Case Studies

The problem

The company processed supplier invoices for independent restaurants and pulled out the line items, so restaurants could track costs by product and supplier. In March 2016, processing cost $1.67 per invoice. Line-item extraction cost another $1.67 and took 24 hours, so it wasn't offered to every customer yet. An invoice with line items cost $3.34 to process, all of what it brought in. Gross margin sat near 20%.

The root cause

Every line item was typed by hand, twice. Two processors each extracted the same invoice, a senior processor checked every line, and the batch was imported by hand. Product names and units were re-keyed for vendors the system already knew. Slow internet at the offshore team meant invoices moved in batches of 25 to 200, and only one person on the day shift was trained to verify them. Queues built up and a 24-hour lag became normal.

What we did

A one-page improvement plan put a cost on every step and set targets for extraction: $0.80 per invoice and a 4-hour turnaround. We built extraction into the app. The database already knew each vendor's products and units, so the app filled those fields and checked the extraction itself. A processor typed 3 fields instead of a whole invoice, and the manual check went away.

With no transit time, there was no reason to hold invoices, so they moved one at a time, first in, first out. We trained the night shift to verify, so two shifts worked the queue. Labor per invoice fell by two-thirds, and a second outsourced team and 3 of 5 US contractors were wound down. Extraction went live in July 2016, and cost per invoice was tracked for 90 days after.

The problem

After the company acquired three regional distributors, fulfillment fell apart. Fill rate dropped to about 75%. Shrink ran 17–18% for the month and hit 28% in one week, with full pallets of food thrown out. On-time delivery fell as low as 24%, about 22% of orders shipped with an error, and a physical count found errors in 90% of storage locations. Nightly replenishment jobs jumped from about 70 to over 300. The operation was burning about $1M a month in contribution margin.

The root cause

It was tempting to blame the crews. The gap analysis found something more precise: the warehouse had been physically rebuilt, with new racking, a new freezer, and hundreds of new locations, but the warehouse management system's logic was never updated. Old locations kept their old rules, new ones had none, and many had no dimensions entered.

So replenishment sent product to slots too small to hold it, left others empty, and scattered single SKUs across as many as 886 locations. The system recreated the mess faster than crews could clear it. There had also been no full inventory count of the acquired businesses, so counts weren't trusted and product was double-bought.

What we did

We ran two tracks at once, because neither worked alone. Track one stopped the bleeding by hand: daily triage to clear staging lanes and trailers, replenish the top 1,000 items, consolidate each SKU to one location, and cycle-count by aisle, with corporate staff trained and on the floor.

Track two fixed the cause. We rebuilt the WMS into a hybrid slotting model (fixed locations for fast movers, dynamic slots for the rest), re-zoned the facility to the new racking, and added logic to consolidate inventory instead of scattering it. Then we added barcode scanning with QA hard-stops across receiving, picking, loading, and delivery, and wrote it all into daily standard work.

The problem

The company had found its market and planned to triple revenue. Asked for his top five goals for the year, the CEO listed six, then remembered a seventh. None were ranked. Two pre-offsite surveys named the same top blocker: prioritization. Five of twelve people asked leadership to start prioritizing, and four wanted to stop treating everything as urgent. Asked for one word to describe the culture, the CEO said "survival."

The root cause

The goals were real, but they lived in one person's head, unranked and not shared, so each function worked toward a different one. With no ranked list and no way to turn goals into owned work, "urgent" was the only prioritization signal the company had. It bought one-off exceptions that added complexity and left no time for the fixes that would have prevented them.

What we did

A four-day offsite, phones in a box. Two anonymous surveys, answered by most of the team, meant the room opened with the team's own words rather than the CEO's.

Seven goals came down to three, the number a team can actually carry. All three had to sit on the real bottleneck, which was collections, not the thirty AI agents on the wish list. A key result reading "test 10–30 agents" was struck out and rewritten as three agent workflows: deal desk, close, and collect.

"We've really gone from survival to thrival. I don't look at the bank balance every week and think, what's gonna happen next week."The CEO, six months later

The problem

First response on support tickets averaged 22 hours, which frustrated customers and drove churn. Requests came in over five disconnected channels, including a ticketing tool, calls to operations, messages to sales reps, and a 1-800 line, and about half were resolved outside any system. Only 10% came through the app, and only 10% were assigned to the right person automatically. Calls went to voicemail, and an executive still took 5–10 support calls a day.

The root cause

There was no structured front door. Customers couldn't say which order or issue a ticket was about, so a person had to read, categorize, and route every message. The most common questions (credits, returns, invoices, delivery delays) had no self-service path, and even a small credit needed escalation.

With several warehouses, agents had to search multiple systems to find which one filled an order. Customers called for delivery times on trucks the company already tracked by GPS. And with so much work outside the ticketing system, no one could see the real queue.

What we did

In the app, the customer picks the order and the issue, which creates a tagged, linked ticket. The system routes it to the right group and agent, with alerts for overdue tickets. Every channel feeds one system: a masked 1-800 line, in-app chat, and calling.

Agents see delivery and billing details on each ticket, including which warehouse filled the order, and can approve credits up to $50 per order and $500 per customer a year without escalating. Customers can see their invoices, credits, returns, and truck location in the app. Root-cause tags on resolved tickets go back to operations to fix what generates them.