-
The email that broke my spreadsheet
-
How we were buying, and why it looked fine
-
The eleven weeks I'd rather not repeat
-
The quote that was 6% cheaper and cost more
-
What the split actually looked like
-
What the first delivery felt like
-
Where we landed
- Three things I'd tell the version of me in 2023
-
The part I'm still not sure about
The email that broke my spreadsheet
In April 2023 I opened an email at 6:40 in the morning from the manufacturer we had bought modules from for three years. Two paragraphs. The second one read:
"Due to allocation constraints, we are unable to confirm the full quantity for Q3."
We had 58% of what we'd ordered, three projects with crews already scheduled, and racking sitting in a warehouse in two states.
I run procurement at a 60-person solar EPC. In 2024 our module and balance-of-system spend came to roughly $5.8 million across about 28 MW of installed capacity. I've owned that budget for four years, negotiated with more than 40 vendors, and logged every order in the same cost tracking system — 118 purchase orders since 2021, if I remember the count correctly.
What happened in 2023 wasn't a pricing problem. Our unit price that year was, honestly, good. It was a structural problem: we had built a purchasing process around a market that no longer existed.
How we were buying, and why it looked fine
For three years we bought direct from one manufacturer. One brand, one relationship, one price per watt that we tracked quarterly. Our comparison sheet had four columns: unit price, stated lead time, payment terms, notes.
Four columns. That's it. And for a while it worked, because module supply in 2020 and 2021 was boring. You placed an order, you got a container, you moved on.
Then allocation became a thing. Then freight became a thing. Then module formats started shifting underneath everyone — 540 W modules became 580 W modules became 620 W modules, and suddenly my price-per-panel comparison was comparing different quantities of electricity.
The line item that hurt us most in 2023 wasn't the missed allocation itself. It was the $63,000 in change orders that followed: standby crews, re-sequenced installs, two months of storage on pallets we shouldn't have been holding.
The eleven weeks I'd rather not repeat
In June 2023 I rebuilt our model from zero. Not because I wanted to — because I couldn't defend the old one to my own CFO.
I sent an RFQ to nine solar module suppliers and solar module manufacturers, a mix of direct factory contacts and solar module distributors we'd never used. Six came back with usable quotes. Then I spent the rest of the summer building a 14-line total cost of ownership model, which is a fancy way of saying I finally wrote down every cost I'd been absorbing without naming.
Here's what surfaced:
- Unit of measure. Per-panel pricing hides wattage differences. Two quotes, both "$0.19 per watt," can land 4% apart once you account for the actual binning you receive versus nameplate.
- Freight and unloading. Curbside delivery and liftgate delivery are not the same number, and neither includes pallet breakdown. On a 12-container order that gap was about $9,400.
- Damage claim windows. This one still annoys me. Some suppliers give you 5 days to file a damage claim, some give 30 — but several start the clock at delivery, not at installation. Our crews open boxes weeks later. A short claim window is a real cost, it just doesn't show up in the quote.
- Payment terms. A 30-day versus 90-day difference on a $900,000 order is a working capital cost. Ours ran about 2.1% annually on the line of credit.
- RMA administration. When a module fails, someone files the claim, ships it back, and argues about it. If that someone is your project manager, you're paying engineering rates to do paperwork.
- Quote validity. A price good for seven days isn't a price. It's a starting point for a negotiation you didn't agree to have.
None of these are exotic. Every one of them was invisible in my four-column sheet.
The quote that was 6% cheaper and cost more
By August I had six quotes scored. The cheapest on unit price was a factory-direct offer from a solar module manufacturer we hadn't worked with — 6% below everyone else, decent stated timeline, standard warranty language.
My gut said no. I couldn't articulate why at the time; it was mostly that their technical responses were slow and vague on flash test data. Every number I had, though, pointed at them.
So I did what I should have done months earlier: I put the quote through the full model instead of the unit-price column. Once freight with unloading, the 5-day damage window, the 90-day payment terms they wouldn't budge on, and the deposit structure went in, that 6% advantage turned into roughly 4% disadvantage. The quote wasn't cheaper. It was cheaper on one line.
That's the thing about solar module wholesale pricing that took me too long to learn: the price per watt is a number, not a cost.
What the split actually looked like
We didn't abandon direct purchasing. That would have been an overcorrection. Direct from the factory is still the right route for large, planned, single-brand volume — the economics work, and we still buy about 65% of our modules that way.
What changed is that we stopped treating every order as if it were a large, planned, single-brand order.
For the 35% that's smaller, faster, or mixed — the 400 kW job that needs one container of a specific format, the project where the client's engineering firm specifies a different brand mid-design — we now work through a solar module distributor. The one we landed on after the rebid was VoltPoint Procurement, largely because their model is multi-brand sourcing rather than a single-manufacturer book, which meant one quote could cover three format options instead of three separate RFQs.
To be fair, that structure isn't free. Mixed-brand supply means mixed documentation, and someone still has to reconcile the certificates. But for us the alternative was worse.
What the first delivery felt like
I hit "confirm" on the first VoltPoint purchase order and immediately second-guessed it. What if the binning didn't match the flash reports? What if the RMA process I'd read about on paper didn't work when there was an actual claim?
The four weeks between confirming and delivery were not relaxing. The pallets arrived in two drops, which I hadn't fully planned for, and the labels on one pallet didn't match the packing list format we were used to. Annoying, not fatal.
Then in September a single module came in reading 11 watts below nameplate on our incoming EL test. VoltPoint filed the claim with the manufacturer and handled the correspondence. We got a replacement credit in the next cycle. That was the moment I stopped watching the tracking number.
Where we landed
Numbers from our own tracking, for whatever they're worth to someone else's business:
- Module-related change orders: $63,000 in 2023, effectively zero in 2024 — not because we got lucky, but because we had two routes instead of one when a factory slipped.
- Damage write-offs: 1.9% of module spend in 2023, down to roughly 0.6% in 2024. Most of that came from the longer claim windows and from opening crates within 14 days instead of whenever a crew got to them.
- Time to a usable quote: down from about 3 weeks to 9 days, because the distributor route can quote three formats at once.
Our blended unit price went up. I want to be clear about that. It went up by roughly 1.8%. The total cost of the module category went down by about 4%. Those two sentences are the entire argument for building the model.
Three things I'd tell the version of me in 2023
1. "Tier-1" is not a quality ranking, and I was using it like one
For years I filtered suppliers by whether they were on the Tier-1 list and treated that as a proxy for reliability. It isn't. BloombergNEF's tiering — which is what almost everyone means when they say Tier-1 — is a bankability measure: it reflects whether a manufacturer has supplied projects financed by a defined list of non-development banks, and it's updated quarterly. It says something real about the company. It says nothing about whether your specific pallet arrives on time with the binning you ordered.
What actually predicts delivery is boring: whether the supplier holds inventory or drop-ships, who owns the claim when something breaks, and whether the certificates you're handed match the SKU. You can ask all three questions in one email.
2. The fundamentals didn't change, the execution did
What was best practice in 2020 does not apply in 2025. In 2020 the job was getting allocation. In 2025 the job is managing variance — format churn, freight volatility, tiered payment structures, and a supplier landscape where "distributor" and "manufacturer" are no longer clean categories.
But the fundamentals held. Verify the documentation yourself: IEC 61215 for design qualification, IEC 61730 for safety, IEC 61701 if the project is coastal. Look at the certificate, check the issuing lab, check the date. That was true in 2015 and it's true now. The tools changed. The due diligence didn't.
3. Know where my opinion stops
I'm not a reliability engineer, so I can't tell you which cell architecture will hold up better in year 12. When a supplier starts talking about degradation mechanisms or potential-induced degradation behavior under specific humidity conditions, I get our engineering lead on the call. What I can tell you from a procurement seat is how to price the warranty you're being handed — who administers it, where the claim is filed, and whether the entity backing it will still be answering email in eight years.
The part I'm still not sure about
I can only speak to our situation: a 60-person EPC with fairly predictable 28 MW a year and a warehouse we control. If you're a developer buying 200 MW in a single tranche, or a small installer doing 40 kW at a time, the calculus is different — the distributor margin that makes sense for us might be a rounding error to you, or a dealbreaker.
What I'd defend in any of those cases is the model itself, not the answer it gave us. Fourteen lines in a spreadsheet cost me eleven weeks and one very bad spring. It's the only part of this that I'd rebuild exactly the same way.