# Marginal gains, sometimes *Find the weakest link and put your weight on it.* --- You've inherited a business doing £80m. The board wants 15% growth and the CEO has briefed in six initiatives to get there: a pricing reset, a churn programme, sales enablement, an ops efficiency push, a brand refresh, and a roadmap reset on product. Each is sponsored by a senior leader with a target of five to ten percent improvement on their domain. You're meant to drive all six. The deck calls it marginal gains. --- You can already tell something's off about being asked to advance six things at once. You can't quite say why yet. Run the maths. Customer experience for this business is a chain. A prospect has to find you (marketing), engage (sales), onboard (implementation), use the product (success), renew (retention), and ideally expand (account management). Give each stage a probability of going well, a stand-in for the share of value-on-the-table that actually reaches the customer. Five sit around 0.9. Implementation sits at 0.3, because IT is bottlenecked and the average new customer waits eleven weeks for a clean go-live. Multiply. 0.9 × 0.9 × 0.9 × 0.9 × 0.9 × 0.3 = 0.18. You're realising 18% of the value the business could be creating. Now imagine the brand refresh succeeds and lifts that stage from 0.9 to 0.95. The product roadmap delivers and does the same. Sales enablement, the same again. Three of the 0.9s have moved to 0.95. Recompute. 0.95 × 0.95 × 0.95 × 0.9 × 0.9 × 0.3 = 0.21. Three initiatives, six months of executive air-time, three percentage points. The implementation bottleneck stays where it was, and it gates everything that flows through it. --- The British cycling case as it usually gets cited worked because the inputs were genuinely independent. A slightly lighter wheel was a separate addition to the same total as a slightly better warm-up routine, which was a separate addition to slightly cleaner sleep. The gains added. Most operating systems are sequential, gated, and multiplicative. A customer who churns at week ten because of a broken onboarding never sees the brand refresh. A salesperson with bad pipeline data prices badly however much pricing capacity has been built. A product release that ships without QA fails on a single integration regardless of how good the code review process is. The weakest link doesn't reduce the average; it multiplies through it. Some parts of a business are genuinely additive, like independent product lines or brand investment, and marginal-gains thinking lands fine there. --- Eli Goldratt made this argument forty years ago (he called it the "theory of constraints") and the marginal-gains era half-buried it. Find the binding constraint. Put your most senior people on it. Subordinate the rest. Don't run the brand refresh, the pricing reset, the sales enablement, and the roadmap reset in parallel; run the implementation programme until the eleven-week wait is a four-week wait and the 0.3 is a 0.7. The business goes from 0.18 realised to 0.41 realised on that single move. Then look around. The moment implementation stops being the constraint, something else becomes it. Sales capacity, perhaps, or the next product release. Whichever it is, you find what actually moves and you put your weight on it. --- Marginal gains do happen across the system. They are downstream of bottleneck-chasing rather than a doctrine in their own right. The phrase to keep is sequential, not parallel. One thing at a time, hard, until the constraint moves; then the next thing at a time, hard, until that one moves. The cycling team would have hated it. They had eighteen months and one race to win.