Go / no-go on a new device launch

Launch or notHard

Problem. A PM proposes a new Amazon device. Leadership needs a go/no-go. Build the economic case. How do you decide?

Before you reveal: say your answer out loud, as if you were in the real interview — get your reasoning across clearly first. There is no single correct answer: reading what the interviewer is really after and defending your own thinking is what makes an answer strong.

Clarify the question first

"What's the decision criterion — downstream NPV over the device lifetime, at what hurdle?"
"Is it a new category (net-new customers) or a portfolio addition (cannibalizes existing)?"
"What's the uncertainty tolerance — this is a multi-million-dollar bet?"

Lay out your approach

Value it in downstream terms, not unit margin. The case is: forecast demand → estimate incremental downstream ecosystem value (net of cannibalization) → minus device cost (often a loss) and launch/tooling cost → downstream NPV.
Net out cannibalization — separate incremental demand (new customers, engagement-expanding upgrades) from substituted demand.
Forecast under heavy uncertainty (no history — analogs, conjoint, diffusion) and present scenarios / expected value, not a point.
Frame the go/no-go on expected downstream NPV vs the hurdle, with the downside scenario sized.

Then check in: "The case is downstream NPV — forecast demand, value the incremental downstream ecosystem spend net of cannibalization, minus device + launch cost — presented as scenarios under uncertainty. Want me to build the model?"

Simulate the back-and-forth

Interviewer

The device will be sold at a loss. How can it be a "go"?

You

Because the hardware loss is only the first term. Amazon's model is to make money on usage, so the case is the downstream ecosystem value the device drives — incremental Prime engagement, shopping, subscriptions, content, ads over the customer's lifetime. If a device sold at a $20 loss drives, say, $60 of incremental downstream value per customer over its life, it's a strong go despite the unit loss. So I'd build the downstream value per unit (causally attributed — the incremental effect, not correlated spend), multiply by the incremental forecast, and net the device and launch costs. The unit margin being negative isn't disqualifying; the downstream NPV is the criterion.

Interviewer

How do you handle no sales history and cannibalization?

You

For demand with no history, I'd forecast from analogs (similar past device launches' adoption curves), conjoint (attribute-based willingness to pay from surveys), and a diffusion (Bass-style) adoption model, with wide uncertainty. For cannibalization, I'd estimate how much of the demand comes from customers who'd have bought an existing model (substituted) vs net-new/upgraders (incremental), because only the incremental downstream value counts — and an upgrade is only incremental to the extent the new device drives more engagement. I'd present the forecast as a distribution/scenarios, not a point, given the uncertainty.

Interviewer

So how do you present the go/no-go?

You

As expected downstream NPV with scenarios and the downside sized. "Base case: incremental downstream NPV of $X at a Y-unit forecast; optimistic and pessimistic scenarios span $X_low to $X_high; the launch is a go if we believe the base case clears the hurdle, and the downside loss is bounded at $Z." I'd make the key assumptions explicit (downstream value per unit, cannibalization rate, adoption curve) and show the decision's sensitivity to them, so leaders can bet knowingly. The senior signal is a downstream-value NPV under honest uncertainty, not a single-point unit-economics spreadsheet.

Land a clear conclusion

You

So the go/no-go is downstream NPV, not unit margin: forecast incremental demand (analogs/conjoint/diffusion, net of cannibalization), value the incremental downstream ecosystem spend it drives, minus device + launch cost, presented as scenarios/expected value with the downside sized and assumptions explicit. Go if expected downstream NPV clears the hurdle with an acceptable downside — a hardware loss is fine if the ecosystem value justifies it.

Key summary — what really matters.

Decide on downstream NPV, not unit margin — devices are loss leaders.
Value = incremental downstream ecosystem spend (causally attributed) net of cannibalization.
Forecast with no history — analogs, conjoint, diffusion; wide uncertainty. Naming this as a cold start and leaning on comparable past launches beats pretending the point estimate is solid.
Present scenarios / expected value + sized downside, not a point.
Make assumptions explicit (downstream value/unit, cannibalization, adoption).

Downstream NPV, not unit margin. Per unit and at the forecast:

Term	Per unit	Note
Hardware margin	−$20	sold below cost
Incremental downstream value (lifetime)	+$60	causally attributed ecosystem spend
Net per incremental unit	+$40

At a base-case incremental forecast of 1.5M units (net of cannibalization), that's +$60M downstream value, minus, say, $25M launch/tooling → +$35M downstream NPV (base case). The hardware loss (−$20/unit) is more than offset by downstream value. Scenarios: if downstream value is only $35/unit or cannibalization is higher, NPV could go negative — which is why the decision is presented as a range with the downside sized, not a point.

Dimension	What interviewers look for	Below the bar ❌	Above the bar ✅
Problem framing	Whether you use downstream NPV.	Unit margin.	Downstream ecosystem NPV over lifetime.
Methodology and math	Whether you forecast + net cannibalization.	Gross demand at unit margin.	Analogs/conjoint/diffusion; incremental net of substitution.
Business judgment	Whether a hardware loss can be a go.	"Loss = no."	Loss offset by downstream value.
Deployment thinking	Whether you handle uncertainty.	Point estimate.	Scenarios / expected value; sized downside.
Communication	Whether assumptions are explicit.	Black-box number.	Key assumptions + sensitivity.

Interview tips

Strong-answer signal: framing go/no-go as downstream NPV net of cannibalization, forecasting with analogs/conjoint/diffusion, and presenting scenarios with sized downside. Common trap: rejecting a device on the hardware loss, or a single-point unit-economics case.