Three fill rates, one argument, ending it.

You cannot run a monthly reconciliation between demand and supply if the two sides of the table are working from different definitions of whether last month went well. Everything downstream of that — the forecast debate, the inventory target, the service commitment to a key account — is unresolvable, and the meeting degenerates into whoever argues best.

The definitional problem

"Fill rate" is not one measurement. At minimum it's three, and they answer different questions:

• Line fill — binary, per order line. Did the customer get the item they asked for, yes or no? This is the number that tracks customer experience, because a customer doesn't grade you on percentages, they grade you on whether the thing showed up.
• Unit or case fill — what share of the requested quantity shipped. This tracks operational depth, and it's the number that flatters you: ship 90 of 100 units on a line and you look like a 90, while your customer experienced a short shipment.
• Dollar fill — the same thing weighted by value. This is the number finance cares about, because it maps to revenue that didn't happen.

Each function had gravitated toward the metric that best described its own concern, which is entirely rational and completely corrosive. The fix wasn't picking a winner. It was publishing all three from one source, scoped identically, so that "fill rate was 94" became a sentence that had to specify which one.

Two scoping decisions did most of the real work. First, the metric was restricted to stocked SKUs — items the company had committed to holding. Including special-order items in a service metric punishes the planning function for lead times it never controlled and drags the number into meaninglessness. Second, those special-order items got their own metric, lead-time attainment, which is the honest question for that category: did we hit the date we promised?

The argument didn't end because someone won it. It ended because the words started meaning one thing.

Building the archive

A definition nobody can reproduce isn't a definition. The three metrics were implemented as SQL archive scripts writing to a governed store, with dashboards on top — so the number in the board deck, the number in the monthly review and the number a planner sees on a Tuesday are the same number, computed by the same code, with the history preserved rather than recalculated from whatever the current data happens to say.

That last part matters more than it sounds. Service metrics recomputed from live data drift, because orders get cancelled, credited and re-entered after the fact. An archive that snapshots what was true at the time is the only version you can hold anyone accountable to.

Then the cadence

With the definitions settled, the SIOP process could be designed around how the business actually decided rather than around a textbook calendar. The business had two genuinely different demand streams — a seasonal retail-facing side with key accounts, and a commercial grower/specialty side with different rhythms, different customers and different failure modes. Running one combined demand review meant half the room sitting through the other half's problems.

So the calendar was built as split demand reviews — one per stream, each with its own owner and its own participants — converging into a single enterprise supply review where the constrained decisions actually get made. Reconciliation, then executive sign-off. The whole cadence was maintained as a working calendar with a tab per stage, which sounds mundane and is the thing that makes it survive: everyone can see what's due, when, and who owns it.

Three design rules I apply everywhere and applied here:

1. One owner per call. Not a committee, not a rotating chair. If nobody's name is on the demand review, the demand review is a status update.
2. Overrides get written down. When a human changes what the system produced, the change and the reasoning are recorded — and revisited later against what actually happened. This is how a planning organization learns, and it's the single most-skipped step in the discipline.
3. The executive session decides, it doesn't review. If the constrained items don't arrive at the exec table with a recommendation and a cost attached to each option, you've built a reporting meeting.

The diagnostic that made the case

Selling a cadence to a leadership team is hard in the abstract. What made it land was a concrete walkthrough of the ERP's peak-blunting failure — showing, with the company's own items, how a three-month rolling average structurally undershoots going into the season and then keeps ordering for a quarter after it ends. Once a leadership team sees its own bad August inventory explained by arithmetic rather than by blame, the argument for a different process gets much shorter.

The same material went to the sponsor. Alongside it, an independent supply chain view fed into the long-range financial plan for a follow-on funding round — including the roughly $850k of run-rate savings modeled from right-sizing the buying team once the manual work came out of it.

What transfers

All of it, and this is the part of my work that travels furthest. The mechanics of a co-man beverage business and an eight-DC distributor have almost nothing in common; the disease is identical. Too many hands on the number, no shared definition of whether last month was good, and no forum where a constrained decision gets made and recorded.

I've now stood this up three times in three operating models. The third one went in materially faster than the first — not because the answer is the same, but because I know which fights to have in week two instead of month five. The definitional fight is always week two. If you postpone it, the process gets built on sand and decays back to spreadsheets inside two quarters.