A service center lost a $400,000-per-year account last quarter. The sales rep was blindsided. "They never said anything was wrong." But the data told a different story. The account's order frequency had dropped from weekly to biweekly six months earlier. Average order size had declined by 30% over the same period. They had started ordering fewer product categories, dropping two of the five product lines they historically purchased. Every one of these signals was visible in the system. Nobody was looking.
The Warning Signals
Customer churn in steel distribution rarely happens overnight. It follows a predictable pattern that unfolds over 3 to 6 months. The signals are measurable if you know what to track.
Order frequency decline is the earliest indicator. A customer who ordered every week for two years and suddenly shifts to every other week is testing an alternative supplier. They have not left yet, but they are splitting their volume. A 25% or greater decline in order frequency over a 60-day period should trigger a flag.
Order size decline follows frequency decline. Even if the customer is still ordering regularly, shrinking order sizes mean they are buying less from you and more from someone else. A 20% decline in average order size over 90 days is a warning.
Product mix narrowing is the most ominous signal. A customer who used to buy HRC, CRC, plate, and tube from you and now only buys HRC has moved three product categories to a competitor. They are keeping you as a backup on HRC while their new supplier handles the rest. You are one phone call from losing the HRC too.
Quote activity without orders means the customer is using your quotes as a benchmark to negotiate with someone else. If your quote-to-order conversion rate with a specific account drops below 10%, they are shopping you, not buying from you.
Building an Early Warning System
You do not need AI or advanced analytics for this. You need four reports run weekly. A frequency report that shows each customer's order count this month versus their 6-month average. Flag any account where the current month is tracking below 75% of their average. A volume report that compares each customer's tonnage this month to their 6-month average. Same threshold: flag below 75%. A product mix report that shows how many distinct product categories each customer ordered this quarter versus last quarter. Flag any reduction. A conversion report that shows quote-to-order rates by customer. Flag any account below 20%.
These reports take 30 minutes to build in any modern system and 15 minutes to review each week. That 15-minute weekly investment is the difference between catching a declining account in month 2 and discovering it in month 8 when the customer is already gone.
What to Do With the Flags
When an account is flagged, the sales rep makes a call within 48 hours. Not a check-in call. A specific conversation: "I noticed your orders have slowed down over the last couple of months. I want to make sure there is nothing on our end that is causing that. Are we meeting your needs on quality, delivery, and pricing?"
The customer will usually tell you what changed. Common responses: "Your competitor offered a better price on CRC so we moved that over." "Your delivery was late twice last month and we cannot afford that." "We changed our production schedule and need Friday deliveries, which you do not offer." Each of these is fixable if you catch it early enough.
If the customer says everything is fine but the data shows continued decline, they are being polite while leaving. Escalate to a management-level conversation. Have the sales manager or VP of Sales call with a more direct approach: "Your business is important to us and I want to understand what we need to do to earn it back."
You will not save every at-risk account. But catching the decline early and making a proactive effort saves enough of them to materially impact your retention rate. A 5% improvement in customer retention, from 80% to 85%, translates to hundreds of thousands of dollars in preserved revenue for a mid-size service center. The data to make it happen is already in your system. You just need to look at it.