What it knows about your industry
Before it has read a single file of yours, a twin has to know roughly how a business of your shape works. So the product carries twenty industry profiles, and each one is a written set of beliefs: how money moves, which levers matter, how companies of that shape get into trouble, what they have to decide, when their year is busy, and what range each figure usually falls in.
Build a twinThe assumption ledger
None of it is research. All of it is this model stating what it assumes, so that you can read the reasoning and replace it.
These are the model's own stated priors. They are not findings, they are not a study, and they are not what companies like yours achieve.
The claim on this page is not that the software knows your industry better than you do. It is that you can open what it believes about a business of your shape, read why it believes it, disagree with a number, replace it with your own, and watch the band around every answer narrow. Software you cannot argue with has nowhere to put what you already know.
Anatomy
A profile is not a row of benchmarks. It is seven pieces, and most of them are prose rather than numbers, because the shape of a business is mostly not a number.
It loads when you pick the industry, and every part of it is replaceable. The numbers are replaced by documents. The written parts are replaced by your own strategy notes, contracts and cost breakdowns, which is how a generic list of failure modes becomes a specific list about you.
The twenty
The middle column is the part that matters. It is the arithmetic the profile assumes connects your revenue to your cost, rather than a description of your sector.
| Industry | How money moves in a business of this shape | Levers the profile leans on |
|---|---|---|
| Software and SaaS | Revenue recurs, so this month is mostly last month plus what was added minus what churned. Cost is people and hosting, and neither falls when a customer leaves. | Price, churn, new customers a month, sales capacity |
| Professional services | Revenue is hours sold at a rate. Capacity is people, utilisation decides margin, and an idle week is gone rather than deferred. | Rate, utilisation, headcount, pipeline |
| Manufacturing | Revenue is units shipped at a price. A large part of cost moves with volume and the rest does not, so margin swings on how full the plant is. | Price, input cost, volume, capacity |
| Retail and ecommerce | Revenue is transactions times basket. Margin is thin, so a small move in input cost or in discounting is a large move in profit. | Price, traffic, basket, cost of goods |
| Healthcare services | Revenue follows episodes of care and who is paying for them. Staffing is the constraint and the largest cost at the same time. | Volume, payer mix, staffing, cost per episode |
| Construction and trades | Revenue is projects recognised as work is done. Cash and profit tell different stories, and one project can carry or sink a year. | Win rate, project margin, capacity, input cost |
| Logistics and transport | Revenue is movements at a rate. Cost is fuel, vehicles and drivers, and utilisation of the fleet decides whether the rate is enough. | Rate, volume, fuel and vehicle cost, utilisation |
| Hospitality and food | Revenue is covers or rooms times spend. Most of the cost is committed before the customer arrives, so quiet weeks hurt more than busy weeks help. | Price, occupancy, food cost, labour cost |
| Education and training | Revenue is enrolments across a term or a cohort, which makes the year lumpy and the intake date decisive. | Price, enrolment, completion, cost per cohort |
| Financial services | Revenue is fees, spread or commission on a book of business, so the size of the book and the quality of it both matter. | Rate or fee, book size, retention, cost to serve |
| Nonprofit and public | Income is grants, contracts and donations rather than sales, and it arrives on a calendar somebody else controls. | Funding mix, contract renewal, cost per beneficiary, reserves |
| Energy and utilities | Revenue is units delivered at a tariff, and the margin is the gap between a price bought ahead on a hedge and the price paid today. The plant, the network and the debt are fixed for decades. | Wholesale cost, hedging, tariff reviews, bad debt |
| Agriculture and food production | Revenue arrives around harvest at a price the market sets. The season is paid for months before it is sold, and one large buyer can decide whether the year pays. | Yield, market price, input cost, the largest buyer |
| Real estate and property | Revenue is rent on leases reviewed on fixed dates. Running cost is small and interest is large, so rates and lease ends move the year more than anything else. | Occupancy, rent at review, interest, lease length |
| Media and publishing | Revenue comes from readers and from advertisers who pay to reach them. Content is paid for before it finds an audience, and platforms control much of the traffic. | Audience, subscription price, advertising yield, platform share |
| Telecoms and connectivity | Revenue is subscriptions on one or two year contracts. The network is a fixed cost already paid for, so every extra customer on it is close to pure margin. | Churn at contract end, revenue per user, cost to win, network spend |
| Wholesale and distribution | Revenue is stock bought and sold on at a margin of a few points, so buying terms decide profit and stock and credit tie up most of the cash. | Buying price, stock turn, payment days, availability |
| Life sciences and medical devices | Revenue comes from approved products sold to institutions by tender. Research is paid for years ahead, and an approval date matters more than any price. | Approval timing, tenders, research spend, months of cash |
| Marketplaces and platforms | Revenue is a take on what buyers and sellers trade. Each side is worth more when the other is larger, so both have to be won and kept. | Value traded, take rate, repeat buyers, sellers leaving |
| Something else | The general shape: revenue, cost, customers and capacity, with nothing assumed about how they interact. | Price, volume, cost, headcount |
Picking the wrong one is not fatal. Everything a profile brings is a prior, is marked as a prior wherever it appears, and is replaced by the first document of yours that says otherwise. How origins and bands work.
Levers and rules of thumb
A profile names the handful of levers that decide the year for a business of that shape. Not because the others do not exist, but because a list of forty levers is the same as no list at all, and the point of the profile is to tell you where to look first.
Alongside them sit three to five rules of thumb, and every rule ships with the condition that breaks it. A rule of thumb with no breaking condition is folklore. One that says this holds until your largest account is more than about a quarter of revenue is something you can check against your own record in a minute.
The levers are also what the sweeps offered on a new twin are drawn from. The product does not start you with a menu of every parameter it has. It starts you with the four to six the profile says matter, and lets you reach the rest when you want them.
The written parts
The numbers in a profile are the easy part. These three are the parts that took the longest to write and are the first things an experienced operator will want to correct.
Three or four of them per industry, each with the earliest month it becomes visible in a run. The month is the useful half. Knowing that a business like yours dies of working capital is common knowledge. Knowing that it shows up in month four of the run, in a line you were not watching, is not.
Five to seven decisions that a business of that shape faces whether or not it wants to. They become the scenarios suggested on a new twin, so the first thing you are offered is a question your sort of company has, rather than a blank field.
Which months are usually busy and which are usually quiet. It matters because a scenario that starts in a quiet month is not the same scenario as one that starts in a busy one, and because a fall in month three is only a signal if month three was not always going to be low. Twelve months of your own revenue replaces it.
On screen
A figure and a band is where most software stops. Six pieces is what it takes for a comparison to be worth acting on instead of worth quoting.
| Part | What it says | Why it is there |
|---|---|---|
| The value | Your figure, as the twin currently holds it. | With its origin chip, so you can see whether it was read from a file or is itself a default. |
| The band | The quartile range this profile assumes for a business of that kind. | A range, not a target. Sitting outside it is information, not a fault. |
| Where it sits | One of five: well below, below, typical, above, well above. | Five buckets rather than a percentile, because a percentile implies a population that somebody measured. |
| What it means | One sentence naming the consequence for a company in that industry. | A consequence, not an adjective. Not low margin, but what a margin that size does to you in the month volume falls ten percent. |
| What it makes fragile | The thing that breaks first if this figure stays where it is. | Which is the part a ranked list of metrics never gets to. |
| The prior note | A sentence saying the band is a prior, and naming the document that would replace it. | On every row, every time. Not once in a footnote at the bottom of the screen. |
The fourth and fifth rows are the ones that took the writing. A comparison that ends at above typical leaves the reader to supply the consequence, and the consequence is the only reason the comparison was worth computing.
A worked row
How a prior behaves
A default is not a number sitting quietly in the background waiting to be ignored. It is drawn with a band on every replication of every run, and the bands on priors are wide on purpose.
So a twin built mostly on industry profiles produces wide answers. That is the correct answer to give somebody who has uploaded two files. The model is not being evasive, it is reporting how much of what it just told you came from your record and how much came from an assumption it is willing to name.
Replace one figure with a document and two things happen at once. The value moves to whatever your file says, and the band around it collapses to the narrow one a read figure gets. The answer at the end narrows because of evidence rather than because the software became more sure of itself.
The least flattering part
Every profile ends with a section most products would leave out: the places where this engine is a poor model of that industry, named on the screen where you pick it.
Take software. The engine treats capacity as revenue per head, which is a fair description of a business where people do the delivering and a poor one where the serving is done by machines. A software company can add customers without adding anybody, and until you tell the model what capacity means for you it will keep assuming growth needs hiring.
The same profile says there is no expansion revenue in the model. Revenue from an existing customer does not grow on its own, and nothing in the engine produces net revenue retention above one hundred percent. For a business whose entire plan is expanding inside its existing accounts, that is a large piece missing, and it is better to read it on the industry screen than to work it out from a strange answer three weeks later.
All twelve carry their own version of this. It is the most useful part of a profile and the least flattering, which is usually the same thing.
No sentence in the product says that companies in your industry achieve a particular number. There is no study behind these bands, no proprietary data set, and nothing from any twin ever feeds them. They are this model writing down what it assumes, with the reasoning next to it, so that the first useful thing you do is disagree with one.
Ranges commonly reported in public benchmark surveys and in textbook industrial organisation, rounded to one decision, then written down as this model's own assumption. They are not derived from anybody using this product. If a band looks wrong to you for your own industry, it probably is, and the fix is one document rather than an argument.
Change it. Everything a profile brought is a prior, so nothing you uploaded is lost or overwritten. The profile supplies starting points and prose. Your documents supply the numbers that matter.
No. The equations are identical for every industry. The profile supplies starting values, bands, seasonality and the written parts. That is deliberate, because an engine that behaves differently by sector is an engine you cannot audit in one place. The engine.
You replace the value, which is the part that changes an answer, and the band narrows to the one your own figure earns. The industry band stays visible underneath, so you can always see what you moved away from and by how much.
The ones the twin already holds: margin and cost structure on the money side, churn, concentration and satisfaction on the customer side, elasticity, competitor reaction and market growth on the market side, supply exposure, and people figures such as cost per head and revenue per head. Each gets a quartile band and the note naming the document that replaces it.
If it is not, use something else. It carries the general shape and assumes less about how your business works, which for an unusual company is the more honest place to start.
No. Profiles ship with the software and are the same for every customer. Nothing you upload trains, tunes or updates them, and no figure from your twin is ever pooled with anybody else. What happens to your files.
By the chip on it. A default carries an amber chip and a named document that would replace it, and a figure read from a file carries a green one and the quoted line. How the chain works.
Pick your industry, read what the software already believes about a business of your shape, and find the first band you disagree with. That is the document to send first.