Illustrative example (not a client engagement)
Which orders to accept when capacity, skill and margin all bind
A business with more demand than capacity must repeatedly decide which orders or projects to accept, at what price and start date — balancing expected margin against downside risk, scarce skills, subcontracting and delivery penalties.
The intervention
We model the choice as risk-adjusted selection under uncertainty: maximise expected contribution margin, penalised for downside (CVaR), delivery instability and client concentration, subject to capacity and skill-coverage constraints.
Stochastic optimisation · CVaR / risk-adjusted objectives · Constraint modelling · Scenario analysis
Result
On a synthetic benchmark, the risk-adjusted policy dominates a first-come / highest-price heuristic on the margin-versus-loss frontier — an illustration of the method. There is no client outcome in it.
This is a deliberately illustrative example on synthetic data, published to show the shape of the method. It is not a client engagement and carries no economic claim. An engagement begins with the client's baseline.
Backtested
This is an illustrative example on synthetic data, published to show how we frame an acceptance decision. It is not a client engagement, and it makes no claim about any company’s results. An engagement starts by establishing the client’s own baseline.
The decision
Accept-or-decline looks binary but is really a joint choice: accept which orders, at what price, promising which start date, staffed how? Take the wrong ones and a busy quarter can still lose money — the cost of a bad acceptance is spread across the whole delivery organisation. So it is often made by habit.
What this is and is not
The objective is not “maximise revenue.” It is expected contribution margin, penalised for what hurts: downside risk (via CVaR), delivery instability, and client concentration — subject to hard constraints on capacity and skill coverage that must never be violated. Crucially, the system returns a small set of Pareto-relevant plans rather than hiding every business preference inside one arbitrary weighted score; the owner still chooses.
What the illustration shows, and what it cannot
On a synthetic benchmark, the risk-adjusted policy sits above a first-come / highest-price heuristic on the margin-versus-loss frontier. That demonstrates the method. It does not demonstrate value for any company, because the model cannot know whether the estimated effort is accurate, whether the objective matches management’s true preferences, or whether a constraint was omitted. Those remain explicit assumptions. So an engagement is paid, scoped and measured against the client’s own baseline before anything is built.
