Ideas (dots) arrive, pass or fail each gate (red squares are rejections) and queue for a limited number of development teams. Add teams or tighten early screening and watch the waiting line and cost per launch change.
The science
The pipeline has three gates with pass probabilities g1, g2, g3. Ideas arrive as a Poisson process with rate λ; screening and market testing take exponentially distributed time (means 1 and 2 weeks) and need no capacity. Only the g1 fraction that survives screening reaches development, which has c parallel teams and exponential service time with mean Td. Poisson arrivals thinned by a random pass/fail remain Poisson, so development is an M/M/c queue with arrival rate λ′ = λ·g1, offered load a = λ′Td and utilisation ρ = a/c.
The Erlang-C formula gives the probability that an arriving idea must wait, Pw = [ac/(c!(1−ρ))] / [Σk<c ak/k! + ac/(c!(1−ρ))], the mean wait Wq = Pw·Td/(c−a) and queue length Lq = λ′Wq (Little's law). If ρ ≥ 1 there is no steady state and the queue grows without bound.
Yield is g1g2g3 launches per idea; cost per launch is (c1 + g1c2 + g1g2c3) / (g1g2g3). Independent checks (below the readouts): a separate 30,000-idea simulation for the yield and cost, and a separate event-driven M/M/c simulation (earliest-free-server rule) for the mean wait, both compared with these closed forms.
Simplifications: exponential times, first-come-first-served, fixed illustrative costs (screening 0.02, development 0.5, test 0.2 M$ per idea) and a user-set product value; real innovation work has rework loops, learning and correlated ideas. The queue is capped at 140 waiting ideas for drawing. Educational only, not management advice.