Horizontal vs Conglomerate Diversification
A business that adds a second activity has done one of two very different things, and one question separates them: can anything the second activity does reach the first? Where it can, name the thing shared and follow the arithmetic. Where it cannot, the combination adds and does nothing else, and there is no fit to test. Only the related side gives the test something to name.
A business that ran one activity now runs two. What separates the two ways that can happen?
The first thing almost everybody does with a sentence like that is count. One became two, so the business must be spread more widely than it was, and if it is spread more widely then surely it is steadier. Hold that reading for a moment without acting on it. The count is where this subject goes wrong, and it goes wrong quietly: every fact inside it is correct.
Ask a different question of the pair, and ask it before any definition arrives: can anything the second activity does reach the first? Answering that for any real business takes about four seconds, and the answer decides almost everything that follows.
Work the two possibilities in the plainest terms, with nobody named yet. Suppose every unit the second activity produces goes into the first. Then the second is not sitting beside the first at all; it is standing in the middle of the first one's work, and a change made inside it reaches the whole of that work. Now suppose none of what the second produces goes into the first. A change made inside the second reaches none of the first's work whatsoever, and the only thing the two have in common is that the same people carry both of them.
Two activities under one business are either joined by a channel or they are not, and almost everything else about them follows from which. Where something is shared there is a channel, and a change made inside one of them travels along that channel and changes the arithmetic of the other. Where nothing is shared there is no channel, and the two sit side by side and add.
The pattern is familiar from a street corner. A snack stall outside one office building starts making the samosas it used to buy in every morning: that is a second activity, and every single samosa it sells now passes through it, so a bad batch or a cheaper supplier reaches the whole counter. The same stall instead starts selling phone covers from the same counter: also a second activity, also one more line on a list, and the samosas do not care. Both stalls went from one activity to two. Only one of them changed anything about the first.
A business that ran one activity now runs two. Which question is asked of the pair before any other?
Is this the same question as which direction to grow in?
The two questions are not the same, and saying so plainly now saves a contradiction later. There is another way of placing a move into a second activity, one that sorts it by direction: along the chain, into a stage the business does not run itself, or across the chain, into more of the stage it already runs. Direction is worked in full under Vertical vs Horizontal Integration: Which Way to Grow, and the conclusions there are about when each direction is the wrong move.
Relatedness asks something else entirely: only whether a channel exists between the two activities. Direction is about where a move went. Relatedness is about what the two ends have in common once it has gone there.
The cleanest way to see that the two questions are separate is to ask both of one move and watch both get answered. Anjani Stationers Private Limited, an invented manufacturer, makes hard-bound registers for schools. Sorting moves by direction places its purchase as a move along the chain, into a stage the business did not run itself, and that is the label the direction question gives it. Binding is a stage every one of this business's own registers already passed through on its way out of the works, so the relatedness question lands the very same purchase at the related end.
Both labels are correct, they answer two different questions, and neither of them corrects the other. A move can be along the chain and related, and holding both answers tells more than either one gave on its own. Reading either as an amendment to the other assumes the two questions were asking the same thing. The two questions were not.
The direction question places the purchase as a move along the chain, into a stage the business did not run. The relatedness question places the same purchase at the related end. What follows?
What is horizontal diversification, and what does its test look like when it can be run?
A horizontal move is a move into a second activity that shares something real with the first. A change made inside either one therefore reaches the other. The thing shared can be almost anything solid: the buyers, the skill, the premises, the machines, or a stage of the first activity's own work. A move is horizontal when the thing shared can be pointed at and named.
Naming gives a test, and the test is a procedure rather than an opinion. Step one, name the thing shared. Step two, name the arithmetic that changes because of it. Two steps and no more. Both steps have to produce an answer somebody else could check, and if the first step produces nothing then the second one never runs at all.
Run it on the one move that was actually taken. The works at Anjani Stationers runs three stages one after another on the same registers, and each has its own stage rateHow many units one step of a works can turn out in an hour, counted for that step on its own and taking no account of what the steps on either side of it can manage.: cutting turns out 150 registers an hour, printing 125 and binding 100. A register has to pass through all three in order, so the works makes 100 an hour and no more. One hundred an hour is the system rateHow fast a whole works produces. It is the rate of the slowest stage and no other rate, so it is never the total of the stage rates and never an average of them., the smallest of the three stage rates rather than their total of 375. Binding is therefore this works's constraintThe step in a sequence that decides the rate at which the whole sequence runs. Speed up any other step and the sequence finishes exactly as much as it did before.. All of that is worked in full elsewhere in these notes.
At the start of its second year, Anjani Stationers paid Rs 21,00,000/- for 70 per cent of a binder, Chitra Binding Works Private Limited. Chitra Binding Works became a subsidiaryA company in which another company holds a large enough share to decide how it is run. The larger company does not have to hold all of it, and usually does not. from that day. Putting a value on a business begins with a price set against a profit, and that is a separate subject, so the Rs 21,00,000/- stays undivided.
Now the two steps. Step one, the thing shared: every register this business makes passes through binding, so the second activity is not standing beside the first, it is standing inside it. Step two, the arithmetic that changes: the stage that had been setting the rate for the whole works is now a stage the business can decide about. Put together, those two steps reach a conclusion already published elsewhere in these notes: read next to the three line rates, the purchase stops being a piece of corporate news and starts being an operating decision.
And in the same breath, before that sentence has time to harden into a verdict: in the same twelve months the standing baseThe spending that arrives in the same size whichever way volume goes, so it has to be earned back out of volume every single year. rose Rs 24,40,000/- and operating profitWhat the business earned from trading, counted before finance cost and before tax. fell Rs 11,50,000/-. The fit is a description of where the move landed and not a judgement on whether it was worth making, and offering it as proof would turn a description into advice. A commitment and its result sit on different clocks, and one year is not long enough to be a verdict on either.
The household version is a tailor who buys the button shop next door. Every shirt he has ever sewn already needed buttons, so the thing shared names itself, and the arithmetic that changes is which of the two of them decides how fast buttons arrive. Whether he paid a sensible price for the shop is a completely different question, and nothing about the fit answers it.
The works cuts 150 registers an hour, prints 125 and binds 100, and the business bought a binder. What does naming the thing shared establish?
What did the second related move show, the one that was tested and never taken?
One worked example teaches a story. Two teach a test. So here is a second move of the same kind, worked out elsewhere in these notes, against a different sort of counterparty entirely.
Bhavani Register Works is promoter runRun by the people who started the business and who still hold most of its shares, rather than by managers appointed on behalf of outside holders. and makes 1,50,000 registers a year. Anjani Stationers carries 1,50,000 registers of idle capacityThe ability to produce that a works carries in a period and does not use. The machines are there, the year's standing cost is being paid for them, and they are switched off.. Before anything at all is read into that: those are the same number, the symmetry is a teaching device and not a discovery, and Bhavani Register Works was drawn at exactly the size of the first's idle capacity on purpose. The machines on both sides therefore cancel to nothing, and whatever is left standing afterwards is the only thing genuinely being bought.
The equality is a teaching device, said first so that nobody carries it away as a finding about how businesses line up in a trade. Real counterparties do not arrive at convenient sizes. Bhavani Register Works was written to.
Now the reading that belongs to the relatedness question. Both businesses make the very same product on the very same three stages, so something is shared here too. The move is related, the test is available, and it runs. Step one names the thing shared without any difficulty at all: the product, the stages, the machines. Step two then asks what arithmetic changes because of it. On the machines the answer is almost nothing: the first business could already do that work on lines it was standing idle.
A related move is not automatically a useful one: the channel is what makes the test available, and the test is what says how much arrives. Relatedness is a gate, not a verdict. If relatedness by itself decided anything, both of these moves would land in the same place, and they do not.
The status of the second move matters more than the reading. The move was never made, and nothing anywhere says it was even considered. A fit test was run on it to show what a fit test does, and no year in anybody's accounts records it.
Two stalls on the same street sell the same snacks. Under one owner they become two counters on the same street, and the second counter adds whatever the first could not already have served. If the first was turning nobody away, that is very close to nothing.
One business makes 1,50,000 registers a year and another carries 1,50,000 registers of capacity standing idle. What explains that equality?
What is conglomerate diversification, and why is there no case to run the test on?
Take the unrelated kind next, and take the missing case with it. A conglomerate move is a move into a second activity that shares nothing with the first. Not the buyers, not the skill, not the premises, not the machines, and no stage of anybody's work. The two activities sit under one owner and under nothing else.
The practice of placing a move into a new activity by what it shares with the activity already being run was set out by H. Igor Ansoff in 1957, and the frame here borrows it. A paper is a source for a way of asking a question, not a source for a figure.
Not one of the businesses worked above has ever moved into an activity unrelated to what it already does, so there is no case to run the test on. Inventing one would manufacture exactly the evidence nobody supplied, and a comparison propped up by a fabricated column is worse than a comparison with a gap in it.
A construction stands in its place, and the label travels with it: arithmetic demonstrating a property and not a business, with no name, no trade, no country and no year, belonging to none of the businesses above. Take an activity turning out 100 units in a period, and a second activity turning out 100 units in the same period. If every unit of the second goes into the first, a change made inside the second reaches the whole of the first's work, and what looks like two activities is one thing with two names. If none of the second's units goes into the first, the combined output is 200 units by addition and by nothing else, and a change made inside the second reaches none of the first's work.
With no channel the combination adds and does nothing else, so there is nothing to name as shared and therefore nothing to test. The channel is the whole difference between the two kinds of move, and the difference is visible in two sentences of arithmetic without a single rupee entering the argument.
What happens to the value put on a business that runs several unrelated activities is covered separately.
The panel below moves the share of one activity's output that passes into the other. Take it to none of it. What happens to the combined count of units?
Hold both blocks completely still, and move only what passes between them
One control, and it moves one thing: the share of the second activity's output that passes into the first. Neither block changes size at any setting. The two together turn out 200 units at every setting, and that total is added afresh at each position rather than printed as a constant. The panel is arithmetic demonstrating a property and not a business: no name, no trade, no country and no year appears anywhere on it.
Passing across 100 per cent units crossing 100 sold outside 0 combined count 200
Held at every setting: both activities turn out 100 units, a demonstration base rather than any business's figure; neither block changes size; and the two together count 200 units by design rather than by oversight.
Educational illustration. The setting at which everything passes across matches the shape of the purchase worked above, and it matches that shape rather than those figures.
Why does the test run in one direction and not the other?
The gap at the centre of this comparison is not an apology. Two related moves have been worked above, one taken and one never taken, and no unrelated move at all. So the fit test gets run twice on one side of this comparison and not once on the other.
The asymmetry is a fact about what has been published, not a fact about the two kinds of move. The asymmetry does not mean that related moves can be examined and unrelated ones cannot, and any reader who carried that away would have learned the shape of the evidence rather than anything about businesses.
There is a second and better reason the unrelated column stays empty, and it is worth earning rather than asserting. The first step of a fit test is to name the thing shared, so a fit test needs one. An unrelated move by definition carries none. So on that side the test does not merely fail for want of a case here; it returns nothing anywhere, in any set of accounts, for any business. The blank is structural before it is local.
A comparison with an empty column is honest as long as it says the column is empty and why, and dishonest the moment somebody fills it. Filling it would take one invented business and about four sentences, and the four sentences would then be quoted by somebody as evidence about a kind of move nobody in these notes has ever seen.
And the honest limit, said once and then left alone: one act is one observation. A reader who has watched one commitment through one year has learned the shape of the question and nothing whatsoever about how often the answer comes out either way. One honest observation is worth more than a second business invented to fill a column with.
Why is the test worked twice on one side of the comparison and not at all on the other?
So is the business spread more widely than it was?
Most readers arrive with that question, and it comes last because it needs everything above it. Take the count first and grant it in full: the business ran one activity and now runs two. The count is true and there is nothing wrong with it.
Now the relatedness question, put to the pair. Every register the business makes passes through binding, so the two activities do not move independently of each other. Whatever slows the register trade slows the binding work along with it, in the same week, without any decision being taken anywhere. The list grew from one name to two. The number of things that could move on their own did not move at all.
A count of activities is a count of names, and what matters is how many of them could move on their own. The two are different quantities that happen to share a unit, and that is exactly why one gets mistaken for the other so easily.
The same shape of test governs spreading money across holdings, where the finding is to count what could move on its own rather than what could be listed, and that subject is covered separately. The boundary of the analogy has to be said in the same breath so the two never blur: there the object is a set of holdings and here it is a set of activities inside one business, and the two are different objects sharing one word.
Ten shops in one shopping centre, all in different trades, look like ten separate things until the shutters come down on a public holiday and all ten take nothing on the same day. The trades were different. The driver underneath them was one.
The business ran one activity and now runs two, and the second is a separate company with its own board and its own accounts. Is it spread more widely than it was?
The note that counted the activities, and every fact in it was right
Somebody is writing the business up at the end of its second year and reaches for a sentence about how widely it is now spread. The writer records that a year earlier it ran one activity. The writer records that it now runs two. The writer records that the second is a separate company with its own board and its own set of accounts. All three are true, all three are published, and any reader could check every one of them.
Then comes the sentence the whole paragraph was heading towards: the business is more widely spread than it was and is therefore less exposed to anything going wrong in one place. Every fact in that note is correct and the conclusion follows from none of them.
The tempting diagnosis is the wrong one, so state exactly what happened. Nobody invented an activity and nobody miscounted. A count of activities was treated as a count of things that could move on their own, and those are different quantities that happen to share a unit. Every register the business makes passes through binding, so the second activity sits inside the first rather than beside it, and whatever slows the register trade slows the binding work with it. The list grew from one to two. The number of things that could move independently did not move at all.
Now the cost, landed somewhere specific rather than left as a complaint. The sentence is short and the test is not, so the sentence travels and the test stays behind. The next reader inherits a business described as spread more widely. The productive question, what here could move on its own, never gets asked at all. Worse, the note has now supplied a reason to expect steadiness that the arrangement never offered, so when both activities move together the movement reads as a surprise rather than as the ordinary consequence of a channel that was on the record from the day of the purchase.
And the part worth sitting with: the fact that the second activity is a separate company is what made the error easy. Two companies, two boards, two sets of accounts and two names all look like separation. A legal boundary and an operating channel are two different things, and only one of them decides whether two activities move together. The separation here is real in law and absent in the work.
The fix is one line and it is not a longer list: before writing that a business is spread more widely, name what the activities share, and if anything is shared then count what could move on its own rather than what could be listed.
What is local here, and what is not local at all
India supplies three things here and nothing more: the currency, the lakh and crore grouping of digits, and the legal form Private Limited, written out only so a reader can tell which set of accounts a figure belongs to.
The mechanism itself is universal, and it would be the same argument anywhere on earth: two activities either share something or they do not, and a change made inside one of them reaches the other only through what they share. No level, threshold, rate, period or filing requirement arises in this subject, and none is needed to run the test. A national figure on how many businesses run several activities, or a study of how such moves turn out, would be a figure of exactly the kind that turns a test into a claim about outcomes.
What a practitioner writes down, in this order, before the claim leaves the desk
Four lines travel with any claim that a business has moved into a second activity, whether the person writing is at a lender, at an analyst's desk, or simply working out whether a supplier is steadier than it looks. The four go in this order, and the order is what makes them useful.
One, what is shared, named as a specific thing rather than as an adjective. A stage, a set of buyers, a shed, a licence, a skill. Relatedness described without a named shared thing is a mood, and a mood cannot be checked by the next person to read the note.
Two, what arithmetic changes because of it. Which rate, which cost line, which set of buyers. A channel with nothing running through it is not a channel, and this is the line that separates a real connection from a passing resemblance between two trades.
Three, what could move on its own, counted rather than asserted. Line three separates a wider spread from a longer list, and it is the one a reader wants when the question is really about steadiness.
Four, where did each of the first three come from, meaning which published sentence supplies it. If the answer to that is nowhere, the claim is an assumption and gets labelled one on the spot rather than three readers later.
A claim about a move into a second activity with line one blank is a count rather than a finding. Notice how much work that single line does: it separates the two moves worked above from the construction two sections back, without anybody needing to know a single thing about a machine.
Somebody writes that a business has diversified into a second activity. Which line, left blank, turns that claim into a count rather than a finding?
One question, and the subjects that sit next to it. The question asked of a business that adds a second activity is whether anything the second activity does can reach the first. The published evidence then decides what can be tested and what cannot.
Which direction a business should grow in, and what a same-size counterpart would add, are covered under Vertical vs Horizontal Integration: Which Way to Grow. The commitment this business made, set against what the year produced, is covered under Corporate and Business Strategy Compared: Where and How to Win. A reported growth rate under the two ways of growing is covered under Organic and Inorganic Growth Compared: One Year, Two Rates. The use a year's result was put to is covered under Growth Investment vs Capital Return: One Pot, Two Uses. How a business is arranged, and what changing that costs, is covered under Organisation Design and Transformation: What Structure Costs.
How one company comes to hold most of another, how that purchase is priced, and how two sets of accounts are put together are separate subjects. The value put on a business running several unrelated activities is covered separately. Spreading money across holdings uses the same word for a different thing and is covered separately too. Neither kind of move ranks above the other, and nothing above settles whether the purchase was worth making.
What stands behind each part of this comparison, and what stands behind nothing
| Source | What it is | How it is treated here | Site |
|---|---|---|---|
| H. Igor Ansoff | Strategies for Diversification, Harvard Business Review, 1957 | Borrowed for one idea and one only: that a move into a new activity is placed by what it shares with the activity already being run. A paper is not a source for a number. Not one figure above came out of it, and none of its own cases is retold. The attribution is a single clause in the section on the unrelated kind. | hbr.org |
| Ministry of Corporate Affairs | The public record on which companies place their accounts | Stands behind a single sentence: a record of that kind exists, and a company most of whose shares are held by another still lodges its own accounts on it. Nothing anybody files states what two activities share or which of them could move on its own. The shared thing above therefore had to be named from the work rather than looked up. No requirement, no limit and no figure is drawn from it. | mca.gov.in |
| Built earlier in these notes | The three stage rates, the rate the works runs at, the purchase, the fit sentence and the confession about the drawn symmetry | Every one of them is quoted rather than recomputed, and no division and no subtraction is performed on any figure above. The direction label on the purchase belongs to the source that published it and is left sitting there. Where a source states something in its own conditional voice, or records that a symmetry was drawn on purpose, its own words travel with the figure instead of being tidied up. | finmaverick.com |
| The construction worked here | Two activities of 100 units each, and the share passing between them | The only invented arithmetic here. The arithmetic demonstrates a property and is not a business: no name, no trade, no country and no year attaches to it, and it belongs to none of the businesses above. A construction is labelled in the same breath as it is used, so the label travels with it at every appearance, inside the drawing and inside the panel at every one of its settings. | finmaverick.com |
Anjani Stationers Private Limited, Chitra Binding Works Private Limited and Bhavani Register Works are invented.
Educational material. Not advice on any investment, tax, budget or market position.