How Professional Investors Actually Build an Investment Thesis

An investment thesis is a single, checkable claim about why a business will do better or worse than the market currently expects, resting on a few named drivers you can watch. The work that produces it happens before any memo is written: you understand how the business actually makes money, map where revenue and profit come from, isolate the two or three variables that decide the outcome, and then say plainly why the market may have those variables wrong.

What follows concerns that thinking, not the document. Once the thinking is done, the investment memo is where you write it up for a committee. Here we stay upstream, in the messy part where a raw idea becomes a claim someone can later grade. Nothing below is advice on any security. Every company is a dated, neutral illustration of a method.

A thesis is a claim, not a description

Most first drafts of an idea are descriptions. “This is a good business with a strong brand and a long runway” describes a company. It does not say anything the market does not already believe, which means it says nothing about the price. A thesis has to contain a disagreement. You are claiming that some specific thing is truer, or sooner, or more durable than the consensus assumes, and that the gap will close.

That is why professionals talk about a variant perception: the exact way your view differs from the shared one. If your view is the consensus view, you have described a nice company and stopped one step short of a thesis. The test is simple. State your idea in one sentence, then ask what a smart person on the other side of the trade believes. If you cannot name their view, you do not yet understand yours.

If you cannot say what the other side believes, you do not yet have a thesis. You have an opinion you have not tested.

The rest of the path is the discipline that turns an opinion into that testable claim.

Step one: understand the business as a machine

Before any numbers, you need to know how the company actually earns a rupee. Not the marketing version, the operating version. What does it sell, to whom, how often, and what has to be true for a sale to repeat? This is the same starting point as the equity research process as a whole, and it is where a moat, if one exists, first becomes visible. If you want the deeper treatment of durable advantage, what an economic moat is goes there; here it is enough to ask whether the way the company makes money can be defended.

The mistake at this stage is treating a company as one thing. Large companies are usually several businesses wearing one name, and they can have wildly different economics inside the same annual report.

Step two: map revenue by segment and driver

The clearest way to see that is to break a company into its reporting segments and look at where revenue sits versus where profit sits. They are often in different places. Take Reliance Industries in the quarter ended September 2025 (Q2 FY26), using its own public segment disclosure:

Segment	Revenue (₹ crore, approx)	Segment EBIT margin (approx)
Oil to Chemicals	1,60,600	9%
Reliance Retail	90,000	6%
Jio (digital services)	42,700	52%
Oil and Gas (upstream)	6,100	83%

Read that table slowly, because it is the whole point of this step. The biggest revenue line, Oil to Chemicals, carries one of the thinnest margins. The fattest margins sit in segments that are far smaller by revenue. A single consolidated topline blends all of this into one number and hides the structure completely. An analyst who only looks at group revenue growth is watching the wrong dial, because the same rupee of extra revenue means something very different depending on which segment produces it.

Mapping revenue this way does two things. It tells you where the profit actually comes from, and it points you at the drivers, the underlying quantities like volume, price, margin, and mix that move each segment. That is the bridge from “what does this company do” to “what actually moves the answer.” For the mechanics of building that map, our companion piece on revenue mapping goes step by step, and for this specific company, the Reliance business model lays out the segments in more depth.

Step three: isolate the few variables that decide the outcome

Once revenue is mapped, most of it turns out not to matter to your thesis. This is the hardest habit to build. A company reports hundreds of numbers, and the instinct is to have a view on all of them. Resist it. In almost every case, two or three variables determine whether you are right, and the rest is noise dressed up as diligence.

For a consumer business the deciding variable might be volume growth and gross margin. For a commodity processor it might be a single spread between input and output prices. For a lender it might be net interest margin and credit costs. The job is to find the small set that the outcome hinges on, then ask a sharp question of each: what does the market seem to assume here, and do I have a reason to assume something different?

A thesis built on one clearly identified driver where you have an edge beats a thesis that gestures at ten drivers you half understand. Concentration is a feature. If your claim rests on twelve things all going your way, it is not a thesis, it is a wish.

Step four: form the variant perception

Now you assemble the disagreement. You have a mapped business and a short list of drivers. The variant perception is the sentence that says which driver the market has wrong and why the gap will close. It usually takes one of a few shapes. The market may be extrapolating a temporary condition as if it were permanent. It may be anchored to an old version of the business that the segment mix has already changed. It may be pricing a cyclical low or high as the new normal. Or it may simply not have looked closely at a small segment that is about to matter.

The point is that the variant view must be specific and it must attach to a driver you named. “The market is too pessimistic” is not a variant perception. “The market is treating this year’s compressed margin as structural, when the compression comes from an input cost that has already started to normalize” is one, because it names the driver, states the consensus, states your disagreement, and implies what evidence would settle it.

Step five: track the drivers, including what management promises

A thesis is not a one-time act. Once you have named your drivers, you watch them, quarter by quarter, and one of the most useful things to watch is management’s own guidance, because it is a forward claim you can grade against reality.

Guidance is a band with a hedge, and the exact words matter. Asian Paints, in its public commentary in recent commentary, guided to volume growth “in the band of about 8-10%” and pointed to an EBITDA margin band of 18-20%, speaking of “maintaining our margin guidance.” That is a driver you can track directly. A serious analyst records the band, then checks each quarter whether volume and margin are tracking toward it or drifting away, and watches how the language shifts, from confident to cautious or the reverse, over time. The number tells you the target; the wording tells you how sure management is. Our piece on management guidance goes deeper on reading that language, but the thesis-level point is simple: guidance turns a vague hope into a scoreboard.

Step six: state what would prove you wrong

The last step is the one that separates research from advocacy. Write down, before you are attached to the idea, the specific things that would tell you the thesis is broken. Not vague risks like “competition” or “the economy,” but observable events tied to your drivers. If your claim is that a margin will normalize because an input cost is falling, then the input cost staying elevated for two more quarters is a falsifier. If your claim rests on a segment’s growth, then that segment stalling is a falsifier.

Naming falsifiers does three things. It forces honesty about what you are really betting on, it gives you an exit that is a decision rather than a panic, and it lets a colleague check your reasoning before capital is at risk. A thesis with no stated way to be wrong is not strong conviction, it is an unfalsifiable one, and those are the expensive kind.

One quieter trap sits underneath all of this. When you test an idea against history, remember that the history you see today is not always the history that was knowable back then. Companies restate prior-year figures for demergers, discontinued operations, and segment redefinitions, so a backtest run on “as reported today” data can flatter a thesis that would have looked far less clean in real time. That is the essence of lookahead bias, and it is worth guarding against before you trust any pattern you find.

Putting it together

The path from idea to thesis is a narrowing. You start wide, understanding the whole machine, then map revenue to see where the money really lives, then cut down to the two or three drivers that decide the outcome, then sharpen those into a single claim about where the market is wrong, and finally name the evidence that would break it. What comes out the other side is short: a sentence or two of genuine disagreement, backed by drivers you can watch.

Everything after that, the valuation, the scenarios, the position sizing, belongs in the memo. But the memo is only as good as the thinking underneath it. Get the thesis right, in the plain and falsifiable form described here, and the document almost writes itself. Get it wrong, and no amount of polish on the page will save it.