it, volume and price, or subscribers and average revenue, or units and realisation, and margins are tied to cost structure and operating leverage. Management guidance is one input here: when Asian Paints pointed to volume growth “in the band of about 8-10%” and an 18-20% EBITDA margin for the year ahead, that is a forward claim with a number and a hedge, something you record and later grade. AI can structure the driver tree and check that a forecast stays internally consistent, but the assumptions, and the accountability for them, stay with the analyst. This connects to how professional investors build a thesis.

Stage five: forensic risk checks

Before anyone commits, the thesis has to survive a deliberate hunt for what could be wrong. This is where profit gets separated from cash, because profit is an opinion and cash is a fact. A gap between the two is a question, not automatically a red flag. Titan is a clean teaching case: its operating cash flow ran around half of net profit in FY24, went slightly negative in FY25 despite positive profit, then recovered to roughly 1.1 times profit in FY26. The swing traces to working capital, a jewellery business ties up cash in gold and store inventory as it grows, which is a structural feature, not a problem in itself. The method is to investigate the gap, never to pronounce a verdict. AI earns its keep here by reading cash-flow statements, related-party notes, and accounting choices across a whole portfolio and flagging the outliers early. More in stock forensics, finding problems before the market.

Stage six: the investment committee

A thesis is not a decision until it has been stress-tested by people who did not write it. The analyst brings the model, the drivers, the risks, and the evidence to a committee whose job is to attack the argument: where is the variant view, what has to be true, what would prove it wrong, and what does the downside look like. The written memo is the artifact that makes this possible, because a claim you have to write down is a claim you have to defend. AI can assemble the evidence pack and surface counter-arguments, but the committee stage is where judgement and accountability live, and that stays human by design. The craft of the write-up is covered in how to write an investment memo.

Stage seven: continuous monitoring

Buying is the beginning of the work, not the end of it, which is why monitoring is a full stage and not a footnote. A thesis is only right or wrong over time, and most of a position’s life is spent held, so the workflow that stops at purchase is only half a workflow. Monitoring means grading guidance quarter by quarter, watching whether reality is tracking toward the band or drifting away, and noticing when management’s language shifts from confident to cautious. It also means guarding against a subtle trap: when a company reports, it often restates prior-year comparables for demergers, discontinued operations, or segment changes, so the history you see today is not always the history that was knowable on the decision date. Testing an idea on “as reported today” data can quietly flatter it, a problem worth understanding through lookahead bias. AI is at its most useful here, watching every new filing, call, and disclosure against your stated assumptions so drift is caught in weeks rather than quarters. We treat this as the real edge in the investment thesis as a living document and why continuous research is the competitive edge.

The workflow is a loop, not a line

Laid out in a table the seven stages look like a straight line, but in practice they close into a loop. Monitoring feeds new facts back to the model, a broken assumption sends you back to the drivers, a forensic flag reopens the business map, and a big enough surprise becomes a fresh idea. The institutions that do this well are not the ones with the cleverest single forecast; they are the ones whose loop keeps turning, testing the same thesis against each new quarter long after the purchase.

What AI changes is the speed and reach of the mechanical stages, gathering, reconciling, reading, and watching, which frees the scarce human hours for the parts that still need a person: the assumptions, the committee argument, and the call on when a thesis has quietly stopped being true.