The Method Was Always the Asset
Software was the hard part.
It was standing in front of something harder.
In July 2026, one of India's larger independent wealth managers in Mumbai, with nearly $1 billion under management, sent us the files behind their performance reporting. We wanted to know whether their calculations could be reproduced from the files that were supposed to explain them. They sent a workbook of index data, a sample report, a goal-planning model, a client proposal, and a text file they described as a prompt.
It was not a prompt. It was nine numbered decisions with their reasons, built by the firm's own lead analyst with a frontier model and a text editor, and intended to drive the calculation.
From the analyst's file
Cost basis
Use the valuation report, not the transaction sum.
Missing index close
Walk backwards up to 20 days and use the last close.
Missing benchmark
Substitute the nearest index and disclose it.
Five of the nine exist only to cope with market data that turns up late, partial, or under the wrong name. These are one analyst's rules for one reporting workflow whose output goes to clients.
The report ends with a total row saying the figures are weighted by amount. Two of the three legs are. The third uses aggregate amounts instead, overstating the return by about eight percentage points. The same error flows through to the reported outperformance.
Elsewhere a benchmark is labelled as one index and priced with another. Several schemes are benchmarked against weighted composites whose gold and silver legs are dollar spot prices, introducing a currency mismatch for an Indian gold fund.
10 of 19 rows could not be reproduced from the supplied data.
That is not one arithmetic error. It is a mismatch between the rules the firm intended to apply and the rules the software actually ran.
Deterministic software will reproduce that mismatch perfectly.
The arithmetic sits in a script and is deterministic; it will return the same number twice. The definition sits in prose, in a file, and prose does not run. Nothing was ever comparing the script with the file, so nobody inside the firm had occasion to notice the two disagreed.
The same firm's retirement planning has the same shape. Their spreadsheet uses a five per cent assumption here and six per cent elsewhere. On one household, that difference more than doubles the required retirement corpus. The decision is embedded in the spreadsheet rather than stated explicitly.
The problem gets harder as firms get smaller. The median Swiss independent asset manager runs on3.3 staff and 68 clients.1 There is no second person to check the total row and no research desk to send the data request to. The method is often spread across spreadsheets, files and the memory of the people who built them.
The files we saw covered only one reporting workflow and one planning model. The index mismatches are the easy fixes. The harder question is how much of the firm's method sits elsewhere across spreadsheets, files and people's heads.
That question becomes more important as software gets cheaper to make and discard.
As applications get cheaper to make, they no longer need to look the same for everyone. A client's software can organise itself around their goals, holdings and decisions. An adviser can organise the same client around maturities, shortfalls and conversations that need to happen. A research analyst can work from a different view. The application can fit the person and the job instead of forcing both into the same app.
The views can change. The logic underneath them cannot. Cost basis, benchmarks, planning assumptions, suitability, refusal and record-keeping are not interface choices. They are the firm's method, and every personalised application has to answer to the same one.
The more personalised the software becomes, the more valuable the firm's method becomes.
The existing platforms were built for the opposite economics. Software was expensive, so one application had to serve thousands of firms. The dashboard could be branded and the workflow customised, but the application remained the common thing and the firm adapted its practice to it.
Direct platforms remove the adviser. Adviser platforms give thousands of advisers better versions of the same software. Both can add generated interfaces and governed AI, but the common thing is still the vendor's application and the shape it expects the firm's method to take.
Cheap generation breaks that model.
The application no longer has to be the common thing.
It can vary by firm, by role and by client.
When something that used to be hard becomes cheap, competition moves to what it was standing in front of. Quartz made accuracy cheap, and accuracy had been the industry's whole claim. Paul Graham calls what followed the brand age.2 In wealth management, software was standing in front of the firm's method.
The method was always the asset.
It has only just become separable from the software around it.
That is what VerdeWealth is built for: applications that fit the firm, the adviser and the client, each answering to the firm's method.