Learn

How this compares

Four other ways to answer what happens if we do this, all of which are genuinely used and three of which are sometimes better than a twin. Here is where each one stops.

Why not just ask a model

Written by the people who built one of the five, so read it with that in mind. The last section is the one where we argue against ourselves.

A spreadsheetone path, your own assumptions, no reactionAsking a modela plausible paragraph, no mechanism, no repeatA consultanta real answer, six weeks later, onceA twina range, a mechanism, and you can ask again tomorrow

Every one of these five gives you an answer. They differ in what happens when somebody asks why.

A spreadsheet answers with a formula. A language model answers with a paragraph. A consultant answers with judgement and a page of sources. An enterprise simulation product answers with a model somebody else calibrated. A twin answers with a ledger, a band, and the name of the agent that caused the damage.

Side by side

SpreadsheetLanguage modelConsultantsEnterprise simulationCounterfirm
An answer in under an houryesyesnonoyes
Runs on the record you already haveyespartlyyesnoyes
Shows a range rather than one pathif you build itnosometimesyesyes
Competitors and customers that reactnoin proseyesyesyes
Names which assumption is doing the worknonosometimesyesyes
Points at the line in the file a number came fromnonoyesnoyes
Same question, same answer tomorrowyesnoyesyesyes
You can ask the next question yourselfyesyesnoafter trainingyes
Knows your industry and your politicsnonoyesnono
Models inventory, tax and working capital timingif you build itnoyesyesno
Cost of the first answera week of somebodyminutessix figuresa projectan afternoon
Can be confidently wrongyesyesyesyesyes

The last row is not a joke. All five can be wrong in ways that are hard to notice from the output alone, and a well presented wrong answer is the most expensive product in this table. What differs is how much work it takes to find the error.

The spreadsheet

Still the most used tool in this list, for good reasons

A spreadsheet is transparent, auditable, and everybody in the room can read it. It is excellent at accounting identities, at timing, at tax, at working capital, and at anything where the structure is known and the arithmetic is the job.

Where it stops is behaviour. A cell cannot decide to leave. A model with a growth rate in it assumes the growth rate, and a price rise in a spreadsheet is a multiplication unless somebody chose a volume response by hand, in which case the answer is that person's opinion with a formula around it.

The second problem is that it gives one path. Three point scenarios help, and they still make you name the scenarios in advance, which means the case you did not think of is the one that happens.

  • Better than a twin at: tax, working capital, covenant tests, anything with a defined accounting structure
  • Worse at: what other people do when you move
  • Honest use of both: model the money in the spreadsheet, model the reaction in the twin
Monthly revenuedocumentGross margindocumentPrice elasticitydefaultMonthly churndocumentCompetitor reactiondefaultContracted revenuedefaultLargest customer sharederivedCost per headderived

The consultant

The only one on this list that can tell you something nobody wrote down

A good engagement brings pattern recognition from companies you will never see inside, judgement about which of your problems is actually the problem, and the political cover to say something uncomfortable. None of that is in this software, and pretending otherwise would be silly.

What you are buying is judgement, and judgement is expensive, slow, and gone at the end of the engagement. The follow up question two months later costs another engagement, and the model that produced the recommendation usually leaves with them.

The useful combination is obvious once you have tried it. Bring a twin to the first meeting, so the argument starts at which assumption is wrong rather than at whose spreadsheet to use.

  • Better than a twin at: knowing your market, reading your organisation, saying the unsayable
  • Worse at: being available on a Tuesday afternoon when the question changes
  • The consultants page on this site is written for consultants who want to bring a model rather than a deck about one
Outside adviserKnows things your files do notObjectivesKeep getting what they signed forRenew at a price they can defendNot run a switching projectPersonalityriskloyaltypatiencecandour

The enterprise simulation product

Serious agent based and system dynamics platforms exist, they have been used for decades in supply chain, logistics, epidemiology and manufacturing, and they are more capable than this is. They model things this does not: inventory, queues, individual machines, physical flow, discrete events at the level of a pallet.

What they ask for in return is a project. Somebody builds the model, somebody calibrates it, somebody maintains it, and the person who wanted the answer is usually three steps away from the model by the time it produces one. The cost is not really the licence, it is the specialist you now need and the six months before the first useful output.

This is built on the opposite trade. It gives up depth at the shop floor level, and in exchange a chief executive with a folder of files gets a defensible answer the same afternoon and can ask the follow up question without booking anybody. If your question is about physical flow through a plant, buy the other thing. If your question is about what customers, competitors and your own executives do next, this is closer to the shape of the problem.

When each of the other four is right and this one is not

This is the section to read if you are deciding whether to spend money here.

Use a spreadsheet when

The question is arithmetic rather than behaviour. Covenant headroom, a repayment schedule, a tax position, a cash flow with known timing, a budget reconciliation. Also when the audience needs to check every step themselves, because a spreadsheet is the most inspectable artefact in business.

Ask a language model when

You need the shape of an argument, a checklist of things you might have missed, a first draft of the question itself, or a reading of a messy document. It is genuinely better than any rule at pulling meaning out of a contract written in prose, which is exactly why this product uses one for that job and only that job.

Hire consultants when

The problem is that you do not know what the problem is. Or when the decision needs organisational authority as much as analysis. Or when you need someone who has watched twelve companies in your position and can tell you which three things actually mattered. A simulation cannot do any of that, and a simulation presented as if it could is worse than useless.

Buy an enterprise simulation product when

The question lives below department level: line balancing, throughput, inventory policy, routing, staffing a physical process by the half hour. Also when the model needs to run for years and be owned by a team, rather than answer one decision this quarter.

Do not use any of them when

The decision is already made and you are shopping for support. Every tool on this list will give you something that sounds like backing if you ask it the right way, and a twin is particularly good at producing an impressive looking version of whatever you wanted to hear. The defence against that is the assumption ledger, the sensitivity and the tripwires, and none of those works if you are not honestly trying to be wrong.

The comparison that matters is on your own question

Take the decision you have been deferring, build a twin, and see whether the answer changes what you would have done. If it does not, you have lost an afternoon.

Build a twinWhat it cannot tell you