a field guide, distilled from MIT OpenCourseWare · system by [email protected]
A system is the machine your business runs on, and the discipline of building one that keeps working. Here are the durable ideas from MIT's systems-engineering course, in plain terms, each one cited to the course it came from.
Write down what "working" means before you build.
Most cost and schedule overruns trace back to missing or over-ambitious requirements, so a system starts with a written, measurable statement of what it must do, separate from how it gets built. That statement is also the baseline you test the result against. The Mars Climate Orbiter was lost because a units requirement was in the contract but never flowed down to the build and verified.Heavier is not better: the DC-3 was specified in a three-page request and a phone call. Match rigor to the stakes. A small growth test needs one measurable target and a deadline, not a 40-page spec.Fundamentals of Systems Engineering 16.842-fall-2015
Separate what the system must do from how you build it.
Requirements say what to achieve and how well; specifications say how it is built. Keep a hard line between them. Operators jump straight to a tool ("we need a chatbot"), which is a how, before naming the outcome the buyer or business actually needs. Name the outcome first and you can swap the tool later without losing the target.This maps to honesty-first work: the outcome is what a client signs off on and what you are accountable for; the build is your engineering call and can change. Do not let a client's named tool become the requirement when the real requirement is an outcome.Fundamentals of Systems Engineering 16.842-fall-2015
Most failures live at the interfaces, not inside the parts.
Teams pour effort into the individual parts while the interfaces between them get neglected and become the weak points: bottlenecks, structural failures, bad hand-offs. The Ariane 501 launcher was lost when one subsystem passed diagnostic data across an interface where the main computer read it as flight data. A growth system is a chain of hand-offs, and leads die silently in the gaps.Holds hardest when the system spans several tools or vendors, which is where most marketing stacks live. The fix is cheap: name every hand-off, define what crosses it, and test it live with real data.Fundamentals of Systems Engineering 16.842-fall-2015
Verification and validation are different questions. Answer both.
Verification asks "was the end product realized right?" (does it meet the written requirements). Validation asks "was the right end product realized?" (does it meet the actual intent). A system can be fully verified and still invalid: every metric green, the wrong thing for the business. You can hit every KPI while the client still books no jobs, because the real intent was missed.Verification can be mechanical and automated; validation needs a real signal from the real world (a live booked call, an actual paying customer), not a self-attested green light. Validation is the check operators skip.Fundamentals of Systems Engineering 16.842-fall-2015
A system's life is mostly after launch. Design for the running.
Lifecycle management treats launch as the start of the longest, most valuable phase, not the finish line, and it starts from the beginning. Daily operation, monitoring, maintenance, handling failures, and upgrading as needs change are all ongoing work. A growth machine nobody monitors or tunes decays as the market, the ad platform, and the audience shift.This is the engineering case for a monthly relationship over a one-time build, made by the discipline, not by a sales frame. Decide who runs and tunes the system, and how often, at design time. Keep the upkeep simple enough to actually run.Fundamentals of Systems Engineering 16.842-fall-2015
The pieces must fit each other, not just be individually good.
A business makes choices across categories (capacity, process, sourcing, people, information systems) and those choices have to be consistent with the strategy and with each other. You can buy a great CRM, a great ad account, and a great closer and still lose, because the ads promise an offer the process cannot fulfill, or the follow-up assumes staffing you do not have. Local excellence does not add up to a working system.Run the fit test before adding any tool or channel: is it consistent with the offer, the capacity, and what is already in place. In fast-moving markets any one configuration is temporary, so re-check the fit as things shift.Operations Strategy 15.769-fall-2010
Defined on the way down, proven on the way up. Every spec gets a matching test. The V-model as taught in 16.842-fall-2015.
A system is a chain of hand-offs. The diamonds are the interfaces. Verify each one end to end. After 16.842-fall-2015.
Sources · MIT OpenCourseWare
Fundamentals of Systems Engineering 16.842-fall-2015 (Prof. Olivier de Weck)
Operations Strategy 15.769-fall-2010 (Profs. Fine & Rosenfield)