work.
- demand planning
- s&op
- inventory policy
- forecast diagnostics
- llm workflows
- design systems
what ido.
Four disciplines, one row each. Open a row for the principles I hold to and the failure mode I get hired to fix.
- The calendar is the product. If the monthly cycle isn’t on a fixed cadence people can plan their own work around, no amount of tooling fixes the meeting.
- Every review needs one number the room agrees is real before it argues about what to do next — consensus demand, not three spreadsheets with three different totals.
- Escalation criteria get written down before the first exec S&OP, not improvised in the room when the first real conflict shows up.
The most common shape this takes: a beautifully documented S&OP process binder that nobody follows past month two, because the design skipped the unglamorous part — who actually owns the number when demand and supply disagree, and what happens if they miss the deadline.
- Configuration should mirror how the business actually plans, not how the tool’s reference architecture assumes a generic company plans — the gap between those two is where most rollouts lose a year.
- Hypercare is where the real requirements surface. Budget for it as a design phase, not a warranty period.
- A planning model with more granularity than the business can act on is a maintenance liability, not a capability.
The common failure: a technically correct blueprint that models the org chart instead of the decision flow — so the system produces a number nobody in the actual meeting knows how to challenge or override.
- Safety stock is a policy decision, not a formula output — the formula just prices the policy you already chose (service level) in units of cash.
- Multi-echelon only pays for itself if the network’s node-to-node lead times are trustworthy; garbage lead-time data makes a multi-echelon model worse than a single-echelon one, not better.
- The right question is never "how do we hold less inventory" — it’s "which SKUs are we overpaying to protect, and which are we underprotecting."
The common failure: an optimization run that frees up cash on paper by cutting safety stock uniformly, then gets quietly reversed SKU by SKU over the next two quarters as service failures roll in — because the cut wasn’t risk-weighted.
- A statistical baseline’s job is to be a credible starting point for a human argument, not a final answer — forecast accuracy is a property of the process around the number, not the algorithm.
- Bias is a more useful diagnostic than error. A forecast that’s consistently 8% high is a coaching conversation; a forecast that’s randomly ±15% is a data problem.
- The consensus meeting should change the number sometimes — if it never does, it’s a status update wearing a planning meeting’s clothes.
The common failure: a forecasting tool upgrade sold as an accuracy fix, deployed against a consensus process that was never actually adjusting the baseline — so accuracy doesn’t move, because the tool was never the bottleneck.
play, don't read.
Two games and three live tools. Open one to run it right here; nothing loads until you do.
Stockout
Kombucha has days on the shelf, not months. Load crates onto the truck and ship them before they turn. A 40 second shift.
The Beer Game
The classic supply-chain game, solo: run one desk of four for 30 weeks and watch a small change in demand turn into a swing upstream.
play →S&OP Scenario Simulator
Move the levers a planner actually controls — watch the trade-offs move in real time. Illustrative model.
Inventory Optimization
EOQ, safety stock and reorder point from your own numbers — the math behind every replenishment policy.
Bullwhip Visualizer
A small wiggle at the shelf becomes a storm upstream. Watch demand variability amplify tier by tier. Illustrative model.
engagements.
Anonymised and NDA-safe — numbers as ranges. Open a row for the full situation → task → action → result.
Loading case studies…
© 2026 lavisahu.com. views are my own.
built by hand — next.js static export on cloudflare pages.
set in bricolage grotesque & space mono. bangalore, ist.
lavi sahu.
