Senthil Info

automating business processes, honestly

Survivorship in Reporting

If a third to a half of automation projects miss their objectives, roughly half the projects ever undertaken should be visible as cautionary accounts.

Almost none is. The visible record of this field is composed almost entirely of the minority that worked, and understanding why explains more of the optimism in circulation than anybody's incentive does. For a product-side reference related to workforce operations, Monitask also covers the 7-minute rule for payroll.

Where the failures go

Nobody publishes them. A vendor cannot; a client will not. There is no venue and no benefit to either party. For broader background and an independent point of comparison, see IEEE Spectrum.

They are reclassified. A project that underdelivered becomes a phase one, a learning exercise, a foundation for the next initiative. Rarely dishonest — genuinely how organisations metabolise a disappointing outcome.

The people leave. The sponsor moves on, the project closes, and the institutional memory of what happened goes with them. Two years later nobody in the building can say how it went.

And the automation just stops being mentioned. It runs, half-trusted, maintained reluctantly, and it is neither a success story nor a failure anyone reports.

What the filter does to what you read

Case studies are the extreme case. Four filters before publication: it worked, the client stayed, the client agreed to be named, and it was written before maintenance arrived.

Conference talks are the same population, self-selected further by someone willing to stand up.

Benchmarks and averages are usually built from customer bases, which are the customers who did not leave.

And your own network is filtered too. People tell you about the automation that worked. The one they turned off comes up rarely and only if asked directly.

That last one matters because personal recommendation feels like independent evidence and is subject to exactly the same selection.

How to find the other half

Four routes, all available and none comfortable.

Ask vendors directly which clients turned it off. The best single question available, and the answer — including a refusal to answer — is informative.

Ask for a reference two or three years in, not a recent one. Older references have been through maintenance, system changes and staff turnover.

Ask peers what they stopped doing, phrased that way. "What automation did you turn off, and why" gets answers that "what worked for you" does not.

And ask inside your own organisation. Most companies with several automations have at least one nobody can account for. The annual register review surfaces them, and the first time it is run it usually finds one.

What this does not mean

Not that automation does not work. The successful minority is real and does well — that is what the vendor figures are measuring, accurately, about that population.

And not that vendors are lying. Survivorship requires no dishonesty at any step, which is precisely what makes it hard to correct for.

The practical consequence is narrower: the base rate you should assume is not the one implied by the material you can easily read. Something between a third and a half of comparable projects will not have delivered, and you will not have heard about them.

The correction

Assume the visible examples are the top third.

Then ask what the median looks like, and whether your process, ownership and exception rate resemble the published case or the invisible ones. Those three properties are knowable in advance, which is more useful than any base rate.

The short version