No. 3 of TenAn Occasional Journal of Supply-Chain PlanningJuly MMXXVI
Lavi Sahu
Notes on Planning
Founded on evidence · Est. 2012Set in Fraunces & Newsreader
§ 1
Abstract
Set down · July 2026
TheThe
notes collected here are the working record of a simple conviction: a
supply chain should be planned from evidence, and the evidence should be
reproducible. Fourteen years of practice — across SAP IBP, o9,
Kinaxis and OMP, in functional and delivery roles inside a Big Four
consulting practice — have mostly been spent building the analysis that
sits underneath planning decisions: which node fails worst, whether the
forecast earns its keep, what a commitment will actually cost.
The industries are particular. Pharmaceutical cold chain, where a
temperature excursion is a write-off and the batch record is a legal
document. Fast-moving consumer goods, where the shelf punishes
hesitation within the week. Energy, where lead times are measured in
quarters and a planning error compounds quietly for years. Different
physics; the same discipline — bounded claims, cited methods, and
numbers that are computed rather than asserted.
The forecast is a decision, not a prophecy.
Increasingly the work extends to applied AI and automation inside
planning workflows — not as spectacle, but as one more instrument that
must show its working. Every demonstration linked from these pages runs
from seeded code on synthetic data, labelled as such: same inputs, same
answer, every time. What follows are three entries from that record.
§ 2
No.I
Entry
Supply-Chain Resilience
MethodTTR / TTS stress test
Year2026
Which node, if it fails, hurts most — and for how long?
EveryEvery
network has a node its planners quietly worry about. This entry
replaces the worry with a ranking. Working in the stress-testing
tradition of Simchi-Levi1,
each node of a synthetic multi-echelon network is failed in turn and
two clocks start: time-to-recover, the interval before the
node can serve again, and time-to-survive, how long
downstream inventory can carry demand without it.
Where recovery beats survival, the network absorbs the failure and no
customer ever learns of it. Where survival runs out first, the gap is
the exposure — measured in days, then priced in margin. Ranked across
the whole network, the result is usually uncomfortable: the most
dangerous node is rarely the most expensive one, and the mitigation
budget is rarely pointed at it. That mismatch, made visible, is the
deliverable.
Is our forecast actually adding value — and where is it worst?
AA
forecast earns its place only by beating the alternative of doing
nothing sophisticated at all. Forecast Value
Added2
makes that comparison explicit: every touch in the process —
statistical model, planner override, consensus meeting — is measured
against a naïve baseline, and any step that makes the number worse is
identified by name rather than absorbed into an average.
The second instrument is segmentation. Classifying demand by interval
and variability — ADI against CV², after Syntetos, Boylan and
Croston3 —
separates the smooth series, where statistics earn their keep, from
the lumpy ones, where the honest answer is an inventory strategy
rather than a cleverer model. Diagnostics before medicine: the map
decides where planner effort should go, and — just as usefully —
where it should stop.
Base against upside against constrained — what do we commit, and what does it cost?
SalesSales
and operations planning tends to fail politely: three functions,
three spreadsheets, three versions of next quarter. This entry
reconciles them in the open, in the tradition of Wallace and Oliver
Wight4 —
a base case, an upside, and a constrained case, each passed through
rough-cut capacity so that the executive choice is between costed
positions rather than optimistic slides.
The point of the exercise is the sentence at the end of it: what we
commit, what we hold back, and what that caution costs. When the
constraint binds, the model states where, by how much, and what buying
out of it would take. A plan that cannot state its own price is not a
plan; it is a wish with a header row.
Simchi-Levi, D., Schmidt, W. & Wei, Y.,
“From Superstorms to Factory Fires: Managing Unpredictable
Supply-Chain Disruptions,” Harvard Business Review,
Jan–Feb 2014 — the origin of the time-to-recover / time-to-survive
framing used in Entry I.
↩
Gilliland, M., The Business Forecasting Deal
(Wiley, 2010) — Forecast Value Added as the discipline of measuring
every process step against a naïve baseline.
↩
Syntetos, A. A. & Boylan, J. E.,
“The Accuracy of Intermittent Demand Estimates,”
International Journal of Forecasting 21 (2005), building on
Croston (1972) — the ADI/CV² classification scheme in Entry II.
↩
Wallace, T. F. & Stahl, R. A.,
Sales & Operations Planning: The How-To Handbook;
with the Oliver Wight Class A tradition — the executive S&OP
process behind Entry III.
↩
§ 7
Colophon
No. 3 of 10
This journal is one of ten identity explorations — the same
practitioner, set ten ways. It is composed in
Fraunces
for display and
Newsreader
for text, on paper the colour of warm limestone, and built as a single
page with no machinery beyond what the type requires. Occasional
writing on applied automation and planning appears at
Writing.
Disclosure: all demonstration data in the linked repositories is
synthetic and labelled as such. No client, and no client's number,
appears anywhere in these pages.