Four Suspects. Nobody Had Measured Any of Them.
- Meridian Industrial Supply distributes MRO parts and fasteners to manufacturers and contractors, selling on net-30/60 commercial credit terms.
- Order-to-Cash Cycle Time had drifted to 38 days — cash sitting in receivables and manual approval queues while Meridian's own supplier terms hadn't moved at all.
- Four problem areas had floated informally for over a year: slow credit approvals, a paper-based credit form nobody liked, inconsistent warehouse SLAs, and collections follow-up that varied by whoever picked up the account. None had ever been measured the same way twice.
- The CFO's question wasn't "where's the friction" — everyone already had a guess. It was "prove which of these actually moves the number, and prove it in 90 days."
That's the exact condition Chain Reactor is built for: a business with more suspected fixes than it can afford to chase at once, and no way to tell a real lever from an office opinion without measuring it.
What "success" had to mean here
The KPI Contract locked one distinction before Day 1: a shorter-feeling process isn't the deliverable. A shorter, measured, traceable cycle time is. And precision was the point — the target wasn't a hopeful round number, it was set because the plan already knew how it would get there.
- Floor: 34 days — must not get worse than this.
- Stretch: 10 days — aggressive on purpose, locked because the plan already scheduled the technology to get there, not softened into a safer guess.
Everything below was built against that line — and landed on it, not near it.
One Number, Locked Before Day 1 — Set Aggressively on Purpose
Diagnose is the only phase off the 90-day clock. The headline metric went to Meridian's leadership team with one requirement: written precisely enough that two people scoring the same week arrive at the same number. Stretch was not softened to a safe, defensible round number — it was set at what the plan's own technology bet said was achievable, and precision, not optimism, is what the rest of the engagement had to prove.
| Field | Value |
|---|---|
| Headline metric | Order-to-Cash Cycle Time |
| Definition | Days elapsed from purchase-order receipt to cash fully collected and posted, averaged monthly across all commercial accounts |
| Baseline (Day 0) | 38 days — the number on Day 1 |
| Floor | 34 days — must not get worse than this |
| Stretch | 10 days — aggressive by design, because the Value Tracking plan (Step 2) already scheduled a GenAI process-mining pass, not a conservative guess |
| Start date | Day 11, upon Day 10 Gate sign-off |
| Owner of the number | VP of Finance & Credit Operations — one named executive, not a committee |
Dimensions and atomic units are how the number moves. The KPI Contract itself doesn't move — it's the one fixed point everything downstream gets judged against. Precision meant this number couldn't move either: not padded lower to guarantee a win, not left vague enough to explain away a miss.
Four Dimensions, Each With Its Own Atomic Unit
Dimension Discovery confirmed which of the four suspected problem areas actually sat on the causal path to Order-to-Cash Cycle Time — and derived one atomic unit for each, before Build could start.
The causal path — nothing enters scope unless it traces back
Order-to-Cash Cycle Time ← Credit Approval (the dimension) ← Auto-approval rule, AU-014 (the atomic unit). Every unit generated during Build and Scale had to complete a chain exactly like this one.
The misfit team — assembled by reach, not title
- Credit Analyst — owns Credit Approval
- Warehouse Ops Lead — owns Fulfillment Handoff
- AR Supervisor — owns Collections
- Sales Ops Analyst — owns Order Processing
Value Tracking, designed before Day 1 of Build
Weekly per-dimension days-saved readout. Bi-weekly full Order-to-Cash re-measure. Source: ERP timestamp export — not self-reported. And one line item that made the Stretch target defensible rather than hopeful: a GenAI process-mining pass scheduled at the Day 60 Checkpoint, on the standing assumption that atomic-unit work alone gets a business like Meridian into the high-20s, and only a systematic scan for repetition — not duration — closes the rest of the gap. The team didn't know which findings that scan would surface. They knew, from having run it before, that it would surface something worth automating.
One Segment First — Not All Accounts at Once
Days 11–14: the first batch of atomic units is generated, all inside the Credit Approval dimension — each singly owned, binary verifiable, closable in days.
| Unit | Description | Owner | Status |
|---|---|---|---|
| AU-014 | Auto-approve orders under $10k | Credit Analyst | Done |
| AU-015 | Route orders over $50k to a senior approver | Credit Analyst | Done |
| AU-016 | Same-day credit-hold review | Credit Analyst | In progress |
| AU-017 | Retire the paper credit form | Sales Ops | Escalated |
Resourcing concentrated on one segment rather than spreading evenly across the whole book of business:
The Value Tracking dashboard itself was generative-AI-built — a single-file tracker, vibe-coded and running within hours of the KPI Contract being signed, not a six-week BI-tool procurement. It's the same tool that surfaces the Day 63 finding in Step 6.
Day 25: The Number Actually Moved
Units get executed. Anything stalling is escalated, reassigned, or dropped — never left "in progress" without an owner accountable for it.
| Reading | Value |
|---|---|
| Order-to-Cash Cycle Time | 38 → 33 days |
| Credit Approval dimension | 9 → 4 days — the mover |
| Verdict | First Proof Point cleared |
One dimension, moving for real, visible to leadership — the gate that had to clear before Meridian's effort was allowed to broaden into Scale.
Reinforcement Gets Built Now, Not on Day 89
Day 41: effort broadens past the priority segment, and each dimension gets a mechanism designed to hold the change after the engagement ends — built during the sprint, not bolted on at the close.
One mechanism per dimension, not a memo
- Credit Approval — the auto-approval rule is wired directly into the ERP via API, owned by Finance Ops, not a person — no human touch on a qualifying order.
- Fulfillment Handoff — an SLA breach auto-escalates inside the WMS through a rules engine, the moment it happens, not on next-day review.
- Collections — the dunning sequence runs on a schedule, no manual trigger required, with each follow-up drafted per-account by a GenAI writing assistant and sent without a human in the loop for routine cases.
Right on Schedule, the Plan's Own Bet Paid Off
Value Tracking kept running on cadence through Scale — stalling units surfaced while there was still time to act, not discovered at Day 89.
| Dimension | Week 7 read |
|---|---|
| Fulfillment Handoff | Escalated — 0 days moved in 2 weeks |
| Order Processing | On pace |
| Collections | Ahead of pace |
Steady, real, and — on atomic-unit work alone — headed for the high-20s: exactly what the KPI Contract's own plan had assumed conventional discipline would deliver. Day 63, exactly as scheduled at Diagnose, the Strike team ran the GenAI process-mining pass locked into the Value Tracking plan back on Day 10: six months of raw ERP timestamps and email-approval threads, every order, every touch, every re-open, cross-referenced for repetition instead of just duration.
The Day 63 process-mining pass — three drains a manual audit never saw
| Finding | What was actually happening | Est. days/order |
|---|---|---|
| Duplicate re-keying | 15% of orders were manually re-entered a second time because the CRM and ERP didn't share a live feed | +7 days |
| Redundant credit-hold reviews | Orders were being re-reviewed 2–3× on a stale-sync flag nobody had ever traced back to its root cause | +6 days |
| The gray-zone email loop | Orders too large to auto-approve but too small for senior review sat in a manual email approval chain with no SLA | +5 days |
The Day 60 Checkpoint that preceded it read the trajectory against Floor and Stretch as "on pace to clear Floor, on schedule for the Day 63 scan to close the rest" — written down as the plan, not revised after the fact. That's the precision the KPI Contract was built to protect: the target didn't move once, before or after the scan.
Floor Cleared. Stretch Hit — Not Approximately, Exactly.
The last Value Tracking read is taken against the baseline locked in the KPI Contract back on Day 10 — including the aggressive Stretch the plan bet on from the start.
Order-to-Cash Cycle Time, Day 0 → Day 90
Bi-weekly full re-measure, ERP timestamp export — not self-reported. The bend after Day 60 is the scheduled automation sprint landing on the line it was aimed at.
The Full Chain, Inspectable
Metric ← Dimension ← Atomic Unit is assembled as one complete, inspectable chain — nothing in the final number that can't be traced back to a named unit of work.
| Link | Movement |
|---|---|
| Order-to-Cash Cycle Time | 38 → 10 days |
| ← Credit Approval | 9 → 1 day |
| ← AU-014, AU-015, AU-016 | the atomic units that started it moving |
| ← CRM↔ERP sync, AI approval-routing agent, document AI | the automation build the Day 63 scan triggered |
The handoff, itemized
- Dashboard ownership → Finance Ops
- Reinforcement mechanisms → each dimension's process owner
- Automation build ownership → AI & Technology lead, with a monthly drift check against the original process-mining baseline
- Cycle 2 candidate → Returns Processing — flagged, not started, same process-mining pass recommended before any atomic units are generated
The engagement ends with a fact and a named owner for what happens next, not a narrative and a filed deck — exactly the distinction Chain Reactor is built to protect.
The Compound That Landed Exactly Where It Was Aimed
One dimension moved first and proved the method. Then a pass scheduled on Day 10, run on Day 63 exactly as planned, found eighteen more days hiding in plain sight. One honest number — Floor cleared, Stretch hit on the number, with the two halves of the gain named separately, not blended into one undifferentiated win.
Day 0 → Day 90, at a glance
The same chain named at the KPI Contract on Day 1 — the fourth link was already scheduled then, not discovered by accident on Day 63.
Precision meant setting the target aggressively, not proving it modestly. Stretch was locked at 10 days on Day 10 — before the scan ran, before anyone knew which three findings it would surface — because the plan already accounted for where the technology would do the heavy lifting. A conservative Stretch that gets blown past by a lucky find isn't precision. A Stretch hit on the number is.
The KPI Contract prevented scope creep. Four suspected problems went in; only Credit Approval and Fulfillment Handoff ever traced cleanly back to the metric at the weight leadership originally assumed.
Concentration beat breadth first, before technology beat anything. One segment — Enterprise accounts, 61% of cycle-time variance — produced the First Proof Point in 25 days, the old-fashioned way. The scheduled AI scan had nothing to point at until that discipline existed.
GenAI found what a manual audit structurally couldn't. A time-and-motion study measures how long one pass takes. The process-mining pass measured how often the same order got touched twice — repetition, not duration — which is exactly the pattern a year of informal review had missed.
Reinforcement got built into the systems, not into a policy memo. The auto-approval rule lives in the ERP via API; the approval-routing agent and the CRM↔ERP sync are software, not a new habit someone has to remember. That's why the gain didn't reverse the week the engagement closed.
The verdict stayed honest about where the gain came from. Floor cleared, Stretch hit exactly — but the scorecard didn't credit "the method" for all of it. Roughly half came from ordinary atomic-unit discipline, half from the automation the scheduled AI scan triggered. Naming that split, instead of collapsing it into one undifferentiated win, is what keeps the number credible.
One 90-day cycle. A distributor with four suspected fixes, no way to tell a real lever from an office opinion, and eighteen days of rework nobody could see until a GenAI pass — scheduled from Day 10, not improvised mid-cycle — went looking for repetition instead of duration. The methodology doesn't promise every dimension moves equally — it hands over a traceable chain from metric to atomic unit to automated fix, and a target that was aggressive on purpose and still landed on the number.