For consultants
The analysis is not the hard part. Getting from a folder of client files to something that can be interrogated is the hard part, and it eats the first third of the engagement. Counterfirm builds the twin from the same files in an afternoon, so the weeks go into the thinking the client is actually paying for.
One twin per client, one account per twin, nothing shared between accounts.
The model was never the deliverable. It was the thing that let you defend the deliverable.
Clients do not buy a spreadsheet. They buy a recommendation they can act on and a reason to believe it. What the build phase bought you was the ability to answer the fourth question in the room, and that is exactly what a twin gives you back without the three weeks.
The change in the work
This is the part to be straight about, because it changes the engagement rather than just accelerating it. The ledger lists every number with its provenance and its band, and the client can open it. There is no back of the model tab where a judgement call quietly lives.
That cuts both ways. A client can now ask why you set pass through at sixty percent, and you will need an answer better than experience. The compensation is that when the answer is good, it is visible, and the recommendation stops being a matter of trusting the firm.
It also ends a familiar argument. When a client says the model is wrong, they can change the number themselves and watch the conclusion either survive or fall over. That is a much shorter conversation than defending a workbook they cannot open.
It compresses the middle. It does not touch the two ends, which is where the value was anyway.
Deciding which question is worth answering is judgement and it always will be. The software will faithfully simulate the wrong question at speed.
The twin reads the files the client already has and tells you which ones it could not use and which gaps are costing the most band. That is a data request list generated from evidence rather than habit.
A twin, nine classes of agent, a graph and a first run in an afternoon. The three weeks of structuring a model becomes an hour of correcting one.
A standard run is four hundred simulations in about a tenth of a second on the machine this was built on. You can ask the follow up question during the meeting rather than promising it for Thursday.
A distribution and a mechanism are inputs to a recommendation. Deciding what the client should do with them, and what their organisation can actually absorb, is the work.
The usual briefs
A sweep across twenty five price points at eighty replications is two thousand simulations and produces a response curve with a best point. The curve, and its flatness near the top, is a better artefact than a recommended number.
Raise priceHold, match or split the difference, each run against the same baseline on the same seeds, with a competitor agent that reacts on a lag and only once rather than instantly and forever.
A competitor movesCost now against share later against an incumbent who reacts. The bands here are wide, and a client who understands why they are wide is better served than one given a single number.
Enter a new marketStructure out, with severance, the month the saving starts, morale, attrition and the revenue attached to the leavers, so the case is not just the payroll line.
Reduce headcountPracticalities
Each twin lives in one account and nothing crosses between accounts. Files sit in a directory outside the media library that is not served over the web, and deleting a source deletes its parsed chunks with it.
Self hosting is two WordPress plugins, which matters when a client contract says their data does not leave their infrastructure. You can run the whole thing inside their environment and hand it over at the end of the engagement.
Honestly
It does not replace judgement, and the failure mode is specific: it is very easy to run a beautiful simulation of a question that does not matter. Choosing the question, knowing which stakeholder has to be convinced, reading the room and understanding what the organisation can actually execute are the parts of the job that no distribution reaches.
It does not know the sector. There is no benchmark set, no market sizing, no competitor database and no view of what anyone else in the industry did last year. Everything it knows about the sector is a table of starting points that get replaced the moment a client document says otherwise. Your sector knowledge is still the differentiator, it is just no longer hidden inside a workbook.
It does not write your deck, and it will not defend a recommendation you have not thought through. A client who can open the ledger can also find the assumption you were hoping nobody would ask about.
How far each approach gets
It commoditises the model build, which was never what clients valued, and it makes the assumptions visible, which raises the standard for the judgement layer. If the value of an engagement was mostly in owning an opaque workbook, that is genuinely at risk. If it was in knowing which question to ask and what to do about the answer, this makes that easier to demonstrate.
The brief is a document you can edit and export, and the product name is a setting. There is no pretending the analysis was done by hand, and it would be a bad idea to try given the client can open the same ledger.
What is true. It is good at shape, order of magnitude, timing and identifying which actor causes the damage. It is not good at predicting what a market will do. Its accuracy is the accuracy of its assumptions, which is why they are all listed with bands, and the defaults sit at the top where they can be challenged.
Yes, or on the client's. Two WordPress plugins, one for the application and one for the site, with an install page that covers both.
Use the files you already asked for. If the readiness score comes back low, the list of gaps is your data request, generated from what the model is actually missing rather than from a template.