I Used Claude and Copilot on the Same Excel Model—HUGE Difference

I ran the exact same Excel model through

Claude and Copilot. And one of them

really struggled. Same file, same

business question, same assumptions, but

two completely different behaviors. And

here's the part that matters. If you use

either one of them the wrong way, you

don't actually save time. You just move

mistakes somewhere harder to catch. By

the end of this video, you'll know

exactly which one to use, when to use

it, and how to stop double-checking

everything you ship.

[music]

Here's what I did. I took a realistic,

messy, multi-tab cash flow model that I

was handed, not a clean demo file, not

something I built myself, something that

looks like real life. We've got multiple

sheets, revenue timing logic, expense

acrruel reversals, scenario toggles,

rolling forecast logic feeding a summary

view. Then I ran three very practical

finance tasks through both Claude and

Copilot in Excel. Number one,

understanding a model I was handed,

simply explaining to me someone else's

work. Number two, debugging errors that

I intentionally inserted into the file.

and three updating scenarios to test new

business issues. Three tests, two AIs,

one declared winner. If you haven't

already, make sure to click the link in

the description and join my free weekly

newsletter, Finance Automation Insider.

I send a copy every Wednesday with tips,

tricks, and hacks just like this. And I

don't want you to miss a thing. Plus,

when you join, I'll send you a free copy

of my guide to 15minute finance

automations. So, click the link down the

description to get started. So, for our

first case study, understanding a

multi-tab cash flow that's just handed

to you. Now, we've all been there.

Someone sends you a model and says, "Can

you take this over?" It works

technically, but you don't understand

it. Multiple tabs, linked sheets,

hard-coded timing assumptions buried in

random places, timing logic that might

make sense or might not. But before I

change anything, I want to understand

it. So, I ran this model through Claude

first and then co-pilot. I'm just going

to broadly ask Claude to explain the

entire workbook and how it functions.

So, let's go ahead and do that. Let's

assume that this is a workbook. So, this

is a 13week cash flow forecast. This is

a 13week cash flow forecast for the F9

Finance coffee shop because I'm obsessed

with coffee. So, that's what we're going

to use. And let's just say that an

analyst has handed this to me. I've

never seen it before and I want Claude

to explain it to me. So, we'll ask

Claude, please explain this entire

workbook, how it's connected and how it

works since I have just taken it over.

We'll go ahead and send this away. Okay,

so this is just a model explanation and

we'll see if it gives back. Let's give a

little bit more room here because we're

really interested in seeing what this

says. Uh so let's see. It's going to

read all three sheets. Very quickly

understood kind of what data was

available in here. It's reading those.

It has a complete picture and it's

already coming back explaining it. So

this is a rolling 13 cash flow forecast

for a business with three locations.

Three sheets flow into each other. It's

got data then forecast and stress

testing or scenarios. So data has the

source data. There's a structured table.

This is what's in the table. It's got

actuals and budget. Here's the forecast

model. It's controlled in the top left.

Then there's the cash flow lag table and

the weekly grid. This is really great.

It's explaining kind of where all the

formula references are, kind of what all

the naming conventions are. And then the

stress testing running five different

scenarios. It even explains the

scenarios for us and how those scenarios

are functioning. Wow, this is really

cool. This would be, you know, this

could be 30 minutes to an hour of

research just to understand this. And

then here's how it all connects. And

this is really cool. Key maintenance

actions. So, how to roll it forward, how

to update actuals, adding capex,

tweaking scenarios, and filtering by

location. So, to be totally fair, I'm

going to give exactly the same prompts

for exactly the same file. So, we're

here. We've got a co-pilot window open.

Now, I'm going to use co-pilot agent

mode. I'm going to give Copilot every

possible resource to get this right

because agent mode is most comparable to

Claude. So, I'm going to go down here.

I'm going to click tools. I'm going to

turn on my agent mode. I'm going to let

agent mode go ahead and spool up. All

right, we're in agent mode and I'm going

to give it the exact same prompt. Please

explain this entire workbook, how it's

connected and how it works since I've

just taken it over. We'll send that away

and we'll see what Copilot does with it.

So, so far I noticed that Copilot is

working on it. It says it's on step one

and now it's in step two. So, it's

working through it. It's going to

examine it. But honestly, by this point,

we already had a pretty robust answer

coming back from Claude. So, speedwise,

I think I'm already seeing this lean

towards Claude. Obviously, though, we'll

wait and see how good the input is.

Again, it's still going and we don't

really have anything back. Claude was

already fully explaining the model to us

at this point. Okay, so it reasoned in

three steps. It took about twice as long

as claw to come back, but we do have

this. So, let's see how good it is. So,

complete workbook documentation. We have

the overview. It's a 13we cash flow

forecast. Sheet one is the data. It's a

central data repository. We've got the

structure categories, key features. So

far so good. Sheet two is the forecast

purpose control panel, core formula

engine, some example formulas. Here's

the data flow. We've got a KPI section.

Scenario three, the sheet stress testing

purpose, what scenarios were running,

how scenarios link back to the main

forecast, the scenario summary metrics,

the results, and here's how everything

connects. Wow. Now, I'm sure this is all

accurate, but this is actually pretty

darn hard to follow and understand for a

regular old user. How to use this

workbook. So, here's some good

information on regular updates, how to

run different views, how to stress test,

and some key considerations. So, it did

it. It explained the workbook, but

honestly, this is a clear win for

Claude. So, I'm going to come back to

Claude's response here, and just look

how much cleaner and easier to read this

is. Again, Copilot got to this point,

but this is just dramatically more

userfriendly. It's not speaking in the

weird code. It's not showing just these

massive formulas. It's got tables. How

it connects is a lot cleaner and easier

to read. It's just it's really condensed

and straightforward. And it just says,

"Hey, here's how you need to maintain

it." So, I'm going to give this as a

clear win to Claude. Both of them

accomplished it, but Copilot just wasn't

userfriendly and didn't really leave you

understanding the model, whereas Claude

was much clearer about walking you

through it. If you haven't already, make

sure to subscribe to the channel. I post

brand new videos every week, and I don't

want you to miss a thing. So, now

debugging an error in the model. We're

gonna break something in this model.

We're going to insert two different

periods where there's an error in the

math. It's going to throw off the cash

flow calculation. It's going to be a

subtly broken formula, one that could

have accidentally been copy pasted in,

and not one that devastates the entire

calculation. Now, this is where most

people manually trace formulas for an

hour. You can tell something's wrong,

but it looks so close to everything

else. So, let's start with Claude, see

how it handles it, and then we'll move

on to Copilot. Bugging is a perfect way

to use this. So, let's go to the model

and let's just break something. Let's

see where we can break something. Uh,

I'm going to go to week seven

and let's just pull out some random

things that should be adding in here.

Okay.

And then our ending cash. Let's just

move our net cash flow here. Let's move

our beginning cash over here and just

kind of throw things off. All right. And

we're going to hide our little triangles

here. All right. And then let's just

tell it that we think there's an error

and see if it can help us find it. So

I'm simply going to tell Claude, I think

there's an error on the cash flow model.

Help me find it and correct it. So we'll

send it away and see if it can find

those errors that we just inserted into

the file.

Uh, let's see. The changes suggest

something is off with the formulas. Let

me check the current state of the key

cells.

I found it. The error is in week seven

net cash flow. The formula is missing

two items. It's only summing five items.

So, it caught this one. It did not catch

this one and E yet. I am going to let it

fix the week seven net cash flow

formula. Let's go ahead and allow that.

All right. We'll come back to the

formula that we broke. Beautiful.

So it correctly caught that and fixed

it.

Now it's saying it now correctly returns

the value. Let me verify that everything

else is also corrected.

So here's the error. It's all fixed. And

let's try again. So we put two errors

in. Let's see if it can find the second

one. So let's just say it still looks

like there's an error. Can you find it?

Let's see if it can find our other error

that we inserted in the ending cache.

All right. It's comparing the formula.

Is it going to catch it?

Okay, so it's saying row 25 all looks

correct now, which it is because we

broke row 27.

I noticed some weeks have hard-coded

values and other are zero. I think we

might have tricked it with this one.

Okay, so it says row 25 checks out. Now

it's checking 26 to 27. That is where we

put the other error. So let's see if it

gets it. It's very interesting to see

that this keeps going when it doesn't

initially find the error. Now, it is

important to point out again when I said

that you need to be checking everything.

It didn't know there was the second

error on its first pass. We had to tell

it it looked wrong. So, you have to work

in partnership and not just take it.

But, it is fascinating to see it's still

looking because we identified an error.

It's trying to find it. Let's see if it

finds it. And it found it. There's an

error in week four ending cache. That

was the other thing we broke. So,

because I gave it approval to fix

everything automatically, it's fixing

the formula. It's making sure it all

cascaded downstream correctly.

All right. So, right now it's checking

everything in the model. It's fixed.

This was the error and this is what

happened. There was a copy and paste

error. So, now we've fixed both of the

errors we put in and we didn't really

have to lift a finger aside from that

one time we confirmed that we wanted it

to make changes. But then, now that

we've given permission, it's just making

our changes for us. So, now coming over

to Copilot, I'm going to break the exact

same two things. So, for net cash flow,

I'm going to delete these out.

All right, that's removed. And then I'm

going to come over to my ending cache

for week four and I am going to drag our

beginning cache over to break this

formula. Okay, so the exact same things

we did. I'm going to get rid of our

little green arrow friends. I know that

clearly it should be able to still

understand it's wrong, but I'm just

going to wipe that out to make myself

feel better about it. And then we're

going to go ahead and give this again

the exact same prompt. Okay, I think

there's an error on the cash flow model.

help me find it and correct it. We'll

send that away and we'll see what we get

back. So again, similar to the last one,

this is already again leaning towards

Claude. We don't have a single thing

back from Copilot yet. It is working on

it. It's thinking about it, but Claude

was already starting to talk through its

process. It was showing us how it was

thinking. It was showing us how it was

analyzing the document. And so far,

nothing back from Copilot. It's just

working in the background. doesn't give

you as much confidence. Of course, we'll

still wait and see if how the result is.

The result could be better. Remember,

Claude only caught one error and we had

to poke it. So, this could be a win for

Copilot if it can get both at the same

time. Now, it has started coming back.

It's just on step one now. And again,

it's it's telling us a little bit about

what it's doing, but it's not really

going into the depth that Claude was.

Still a chance for Copilot to pull out

ahead if it can found them at once. Oh,

and now this is interesting. So I said

there was a chance for it to pull it

ahead and it did. It found two errors

and it already fixed the two errors. So

okay, this might have come back for it.

There's corrections made. Fix one cell

H25.

The formula was off and has been

corrected. Let's go ahead and check

that. So this was the formula we broke.

This was the formula we broke right

here.

And it was corrected. Then we also broke

this formula. Let's see if it fixed it.

It did. Interesting. So it it did kind

of hide in the background a little bit

of what was doing, but it caught both

errors without having to be prompted.

That ended up being about the same

runtime and it gave us a nice summary of

the before and after fix impact on the

KPIs. So started off a little bit slow,

but this one is going to be a win for

Copilot. If this is helpful so far, drop

a comment and let me know which side

you're on right now, Claude or Copilot.

I'm genuinely curious how you're using

these and which one you think is going

to win. Now for our third example, the

most complicated one. We're going to

update scenarios to test new issues. We

want to simulate a real world pressure.

Leadership wants to test what happens if

we have to make some capital payments

earlier than expected. Some changing in

payment terms, some, you know, timing

shift. We have those things already

built in. But this is really going to be

about unplanned capital spending move

forward. We need the updated scenarios

fast. And this is all about the ability

to iterate. cloud first, then co-pilot.

Let's see if it can adjust our

scenarios. So, I'm going to go over to

our scenarios tab. And so, we've got

these five scenarios. We have the base

sales down 10%, cost up 10%, AR delay,

and combined stress across a bunch of

them. So, that's all well and good, but

let's toss something else at Claude. All

right, so I'm going to give it a pretty

broad prompt, and I'm just going to say,

"Help me adjust the scenarios model for

an early capital payment of $100,000

coming due." Very broad. Let's see how

Claude handles this. We'll send that

away and we'll let it do its magic. All

right, so it's examining the scenario

sheet to understand the current

structure and how it's set up. It's

saying it understands it and it wants to

ask me a couple of questions before

making changes. I certainly don't mind

that. Okay, so if there's an early

capital payment, when would it hit? It's

actually something else. Um, this would

come into week 10 or 11. So we're saying

one would pull into this period as an

additional

payment. So we'll send that away. We're

saying we want an additional capital

payment on top of what we've already got

to come in. Uh how should this be

modeled? We want to add this to all

scenarios. So that way we can see the

combined stress of all the different

scenarios. So it's an additional capital

payment. It don't have a preference on

approach. Here's the plan. It's going to

add a new parameter column. It's going

to update the 10 formula. It's going to

set the $100,000 value. It's seeing if

column L is empty. It is. So, there's

the potential additional capital. All

right. So, we can actually just watch

it. I mean, we're watching it change our

page as it goes.

It's updating the week 10 column.

It's changing the cash flow formula for

us. Here we go. We can watch it updating

each of the individual scenarios.

Wow, this is cool.

Yep. So there's all five of the

scenarios updated. It's verifying that

the cache is cascading and adjusting.

I like that it goes back and verifies

these new this new agent technology is

just fantastic

because if you remember this used to be

kind of prompt response, prompt

response, but now we're getting all of

the work and it's even checking its work

as it goes. So that's really cool. All

right, so it's now done and updated. It

made two changes. It updated the week 10

formula. It updated the parameter

column. Here's the impact. And you'll

see everything came down by exactly

100,000. And there are two scenarios

that have negative cash that we want to

watch out for. So at some point during

these scenarios with either combined

stress or an AR delay of two weeks or

more, we go into negative cash. But all

of that was done automatically for us.

And then it looks like Claude goes and

kind of condenses the conversation so

they can understand what we've talked

about so far. All right. And again,

we're going to go ahead and pop in here

the exact same prompt. help me adjust

the scenario models for an early capital

payment of $100,000 coming due. We'll

send it away and we'll see who wins. So,

Copilot's moved through steps one and

step two. Now, it's on to step three. It

says it understands how the model works.

It's looking for rows and places that it

can work. It's adding a new scenario

early capex week five that models the

$100,000 payment coming in week five

instead of week 9. Interestingly, Claude

at this point came and asked me how I

wanted the $100,000 treated. Agent

mode's just kind of gone ahead and

decided. So, we'll have to see how

that's handled and how that works, but

it is working through it so far. Uh,

right now we're running about the same

time. We'll see if it extends out or if

it takes any longer or shorter. Okay, so

Copilot says it's added the new

scenario. Now, it just wants to verify

the results and provide a clear summary

of what it did. Now, interestingly, when

we worked with Claude, it would switch

over and it would kind of show us

working live. This all happened in the

background. You notice I'm on the 13week

cash flow tab. Same place I started with

Claude. It didn't flip over to the

scenario. So, we'll have to see kind of

what happened. So, now what it's saying

new scenario added early capex week 9 to

week five. The model's coming due four

weeks earlier. Here's our starting

cache.

All right. So, it's saying there's a

change or no change. Instant comparison

base versus early.

have a cash position comparison. It's

saying it's all okay. And here's some

key insights. Now, let's go look at the

scenarios.

So, where is this extra scenario we're

looking at?

It looks like it just shoved it right

here. So, [snorts] no formatting. It was

actually a little hard to tell that it

was one of our scenarios.

Again, it kind of just decided on its

own it was going to do this. Now, I do

see that it's Now, I do see that it's

moved the $100,000. So, you can see our

negative cash flow in week five where it

said it moved it, but it also did this

as an isolated scenario, whereas Claude

after confirming how we wanted to handle

went into all the other sections. It

also worked in the background. It didn't

really explain its thought process. The

work is fine and technically accurate,

but it's not formatted. It wasn't clear

where the work went. Even though

everything else was kind of cleanly

formatted. So, I'm going to say by and

large, even though it accomplished it,

it didn't clarify. It kind of made more

decisions than it should have. And I'm

intentionally prompting lightly to see

how they behave. Claude saw that the

prompt was light and asked for

clarification. Copilot just went for it.

And the feedback and response is a

little bit sloppy because it treated

this as a totally independent scenario.

But if you notice, it didn't do this

$100,000 adjustment in the other

scenarios. It did this isolated scenario

appear again that's not even clearly a

scenario. So I think this is a clear win

for Claude. So in our test, we had two

clear wins for Claude with our first

scenario explaining the model and our

third scenario running different

scenarios and we had a winner for

Copilot which was our second one where

we adjusted errors in the model. Now, I

think overall, even though Copilot won

the second one, the responses, the

answers, the way that Copilot works is

just not as clear. You can essentially

watch Claude work with your workbook.

You can track what it's doing. It gives

you better context and explanations, and

it's using much more natural language to

explain it. Whereas Copilot's coming

across a little bit robotic, a little

bit more technical, and probably less

userfriendly for the average Excel user.

So overall, even though co Claude

already won two out of the three, I

think it's a clear winner for ease of

use, for speed of response, and also for

just showing its work as it goes because

the ability to kind of track and trace

through this is absolutely huge. And

honestly, I don't know about you, but

I'm really surprised by this result. I

thought Claude for Excel would really

kind of struggle because it's not a

native Microsoft product, but Copilot

is. is literally from the maker of

Excel. So, it's really surprising to see

just how much more natural claw for

Excel was, just how much more dynamic,

how it showed you its work. I was really

surprised to see this, but Claw's the

clear winner. So, if you enjoyed this

video, then you need to check out my

video on getting started with AI agents.

It will take things even further than

this to save you time. I'll catch you

over there. Until then, this is Mike

signing off from F9 Finance. Cheers.

[music]