What changed in a customer's process before the problem started?

This guide is for a supplier's technical service person working out what changed at a customer's site that uses a chemical product, such as a coolant sump, a cooling water system or a lubricated line. It does not replace an incident investigation after an injury or a chemical release, which follows the site's own procedure and the regulator's rules.

What do you need before you look for what changed in a customer's process?

Before you look for what changed in a customer's process, you need a measurable definition of the problem, a named contact at the site and access to every dated record that touches the system. The US Department of Energy guide to change analysis starts with the question "What is the condition?", and then asks what occurred before and after it (DOE-NE-STD-1004-92, 1992). Write the problem as a reading the site can repeat, such as a pH value or a reject count, not as "the product stopped working". Then collect the record sources: condition readings with their sampling point and method, product delivery and application logs, the certificates of analysis for the lots in use, maintenance and inspection logs, and the site's change records if it keeps any. Ask who on site knows about shift and contractor changes, which rarely sit in a file. If you will take samples yourself, follow the site's permit rules and the safety data sheet (SDS) for each product you handle. Agree that no deliberate changes are made while you look.

Step by step: finding what changed in a customer's process

Finding what changed in a customer's process follows four steps, and each step produces a record the next one uses.

1. Date the onset. Find the earliest dated sign of the problem, which is usually earlier than the complaint, and write it as "not later than" that date.

2. Collect dated records. Gather every dated reading, product application, supplier switch, maintenance entry and staffing change on both sides of the onset, and check each against the table below.

3. Compare before and after. Put the trouble-free period and the problem period side by side and list every difference, whether or not it looks relevant, as DOE change analysis recommends (DOE-NE-STD-1004-92, 1992).

4. Test each change. Keep a change as a candidate only when it came before the onset, has a plausible mechanism and matches where the problem appears; then confirm it by reversing or isolating it where that is safe.

An illustrative window is 60 days on each side of the onset; widen it when readings are sparse.

When did the problem in the customer's process actually start?

The problem in a customer's process usually started before anyone reported it, so the onset date is the earliest dated sign in the records, not the complaint date. Small drifts are slow to show: in the NIST/SEMATECH run-length table, a Shewhart chart takes on average 44 samples to signal a mean shift of 1 standard deviation of the sample mean, against about 371 samples between false alarms when nothing has shifted (NIST/SEMATECH, 2012). At one reading a week, an illustrative case, a drift of that size could run for months before a chart flags it. The same handbook notes that cumulative sum (CUSUM) charts are more efficient than Shewhart charts at detecting shifts of 2 sigma or less (NIST/SEMATECH, 2012), so replotting old readings as a CUSUM can move the onset earlier. Walk back to the first reading outside the site's normal band, then ask for the first rejected part or operator remark. The UK Health and Safety Executive tells investigators to record the chain of events in chronological order where possible (HSE HSG245, 2004). Write the onset as a date range when the records only bound it. The US DOE guide warns that a gradual change can be missed when the analyst looks only for an immediate one (DOE-NE-STD-1004-92, 1992).

Which records show what changed in the customer's process before the problem?

The records that show what changed in a customer's process are the ones with a date: condition readings, product application dates, supplier or raw material switches, maintenance entries and staffing changes. A record placed in time can be put before or after the onset; an undated one cannot, however detailed it is, and stays as background. The table below decides whether each record can place a process change before or after the problem's onset.

The key row is the product application date. A delivery date tells you when material arrived, not when it entered the system, and in an illustrative case a tote that sat unopened for three weeks moves the change by three weeks. The UK Health and Safety Executive gives a related warning about equipment: shop floor process and layout changes are a regular occurrence, so an entry without a precise equipment identifier can point to the wrong machine (HSE HSG245, 2004).

How do you compare process conditions before and after the problem started?

Compare process conditions before and after the problem started by putting a trouble-free period and the problem period side by side, using the same measurements from the same sampling points, and listing every difference between them. The US DOE change analysis method sets this out: consider a comparable activity that did not have the undesirable consequences, compare it with the problem, and "set down all known differences whether they appear to be relevant or not" (DOE-NE-STD-1004-92, 1992). Use periods of similar length and production, so a shutdown or a seasonal change in make-up water does not pass for a cause, and compare like with like: the same method on the same sump. Lay the differences out in one table with four columns, illustrative: item, before, after, date of change. Include items that look harmless; the DOE guide's own example is a change in color or finish that alters heat transfer and so the system temperature (DOE-NE-STD-1004-92, 1992). Compare across space too: a change that a healthy sister machine also received is a weaker candidate.

How do you check whether a process change caused the problem?

Check whether a process change caused the problem by testing each candidate against four questions: did it come before the onset, is there a plausible mechanism, does it match where the problem appears, and does the problem recede when the change is reversed or isolated. Timing is the cheapest test: a change dated after the onset is out, unless the onset is only a range. Mechanism comes next: write one sentence on how the change could cause the symptom, marked as reasoning. Scope separates candidates that pass the first two tests: a raw material switch that went to three lines but shows trouble on one points to something specific to that line. Reversal is the strongest test, but run it only when it is safe and agreed with the site, and change one thing at a time so the result can be read. The US DOE guide names two traps: the compounding of changes, such as one made five years earlier combined with a recent one, and a gradual change mistaken for an immediate one (DOE-NE-STD-1004-92, 1992). The same guide says change analysis may be adequate for a relatively simple condition but is generally not thorough enough to find all causes of a complex one (DOE-NE-STD-1004-92, 1992).

Why do process changes go unrecorded at customer sites?

Process changes go unrecorded at customer sites mainly because the formal change rules cover only some processes, and everyday adjustments fall outside them. In the United States, OSHA 29 CFR 1910.119 requires written procedures to manage changes to process chemicals, technology, equipment and procedures, except replacements in kind, for processes that hold a highly hazardous chemical at or above its threshold quantity or 10,000 pounds (4535.9 kg) or more of certain flammable liquids or gases (OSHA 29 CFR 1910.119, 2026). EPA's Risk Management Program rule carries the same requirement for Program 3 processes at covered stationary sources (40 CFR 68.75, 2026). A machining coolant sump or a cooling tower outside those thresholds has no such rule, so a new water source, a top-up habit or a different cleaner can enter without a record. Guidance fills part of the gap: OSHA's metalworking fluids manual recommends that all system additions be controlled and recorded by a designated person (OSHA Metalworking Fluids Best Practices Manual, accessed 2026). Staffing is the other gap: a new operator lives in the site's memory, not its files, so list what the contact reports and mark it as reported, not documented.

When should you call a specialist about a change behind a customer's process problem?

Call a specialist about a change behind a customer's process problem when several changes coincide with the onset, when the readings show a slow drift with no clear start, or when the problem involves a release, an injury or a process under a management of change rule. Coincident changes need designed testing, and a slow drift may need a statistician to fix the onset. For processes covered by OSHA 29 CFR 1910.119, the employer must start an incident investigation as promptly as possible and not later than 48 hours after an incident that resulted in, or could reasonably have resulted in, a catastrophic release (OSHA 29 CFR 1910.119, 2026); a supplier's change analysis then feeds that team rather than replacing it. Bring in the formulator for a raw material or lot difference, and a laboratory when a deposit or failed part needs identification. In Lubinpla, a customer site's condition readings and product application dates are kept by day in one project, so the period before and after a problem can be compared from the same record. Whoever you call, send the onset date range, the difference table and the result of each test.