The Spreadsheet That Said Everything Was Fine

In 2023, I built a beautiful spreadsheet for my product launch.

Nine sheets.
Costs. Margins. Pricing ladders. Stock levels. Shipping. Supplier lead times. Forecasts.

It looked professional.
It looked serious.
It looked like I knew what I was doing.

It told me everything was fine.

It was completely useless.

Not because the numbers were wrong.
Because the most important numbers weren’t there at all.

The test I ran

I wanted to see what AI in Excel is actually good for — not the theory, the actual grunt work. The spreadsheets you look at every day.

So I used a sourcing project I am working on as a test case. I took three supplier emails — full of specs, prices, MOQs, lead times, payment terms — and pasted them raw into Claude inside Excel. I explained the context once.

It populated the tracker, pulled the relevant fields, and restructured the whole thing into a supplier comparison view in about twenty minutes.

That would have taken junior-me two days. It’s not magic. It’s just genuinely useful.

Then I got curious about 2023

Here’s where it got interesting.

If AI can extract, compare, and restructure operational data that cleanly — what would it have done with my 2023 product launch spreadsheet?

So I opened it. The one I’d built for my illustrated product launch. Nine sheets. Professionally laid out. The kind of document that makes you feel like you know what you’re doing.

But before I handed it to Claude, I added something that didn’t exist in the original — a small assumptions table at the top. Five numbers I now know matter:

Total website visits at launch: 622

Organic stranger visits (Google + unknown): ~100

Conversion rate: 0.96%

Average order value: £27

Expected orders from strangers: ~1

Then I asked Claude: based on these figures, what was the maximum revenue realistically achievable at launch? And what traffic would I have needed to generate £2,000?

What it said

Maximum revenue from all 622 visits: £161. That’s the ceiling. Not what I made — what I could have made if every single visitor converted at the average rate.

Maximum revenue from organic strangers only — people who had no idea who I was: £26. Essentially one order.

Traffic needed to hit £2,000: 7,716 visits.

That’s 12x my total launch traffic. And 77x my organic stranger traffic.

It also noted: “The core challenge wasn’t the product or the conversion rate — 0.96% is actually reasonable for a new e-commerce site. It was traffic volume.”

Which is not nothing. That’s a useful, specific, data-backed diagnosis. In about thirty seconds.

But here’s the part that matters

Would Claude in Excel have flagged this in 2023, on its own, with the spreadsheet I actually had?

No.

And this is the bit I want to sit with, because I think it’s the most useful thing I can tell you.

I only knew to add that assumptions table because I had spent the past year dissecting this business — because I’d gone back through my website analytics, pulled the traffic data, calculated the actual conversion rate, and faced the numbers properly. I knew what had gone wrong. And I knew, because of that, which data to add to the spreadsheet before asking the question.

If I had handed Claude the 2023 product range spreadsheet with no additions — the version that existed at the time — it would have looked at my pricing ladder and said: “margins look reasonable.” It would have looked at my production units and said: “MOQ is sensible for a first run.” It would have looked at my cost structure and said: “you need X sales to break even.”

What it couldn’t have said: “You only have 18 email subscribers. This is mathematically impossible.”

Because that number wasn’t in the spreadsheet.

The column that was missing

This is what I keep coming back to.

The 2023 spreadsheet had nine separate sheets. It had everything a product launch is supposed to have. And not one of those sheets asked:

How many people actually know you exist?

Audience size. Email subscribers. Organic traffic. The number of strangers who would realistically see this product.

It wasn’t in any operations course I had taken. It wasn’t in any business planning template I used. It wasn’t something my years of luxury fashion operations had taught me, because at Mulberry the audience wasn’t my problem — I was working inside a business that already had one.

Every operations template I have ever used assumed demand already existed.

When I built my own thing, I built the operations without the audience and then launched into a void. And my beautifully constructed spreadsheet gave me no signal that anything was wrong — because it was only tracking the things I’d learned to track.

What this actually means about AI in operational work

AI in Excel is genuinely useful. I’m not walking that back.

The supplier email extraction was real. The comparison formatting was real. The sensitivity analysis — once I had the right data — was fast, accurate, and would have taken me hours to build manually.

But the honest version of “AI will help you catch mistakes” is:

AI will help you catch mistakes in the data you give it, if you ask the right questions, if you know which numbers need to be in the same document.

The tool amplifies your operational knowledge. It also amplifies your blind spots — very efficiently, without complaint, giving you clean outputs that look completely fine right up until they aren’t.

That’s not a criticism of the tool. It’s just an honest description of how it works.

Which means the conversation about “AI in operations” can’t just be about the tool. It has to be about the person using it — what they know to track, what questions they know to ask, and what data they know belongs in the spreadsheet before the AI ever sees it.

The practical takeaway

If you’re building a product business — or planning a launch — here’s the column that needs to be in your product range spreadsheet before you model anything else:

Current audience size (email list, engaged social followers — people who will actually see this)

Expected traffic at launch (not optimistic, not worst case — realistic, based on what you can see right now)

Conversion rate assumption (e-commerce averages 1-3%; new stores with cold audiences should model 0.5-1%)

Expected orders (traffic × conversion rate)

Revenue ceiling (expected orders × average order value)

If your revenue ceiling is below your break-even, stop tweaking the product. You need an audience, not a better margin.

And no amount of refined pricing or better margins or improved cost structure will fix that. The maths just won’t work.

Ask the question before you build the product. AI won’t ask it for you.