What happens if the forecasts score badly?
Then you find out early and cheaply, which is the entire point. A bad Brier score in
quarter one is worth more than a good anecdote in quarter four, because it tells you the
underwriting is wrong while the cheque is still small. The demo ledger opens at 0.323,
which is worse than a coin flip. It is shown rather than hidden because a measurement
system that only ever reports good news is not a measurement system.
We already have a data and analytics team. Why do we need this?
Your analytics team reports what happened. This is a discipline for committing to what
will happen, in advance, with a probability attached, and then being scored on it. Those
are different jobs and most analytics teams are not asked to do the second one. In practice
this runs alongside them: they usually own the data access, we own the underwriting and the
deployment. If your team already files dated probabilistic forecasts and computes calibration
on them, you genuinely do not need me.
We tried AI consultants and got a deck. Why is this different?
Two structural differences. First, deployment cost is booked against the forecast whether
or not it lands, so I carry the consequence of my own optimism. Second, there is a number
at the end that I cannot spin. A consultant's engagement ends with a recommendation; this
one ends with a score that follows me into the next quarter. Ask any firm that pitched you
for their Brier score across prior engagements. They will not have one.
Will the portfolio company CEOs actually cooperate?
They cooperate when the forecast is co-signed rather than imposed. The claim is sized with
their CFO and the metric comes out of their systems, so it is their number and not a number
the fund invented. The failure mode to avoid is using the ledger as a performance-management
stick on the CEO. It is a measurement of the AI programme, not of the operator, and it has to
be introduced that way or you will get gamed metrics within a quarter.
How do you handle data security and access?
Inference runs on owned GPU hardware on a private network, not on a third-party API, so
portfolio company data does not leave infrastructure you can inspect. Access is granted per
company, scoped to the systems a specific forecast depends on, and revoked at resolution.
Standard fund NDA plus per-company MSA. If a company requires full on-prem deployment, that
is supported and priced as an add-on.
$18K for three weeks. Why should the diagnostic cost anything?
Because a free diagnostic gets treated like a free diagnostic. The paid version gets the
CFO in the room, gets the data access approved, and produces a slate you can actually act on.
It is also the cheapest position in the market against the $500K MBB alternative, and it
credits in full against the Deployment setup fee if you continue. If it does not surface at
least $500K of measurable annualised upside, you do not pay for it.
