Category: ERP

Six weeks after go-live at a 240-person equipment manufacturer, the usage report looked great: 180 of 190 licenses active. The plant director was pleased. Then I sat in the scheduling meeting.

The planner opened a spreadsheet on the second monitor. “I check the system,” she said, “then I do the real schedule here.” Six weeks in, the system was a reporting tool nobody trusted.

The difference between logging in and using

Logins count activity. They don’t count decisions. The reliable measure is simple: how many decisions in a week were made from system data, versus from a personal file?

We asked 12 people one question — “show me the screen you looked at before your last real decision.” Nine pulled up a spreadsheet. That is your adoption number, and it was 25%, not 95%.

• Pick one decision per role and track whether it was made in the system this week. Planner: the schedule. Buyer: the expedite list. Supervisor: the labor assignment.
• Delete or rename the old spreadsheets. Not blocked — renamed. A file called “OLD do not use” still gets used.
• Have the plant manager run the meeting from the system screen, projected. Two weeks of this changes more behavior than any training class.
• Name a process owner per department who is measured on usage, not on project tasks.

The training mistake we made

We had trained 190 people in four days, back to back. Retention after two weeks was close to nothing, because nobody used the screens in between. Training without a live task is entertainment.

The second attempt was different. Each person got one real transaction from their own queue, with a coach beside them, in their own work area. Ninety minutes of doing beat two days of watching by a wide margin. Weekly active use went from 62% at week six to 94% by week fourteen.

A 30-day plan that works

1. Week one: name one process owner per department and give them a single usage number to move.
2. Week two: run the “show me your screen” check with eight to twelve people. Write down what you find.
3. Week three: replace classroom sessions with in-queue coaching, one real transaction per person.
4. Week four: run one management meeting entirely from the system and let people see the data being trusted.

Adoption is a management behavior, not a training deliverable. If the plant manager still asks for the spreadsheet version, everyone below will keep making it. The fastest fix in ERP Consulting is often to stop printing the old report.

Saturday, 6:40 a.m. Eleven thousand SKUs, 2,400 bin locations, one warehouse manager holding a clipboard and a printed pick list. The system switch was scheduled for Sunday midnight.

By 9 a.m. we knew the count was off by 6% in the fast-mover zone. At a food-grade packaging distributor, 6% wrong in the wrong aisle means trucks leave short. This is the part of ERP Consulting nobody puts on the slide deck.

The Thursday decision that saved the weekend

Two days earlier we had set a go/no-go rule in writing: if fast-mover accuracy was below 97%, we would freeze the top 300 SKUs manually and switch anyway, with a paper backup for one week.

Everyone hated the paper backup. It felt like admitting failure. It also meant the warehouse could keep shipping while the team fixed counts in parallel. Having a pre-agreed fallback is what makes a cutover survivable, not heroics at 2 a.m.

• Define go/no-go thresholds in writing at least 10 days out, with the specific numbers and who decides.
• Freeze transactions in the last 24 hours for physical moves only. Receiving and shipping continue; put-away waits.
• Print the top 200 to 300 SKUs by volume as a paper fallback. Two hours of printing beats a day of lost orders.
• Staff the floor with two people per zone who know the product, not just the system.

What went wrong anyway

The receiving dock had three trucks booked for Monday morning, and the new system required a purchase order match that two suppliers never sent. The receiving clerk did what any sensible person does under pressure — she keyed the receipt against a fake PO.

We caught it Tuesday on the exception report. The lesson is not “train harder”. The lesson is to give people a sanctioned workaround. We added an “unmatched receipt” path with a 24-hour cleanup queue. Keyed errors dropped from 40 a week to 6.

The checklist we now use for every cutover

1. Ten days out: agree the go/no-go numbers and write them on one page.
2. Five days out: freeze physical moves, run a full fast-mover count, print fallback picks.
3. Two days out: walk the floor with the warehouse lead and test three real orders end to end.
4. Day one after go-live: sit a person at the dock for the entire shift. Do not leave it to a help desk ticket.

The warehouse hit 99.1% accuracy six weeks later and the paper backup was retired in nine days. The habit worth copying is the written go/no-go page, signed before anyone is tired. That single document is the cheapest insurance in ERP Consulting.

Ask a controller how long the close takes and you’ll usually get a defensive answer. At a 3-plant components maker, the honest number was 18 working days. The board saw the pack on day 22.

Everyone assumed the problem was the accounting team. It wasn’t. Eight of those 18 days were spent waiting for information that existed somewhere else in the business.

Where the 18 days actually went

We time-stamped every task for one close cycle. The pattern was brutal and repeatable. Inventory valuation waited 3 days on a receiving file from the warehouse. Accruals waited 2 days on purchase order confirmations sitting in email. Intercompany matching took 4 days because two plants keyed the same transfers differently.

Only 4 days were real accounting work. The other 14 were queue time. Nobody had ever measured it, so everybody blamed the wrong thing.

• Time-stamp the close for one cycle. Log each task’s start, block, and finish. You will find 60-75% of the elapsed time is waiting, not working.
• Move the three worst blockers to a daily rhythm instead of a month-end rhythm. Receiving confirmations, for example, get cleared every morning, not on day 3.
• Set a cutoff rule with teeth: transactions after the cutoff post to next month, no exceptions, no “just this once” from the sales director.
• Assign one owner per blocker with a name, not a department.

The intercompany fix that saved 4 days

Two plants were recording the same transfer with different reference formats — one used the sales order number, the other used an internal transfer ID. Matching them by hand took 4 days every month.

The fix took one afternoon of argument and a one-page standard. Same reference format, same timing rule, one shared reconciliation view. The next close matched in 3 hours.

What day 4 looks like in practice

The close now runs on a 4-day calendar with a published schedule. Day 1: subledgers and cutoffs. Day 2: inventory and accruals. Day 3: intercompany and reconciliations. Day 4: review and pack.

The controller got 14 days back per month. She used them on margin analysis by customer, which found $190,000 of low-margin orders that had been quietly renewed for two years. That is the real payoff of a faster close — not the speed, but what you can finally look at.

1. Run the time-stamp exercise on your next close. One spreadsheet, three columns, no new software.
2. Circle the top three waiting periods and ask who owns each one by name.
3. Move those three to a weekly or daily routine and measure the next two closes against the old baseline.

If your close is over 10 days, resist the urge to add staff. You almost certainly have a handoff problem, not a capacity problem.

A distribution company with $84M in revenue asked for one small thing during design: their pricing screen should show the customer’s last three orders. Reasonable request. It took four days to build.

Eighteen months later we counted the change log. 210 modifications, 62 custom reports, and a $310,000 annual bill just to keep them working through upgrades. The original request had cost about $2,400.

How small requests become permanent overhead

Nobody signs off on $310,000. They sign off on 210 decisions of $1,500 each, spread over a year, each one approved by a different manager who never sees the total.

The pricing screen was the gateway. Once the team learned that screens could be changed, every department had a list. The warehouse wanted a custom label. Finance wanted a custom aging report. Sales wanted a custom quote PDF. Each one was small. Together they made the system un-upgradeable.

• Put every change request on one shared log with an estimated annual maintenance cost, not just a build cost. Multiply build hours by 0.25 for the yearly upkeep.
• Require the requester’s manager to sign the maintenance line, not just the project line.
• Set a hard ceiling before design starts — for example, no more than 15 modifications in year one — and publish it.
• Review the log in a 30-minute monthly meeting with the finance controller present. Visibility alone kills about a third of requests.

The 20-minute test that filters most requests

Before anything gets built, ask the requester to run the process manually for 20 minutes using the standard system plus a spreadsheet. Two things happen. About half the requests disappear because the standard way is fine once someone actually tries it. The other half come back sharper and cheaper.

On the distribution project, we re-ran the pricing request this way. The real need was not a screen. It was a monthly price review meeting with a one-page exception report. We built a report instead of a screen and saved 200 hours of upgrade work over three years.

What a healthy change log looks like

Aim for under 20 modifications for a company of 200 to 400 users. Above 40, your upgrade path gets expensive and slow. That is the line where ERP Consulting pays for itself — not by saying no, but by pricing the yes.

1. Export your change log and add two columns: estimated annual maintenance hours and who approved it.
2. Sort by maintenance hours. The top 10 items are your candidates to retire at the next upgrade.
3. Book a 45-minute session with the top three requesters and ask what decision the customization supports. If nobody can name one, retire it.

One counterintuitive rule worth trying: make the requester write the help-desk ticket template for their own customization. People who have to support their own idea tend to scope it tighter.

Three weeks before go-live, the materials manager at a 280-person pump manufacturer printed the item master. It came out at 6,900 active part numbers. She expected about 2,900.

Nobody had done anything wrong on purpose. Ten years of “just add a new number, it’s faster than fixing the old one” had quietly piled up. In ERP Consulting, this is the single most common reason a project date slips.

What 6,900 part numbers actually costs you

We pulled a sample of 400 records and had two buyers classify them by hand. 1,900 of the 6,900 were true duplicates — same spec, same vendor, different number. Another 640 were dead: no purchase, no sale, no movement in 24 months.

The damage shows up in odd places. Roughly 11% of purchase orders in the prior year had gone to a secondary vendor record for the same part, at an average 7% higher price. That is real money, and it never appears on a project risk log.

• Freeze new item creation 60 days before go-live. One form, one approver, one reason code.
• Sort records by spend, not by count. The top 20% of parts by spend usually hides 80% of the duplicate damage.
• Give the cleanup team a hard target: 120 records per person per day, with a second person spot-checking 10%.
• Publish the kill list — the 640 dead numbers — to purchasing and sales before you delete anything.

The part that most teams skip

Cleaning the data is the easy half. The hard half is changing the behavior that created it. If a buyer can still create a part number in 40 seconds with no check, you will be back to 6,900 within three years.

On this project we added one rule: any new item request had to name an existing similar item and say why it wasn’t good enough. Requests dropped from about 90 a month to 12. That single form saved more money than the entire cleanup.

What to do this week

1. Export your active item list and count how many records had zero transactions in the last 12 months. If it’s above 8%, you have a cleanup project, not a data migration task.
2. Pick 50 duplicate-looking pairs and check the last purchase price on each. Multiply the gap by annual volume. That number is your business case.
3. Write the one-page new-item rule before go-live, not after. Post-dated rules get ignored.

The lesson from this go-live: data quality is not an IT chore, it is a purchasing negotiation. Budget 3 weeks and 5 people for a company this size — and put the freeze in place before anyone opens the migration tool. Good ERP Consulting always starts with the messy list, never with the software.

The owner of a 120-person plastic components manufacturer told me, with real confidence, that his margins were healthy at around 22%. Two weeks later the costing review showed that 34% of his active products were selling below full cost, and 12% were below the raw material price.

He wasn’t lying. He was reading a report built on numbers that hadn’t been true since 2019.

The quote that hadn’t moved in four years

We pulled every active product and compared the standard cost in the system against what the last six months of production actually consumed.

For 61 of about 900 SKUs, the gap was over 15%. Those were mostly small, awkward parts — the ones nobody wanted to quote, so they’d been priced years ago using an old material cost and a scrap rate of 2%, when the real rate was closer to 11%.

Nothing about this is unusual. Resin prices moved, machine speed assumptions drifted, and the products that get the least attention are the ones quietly bleeding.

Standard cost versus actual: the gap that hides in plain sight

Most plants track material variance closely and treat labor and overhead as a rounding exercise. That’s backwards in a business with high setup times.

On a part with a 45-minute changeover and a 6-minute cycle, setup is a bigger cost driver than the resin. If your standard cost assumes setup is absorbed across the batch, and you actually run batches of 200 instead of 2,000, the true cost of that part is roughly 40% higher than the system says.

Batch sizes shrink, but standard costs don’t follow. That’s where the money hides.

A three-step costing reset you can do this quarter

1. Pick the 50 SKUs with the highest revenue and the 50 with the highest volume. Ignore the rest for now — that’s 10% of the file and most of the money.
2. For each, pull actual material consumption and actual machine hours from the last 12 months, then rebuild the cost from scratch on one spreadsheet. No system changes yet.
3. Sort by margin and look at the bottom 10. Ask one question per part: raise the price, change the batch size, or stop selling it?

What they did about the losers

They discontinued 41 products — 4% of the file. Those parts had consumed 15% of machine capacity and generated a combined margin of almost nothing. Freeing that capacity let them bring in two new accounts without buying a machine.

Another 28 got a price increase of 6% to 12%, and 19 of those held. Two customers left. The revenue lost was $340,000; the margin gained was $410,000. That’s the kind of trade a lot of owners can’t bring themselves to make until they see it on one page.

If you’re bringing in ERP Consulting to work on profitability, insist that the first deliverable is a rebuilt cost for your top 50 parts — not a dashboard. Dashboards built on wrong costs just make the wrong answer look credible.

This week, pick your five lowest-margin products and rebuild the cost from actual consumption. You’ll either find a pricing decision you’ve been avoiding or a process assumption that’s costing you real money. Both are worth the afternoon.

The go-live party had catering. There were t-shirts. Fourteen months and a lot of money later, six weeks into “live,” I sat in a meeting where a warehouse supervisor admitted she kept her old spreadsheet running because she didn’t trust the new system and nobody had asked her opinion.

She wasn’t being difficult. She was being rational.

What the project plan never mentioned

This was a 250-person specialty chemicals distributor. The project plan was thorough — 340 tasks, weekly steering meetings, a color-coded RAG report. What it didn’t contain was a single line about who would do what differently on the Monday after go-live.

The process design had been built by a project team of nine, seven of whom were managers. The people who actually processed orders, receipts and returns saw the system for the first time in a two-hour training session, four days before launch.

Nobody had asked a single order-entry clerk how the old process worked. Three of the new screens required information the clerk didn’t have at the moment of entry, so she learned to type placeholder values. Those placeholders are now in your data.

Three signs in month one everyone ignored

• Order entry times rose from 4 minutes to 11. The project team called it a learning curve. It didn’t improve, because the screen order didn’t match how customers actually placed orders.
• Nobody escalated anything in the first three weeks. That’s not satisfaction — that’s people quietly giving up and going around the system.
• Two resignations in the operations team, both good people, both citing “the new system.” Attrition is the most expensive bug report you’ll ever get.

The recovery: 90 days of undoing

They stopped the rollout to remaining sites and did something painful — they sat four end users in a room with the process design and asked what they’d change. That list ran to 60 items. Half got fixed within six weeks, mostly screen order, mandatory fields that shouldn’t have been mandatory, and one report that had to print the way the old one printed because the customer’s receiving dock expected it.

Six months later, adoption was measurable: 94% of orders entered in the system, and average entry time down to 5 minutes. The win wasn’t the software. It was that people had been given a reason to trust it.

Any ERP Consulting engagement worth the name will spend more time on this than on configuration. Configuration is knowable. Whether a supervisor at 6 a.m. trusts the screen is not, and it decides your outcome.

The lesson worth stealing

1. Before you design a process, spend two full days sitting next to the person who does it today. Not in a workshop. At their desk.
2. Pick your two most skeptical users and put them on the design team. They will find more problems than any consultant, and once they’ve fixed something themselves, they’ll defend it.
3. Set a go-live metric that isn’t technical: order entry time, first-time-right rate, or the number of people still using a personal spreadsheet. Then track it weekly for 90 days.

The hard truth is that this company didn’t fail because of the software. They failed because they treated a change in how people work as a change in what software they use.

The argument in the meeting was between the sales director and the credit manager, and it was not really about customers. It was about one account that existed four times in the system, with a $50,000 credit limit on each. The customer had figured it out. Nobody else had.

How you get 900 duplicates without noticing

This was a B2B industrial supplier, about $70 million in revenue, 4,300 customer records. We matched them on name, address, phone and tax ID. About 900 were duplicates — 21% of the file.

None of it was deliberate. A rep types “Acme Industrial Supply Co.” and creates a new record because the search didn’t find it. A web order arrives and the system builds a fresh account. A branch buys a competitor’s customer list and imports it with slightly different formatting. Three years of that and your customer file is a museum of good intentions.

The real damage: credit exposure was understated by roughly $240,000, and the same customer got quoted four different prices by four different reps in the same month.

Who owns a customer record?

This is the question that decides whether the cleanup lasts. In most companies nobody owns it, which is exactly why it rots.

Here they made one decision that mattered: sales owns the relationship, finance owns the record. A rep can request a new customer, but it’s created by one named person in finance within one business day, after a mandatory duplicate check.

The one-day promise is the whole trick. If creating a customer takes a week, reps will work around the process every time — and they’d be right to.

The rules that prevented round two

• Merge duplicates in batches of 50, never in one giant run, and reconcile the open order and invoice history for each before merging.
• Make the tax ID a required field. It’s the one identifier that’s hard to fake and easy to match.
• Search by phone number and postcode, not just name. Half the duplicates we found had names spelled differently and identical phone numbers.
• Block account creation from the web store and route it to finance for a 24-hour match.

What good looks like six months later

Duplicates dropped to under 40, and the residue gets cleaned monthly in a 30-minute review. Price consistency across accounts improved enough that the pricing team found $610,000 of annual revenue leaking through inconsistent quotes to the same customer.

That’s the part people don’t expect. A customer master cleanup looks like housekeeping and turns into a margin story. When ERP Consulting work is scoped narrowly as data migration, this benefit usually gets left on the table.

Do this today: take your 50 largest customers by revenue and search each one by phone number in your system. Count how many records come back. That number, multiplied across your whole file, is your real starting point.

A 180-person contract manufacturer of metal enclosures was running at 68% on-time delivery. Customers were escalating, the sales team was discounting to keep accounts, and the production planner had developed a personal filing system of sticky notes to remember what the system got wrong.

Everyone assumed they needed better scheduling. They needed better data.

Late orders usually wear a scheduling costume

We pulled 90 days of late shipments and asked a simple question for each one: what was the first thing that went wrong? Not the last — the first.

The answer was uncomfortable. Forty-one percent of late orders traced back to a component that was short because the bill of materials was wrong. Not late from a supplier. Wrong on the drawing. A missing bracket, a quantity of 4 where the build needed 6, a part superseded two years ago that still lived in the system.

The plant was scheduling against a recipe that didn’t match the product it was building. No amount of clever sequencing fixes that.

The audit that took four days, not four months

Don’t audit everything. Pick the 30 products that make up the bulk of your revenue and go deep.

1. Take the top 30 products by shipped revenue. For each one, pull the last five work orders and compare the components actually consumed against the components the system said to consume.
2. Walk one physical build with the shop supervisor and the BOM in hand. Mark every discrepancy on paper, live. Two hours per product is realistic.
3. Classify each error: missing part, wrong quantity, obsolete part, wrong unit of measure, phantom assembly that shouldn’t exist. Count them by type.
4. Correct the top 30 BOMs first, then freeze them for 60 days except through a formal change request signed by engineering.

What changed on the floor

Within eight weeks, material shortages caused by data errors fell from 41% of late orders to 9%. On-time delivery moved from 68% to 89% over six months — with the same equipment, the same people and no new headcount.

The planner stopped keeping sticky notes. That sounds small. It meant she spent her mornings on capacity and constraints instead of re-checking quantities, and that shift alone recovered roughly a day and a half of planning time per week.

When ERP Consulting teams get called in for delivery performance, the conversation tends to jump straight to advanced planning tools. Ask instead for a 30-product BOM audit before anything else gets scoped. If the recipe is wrong, the schedule is fiction.

The one rule that keeps it fixed

Engineering changes are where BOM accuracy goes to die. This company now requires a signed change request plus a first-article check before any BOM line is edited. It slowed changes down by about a day and cut rework by far more than that.

Next Monday, pick your single highest-revenue product and walk one build against its BOM. If you find three or more discrepancies, you’ve found your quarter.

Ask a plant manager why so much spend goes outside contracted suppliers and you’ll usually hear about the system. In my experience, roughly seven times out of ten, the cause is a single approval threshold somebody set in 2018 and nobody has revisited since.

The $500 threshold that created 4,000 tiny purchase orders

A 400-person equipment manufacturer, about $60 million of annual spend, was drowning. Their buyers were issuing close to 4,000 purchase orders a year under $500 each. Average value: $187.

Every one of those needed a manager’s signature. The managers, sensibly, stopped reading them. So the approval existed on paper and did nothing except add two days of delay — while 62% of total spend went to suppliers with no contract and no negotiated price.

That’s the trap. Tight controls on small purchases, loose reality on big ones. The paperwork grows and the leverage disappears.

Three rules that cut off-contract buying by two-thirds

• Raise the approval threshold to $2,500 for anything from an approved supplier on a contract. Below that, buyers just buy. Admin cost per PO is typically $60 to $90 — you were spending $80 to control $187.
• Require a reason code for any purchase outside contract, chosen from four options: urgent breakdown, contract supplier out of stock, one-off specification, customer-mandated brand. Anything else gets rejected at entry, not three weeks later.
• Publish a monthly list of the top 20 off-contract suppliers by spend. Not as a punishment — as a buying opportunity. Four of them became contracted suppliers within a quarter, worth about $2.1 million of spend.

What surprised everyone

The buyers weren’t the problem. Once the reason codes existed, the biggest source of off-contract spend turned out to be maintenance, ordering parts at 6 a.m. from whoever would answer the phone. The fix wasn’t discipline — it was a stocked critical-spares cabinet with a simple reorder point.

This is the kind of thing a good ERP Consulting review surfaces in the first two weeks, because it isn’t hidden in the software. It’s visible in the purchase history the moment you sort by supplier and value.

Cash impact over the following year: payment terms on the newly contracted categories went from 30 days to 45, freeing roughly $480,000 of working capital. All of it from rules, not from a new module.

Start with a sort, not a meeting

Export every purchase order from the last 12 months. Sort by supplier. Then sort again by value, smallest first. Count how many POs sit under your approval threshold and what they cost you to process.

That one spreadsheet usually makes the case better than any presentation. Bring it to your next operations review and ask one question: what would we do with two extra days and $80 per order?