The Spreadsheet That Was Costing $180K a Year

The Problem

The company had been in business for 22 years. Their two dispatchers, Ray and Tom, had been there for 15 and 11 years respectively. Every morning, they sat down with Excel, a whiteboard, and a pot of coffee, and spent three to four hours building the day's routes by hand — balancing delivery windows, truck capacity, driver hours, and customer priorities based on experience and memory.

"Ray and Tom were amazing at what they did. They could look at 80 stops and build routes in their heads that nobody else could. But the problem was, it all lived in their heads. If one of them was sick, we were in trouble. And even on their best day, they couldn't factor in real-time traffic, weather, or capacity constraints the way a computer can."

— Bill Hartley, Owner

Demand forecasting was similarly manual. Bill and his sales manager would review last year's numbers in a spreadsheet, adjust up or down based on what they were hearing from customers, and order inventory accordingly. The result was a forecast accuracy of about 60% — which meant they were either sitting on excess inventory or scrambling to expedite shipments when demand spiked unexpectedly.

The Engagement

Senior AI consultant Ryan Cooper led the discovery phase, spending two weeks embedded with the dispatch team, warehouse managers, and drivers. He rode along on delivery routes, sat in on morning planning sessions, and pulled three years of order history, fuel data, and delivery performance records.

"The first thing I noticed was how much tribal knowledge was running this operation. Ray could tell you that Customer X always needs an extra pallet on Tuesdays in March, or that Route 7 adds forty minutes if you run it after 2pm because of school traffic. That's incredible institutional knowledge — but it doesn't scale, and it's walking out the door when Ray retires."

— Ryan Cooper, Senior AI Consultant

What was built

1. Dynamic route optimization. An AI routing engine that generates optimized routes every morning using live traffic data, delivery windows, truck capacity, driver hours-of-service constraints, and historical delivery performance. Dispatchers review and approve in 20 minutes instead of building from scratch in 4 hours.
2. Demand forecasting model. A machine learning model trained on three years of historical order data, combined with external signals: weather patterns, day-of-week seasonality, customer purchase cycles, and regional economic indicators. Produces 2-week rolling forecasts with 85% accuracy.
3. Warehouse pick optimization. AI-optimized pick paths and batch picking logic for the three warehouses, reducing average pick time by 23% and improving order accuracy from 96% to 99.5%.

The Surprise: The Weather Pattern Worth $180K

The team expected the route optimization to save fuel. What nobody expected was the demand forecasting model's first insight.

Within the first 30 days of running the model, Carlos's team noticed something in the forecast data: the company's largest customer — a regional HVAC distributor accounting for 18% of total revenue — had an ordering pattern that correlated strongly with weather data. Specifically, when the 10-day forecast showed temperatures crossing certain thresholds (above 88°F or below 25°F), orders from this customer spiked predictably 2-3 weeks later.

The pattern had been there for years — buried in the spreadsheets — but it was invisible when you were looking at month-over-month totals. The AI found it because it was designed to look for exactly these kinds of correlations.

"When Carlos showed me the weather correlation chart, I literally laughed. Twenty-two years in this business and I never connected those dots. We'd been scrambling every time the HVAC orders spiked — paying for rush freight, pulling drivers off other routes, burning overtime in the warehouse. And the whole time, the data was telling us exactly when it was going to happen. We just weren't listening."

— Bill Hartley, Owner

By pre-positioning inventory two weeks before predicted spikes — instead of scrambling after orders arrived — the company eliminated nearly all emergency expedited shipments. The savings: approximately $180,000 per year.

Account manager Tessa Franklin helped the company recalculate their ROI projections after the weather discovery.

"Bill's original ROI case was built entirely around fuel savings. The weather pattern wasn't even on the radar. When we added the expedited shipping elimination, the demand-driven inventory reduction, and the warehouse pick improvements, the total first-year ROI was over 400%. Bill said it was the best investment he'd made since buying his third warehouse."

— Tessa Franklin, Account Manager

The Results

Metric	Before	After	Change
Route planning time	3-4 hours/day	20 minutes/day	-90%
Fuel costs	$1.2M/year	~$984K/year	-18%
Empty miles	22%	12%	-45%
Forecast accuracy	~60%	85%	+42%
On-time delivery	82%	94%	+15%
Stockout rate	8-12%	2-3%	-75%
Warehouse picks/hour	100	135	+35%
Emergency shipments	$180K/year	Near zero	Eliminated

What Happened to Ray and Tom

One of Bill's biggest concerns going in was what would happen to his dispatchers. Ray and Tom had built their careers on route planning expertise. Would AI make them obsolete?

The opposite happened. Freed from the daily grind of building routes by hand, Ray and Tom became the company's operations strategists. Ray now focuses on customer relationship management and service level optimization — using the AI's delivery performance data to have more informed conversations with key accounts. Tom manages the dispatch exception queue and trains new drivers using AI-generated route analytics.

"I was worried I was automating my best guys out of a job. Instead, I automated the worst part of their job. Ray told me last week that he's having more fun at work than he has in ten years because he's actually solving problems instead of staring at a spreadsheet every morning."

— Bill Hartley, Owner

Why It Worked

Logistics is one of the highest-impact verticals for AI because the problems are highly structured and data-rich. Route optimization is a mathematical problem with known constraints. Demand forecasting improves with more data. Warehouse picking follows measurable patterns. The ingredients for AI success were already in place — they just needed to be connected.

The critical factor in this engagement was data quality. The company had 20 years of order history in their ERP system and three years of GPS tracking data from their fleet. That data had been sitting unused — reports were run monthly, if at all — but it contained patterns that were worth hundreds of thousands of dollars once someone knew how to extract them.