The Art of Business: Designing the Machine

← The Principle of Ascent · Series index

The mill that required its maker

A craftsman built a mill that ran unusually fast. Only he understood the levers. When the river changed, he adjusted them by ear. Visitors praised the machine. The town gave him more grain.

Soon he slept beside it. Every improvement pushed more volume through his hands. He could leave for an afternoon only by stopping the wheel.

A younger engineer asked a different question: how much grain could the mill process in a week when its maker was elsewhere? He found the question insulting until illness closed the mill at harvest.

When he returned they marked the levers, measured the flow, trained two operators, and replaced the adjustment that only his ear could judge. Output dipped while they learned. Then the mill ran through his first uninterrupted trip in years.

He still owned the mill. At last the mill stopped owning every hour of him.

The engine of a growing company

A durable company needs a loop: attract a fit customer, make a promise, deliver, collect, keep the learning, improve the next cycle.

Each stage should strengthen the next. Sales should bring customers you can serve. Delivery should produce evidence that improves sales. Support should surface recurring faults product work can kill. Collection should fund the capacity that protects quality. When stages undermine each other, growth adds friction.

Sales that promise exceptions to close deals dump expanding debt on operations. Operators who refuse all variation can miss a real segment. Design a route for controlled exceptions: who approves, how you price them, what they cost, and when they become a supported part of the offer.

Your job is to connect the cycle. Local efficiency cannot save a design whose incentives fight.

Four design laws

Complete the circuit. Every important process ends in an observed result. Sending an invoice is activity. Receiving and reconciling payment finishes the financial event. Creating a backup is activity. Proving you can restore the required information makes recovery real. Name the outcome before you measure the work.

Explicit authority. A decision needs an owner whose authority matches the consequence. Routine work should move without constant intervention. Ownership changes, major obligations, and critical service commitments need defined authority. "Use good judgment" is not enough when people have different skin in the result.

Transferable knowledge. Work becomes institutional when another qualified person can do it with the company's records and tools. Document the ordinary route and when to escalate. A wall of screenshots nobody can navigate preserves activity without transferring judgment.

Deliberate friction. Remove friction from repeated, understood work. Keep review where an error is expensive or hard to reverse. Spend attention where attention changes the outcome. Universal approval and universal automation waste that resource in opposite ways.

Design around the real constraint

Young businesses often assume cash is the only constraint. Founder attention often runs out first. A service that looks profitable after supplier costs can demand so many exceptions that you cannot sell or improve anything else.

Track what repeatedly interrupts planned work. Group by cause. Password reset, failed payment, customer misunderstanding, unreliable vendor: each needs a different fix. Calling all four "support volume" hides the design problem.

The best early win often removes a recurring decision. Clear product boundaries cut unsupported negotiation. A tested onboarding sequence kills avoidable setup failures. A renewal register beats memory. Each returns attention without hiring first.

Be careful building an elaborate internal platform to solve a problem you could remove with a clearer promise. Customers who need endless explanation: inspect the offer. Operators who need endless permission: inspect authority. A workflow that constantly breaks: inspect inputs before stacking another automation layer.

Shared infrastructure, separate accountability

Several businesses can share finance, tech, purchasing, or support. Sharing can also hide who is paying for whom. One venture can look successful because another pays its staff or carries its infra.

Keep shared work visible. Record which business consumes the resource and how costs are allocated. Simple enough to understand, stable enough to compare over time. Perfect allocation may be impossible. Invisible subsidy is worse.

A fictional group runs hosting, a publishing imprint, and a game project. They can share some admin and technical knowledge. Customers, obligations, and revenue cycles differ. Hosting must keep its commitments while the game experiments. Publishing needs clear rights and production control. The game must justify spend through learning milestones of its own.

You can subsidize an emerging venture. That is a capital allocation decision: amount, duration, purpose, and a condition for continuing. Without those limits, your strongest business quietly becomes the captive bank for the most exciting one.

Feedback that can change the design

A dashboard matters when a reading changes a decision. Before adding a metric, name the action it could trigger. Complaints may force a product correction. Delayed collections may change terms for new accounts. Rising delivery time may pause acquisition until capacity improves.

Use leading and completed signals together. A growing pipeline suggests future demand; it cannot pay today's supplier bill. Closed sales establish revenue; they may reveal trouble too late if support is already strained.

Hold a design review separate from status theater. Where did we need heroics? Where did a customer repeat information? Where was ownership unclear? Where did a process produce a result nobody wanted? Pick a few structural corrections. Too many simultaneous changes make effects impossible to read.

Design stays open to revision. A process for ten customers can fail at a hundred. Controls for a founder and one contractor can fail with several managers. Growth changes load. Keep inspecting the structure.

First operating blueprint

Draw the path from first contact to renewal in ordinary language. For each stage: owner, inputs, promised output, source of truth, failure route. Test with a real case. Every place the explanation becomes "then I usually handle it" is unfinished design.

Pick one piece and make it transferable. Have someone else run it while you watch. Their questions are evidence. Do not rescue the exercise with information that only lives in your head. Put it in the process.

Field order. Build one complete operating circuit this week. Verify the result from the customer's side and the financial side. Record the exceptions. The company starts expanding beyond its founder when ordinary success becomes repeatable without extraordinary intervention.

MSP standards: how GlassHosting shows up for managed care
Who we are
Art of Business series index