How My App Is Doing (2 Month Update)

Two months ago, I launched my AI calorie

tracking app. And if I'm being honest,

it has not been the smoothest

experience. The profit margins were

rough because the AI costs were so high.

Most people stopped using the app after

a week. And I just wasn't sure if I

could turn this into something

sustainable. It's been about 2 months

and a lot has changed. In this video,

I'm going to share the real numbers,

revenue, profit, retention. I'll show

you how I dramatically improve my AI

bill, which then increased profit

margins, and all the changes that I made

to start moving the needle on retention.

If you're new here, welcome to the

video. My name is Chris and I build

productivity apps. I usually focus on

one productivity app per video and today

we're focusing on Amy. Quick context.

Amy is a calorie tracker in the style of

Apple Notes. You type the food that you

ate and we will magically calculate the

calories on the right. Probably the most

frictionless calorie tracking app out

there. So, let's talk numbers. We just

crossed $1,500 in monthly recurring

revenue. And that's on 148 paid

subscribers paying around $10 a month.

The last 28 days was $1,968.

For an app that's only 2 months old, I

am very happy with this. The revenue is

only half the story. The other half, the

part that keeps me up at night is the

cost. And if you're following along, you

know that the costs were really high,

but I'm proud to say we actually got it

down. My AI bill for the first month was

about $700, but this month we got it

down to $221. And that's with even more

users using the app. Reason the costs

are so high in the first place is I'm

using a very expensive AI model, Plexity

Sonar, to power the search and calorie

calculation. Every time someone logs a

food or makes an edit, we call this

model, which then cost me about half a

cent. I tried a bunch of different

models, but this is the best one I could

find in terms of speed and accuracy.

It's just kind of expensive. So, this is

the big thing that I've been working on

is getting this cost down. And I made

two major changes to do this. The first

is that I introduced a way cheaper model

for editing. Something I was doing that

wasn't good was I was using Perplexity

Sonar for everything. So, if someone

typed in food, we would call Perplexity

Sonar. It would do a web search. It'd be

really expensive. That makes sense. But

if they edit it, so let's say they write

burrito and they change it to half a

burrito, we call perplexity sonar again.

It does a whole web search and

recalculates it to make an edit. Super

unnecessary because for edits like this,

it's really just basic math. So I made a

modification where now when we make an

edit, we check is it just a portion

change? And if it is, we will send it to

a way cheaper model, which is Gemini 2.5

flash light, and it will go ahead and

recalculate everything for us. Instead

of half a cent, this costs like a tenth

of a cent. And a quick note because I

got a lot of questions after I posted

this online. Why couldn't I just do this

stuff programmatically? Why did I have

to use AI for the recalculation at all?

I did originally try that, but there are

so many edge cases because it's just a

free form text input. Like you can type

anything you want in here. So sometimes

someone would change one burrito to half

a burrito. Very easy to figure out, but

sometimes they change it to a couple

bites of a burrito. So I ended up using

natural language processing anyway to

figure out what was the portion change.

And at that point, I thought, might as

well just use Gemini 2.5 flashlight and

have it figure out the portion and do

the recalculation. Anyway, luckily the

model's super cheap. So, when you look

at the total cost for my bill, Gemini

actually ended up only being about $3 of

the bill. The second change I made,

which I believe had a way bigger impact

on the cost, was I added cashing. If

someone typed in burrito, we would go

run an expensive web search and call

perplexity and that would cost half a

cent. Now, instead, we're doing caching.

So, let's say they have typed burrito in

the past and they type it again. This

time it's going to check Supabase, which

is the database we're using, and we'll

see, okay, we already have the nutrition

info. We're going to pull that, and then

we'll use the cheaper Gemini model to

then just figure out all the portions,

which saves us from this very expensive

perplexity sonar call. I knew this would

make a difference, but I was not aware

of how big a difference that this made.

I think this was the bulk of why the

bill went from $700 to $200. And the

reason is because it turns out people

eat a lot of the same stuff every day.

60 to 70% of my calls are actually

hitting the cash now. I was anticipating

it would save like 20 30% in cost, but

it ended up reducing my bill by almost

70%. Now that I had a little bit more

breathing room on the profit margin

side, it gave me the confidence to work

on something I've been wanting to build,

but I kind of held off on because I was

scared of the costs. And that feature is

menu scanning. And it's a feature I have

not seen any other calorie tracking app

do. Here's why I built it. I noticed

that when I'm at a restaurant, yes, I

have the ability in Amy to scan a

picture of your food and then we'll

calculate the calories that way. But

sometimes I found that the menu actually

had really good descriptions of the dish

and I kind of just wanted to scan the

menu directly instead. But I held off on

building the feature because I didn't

want to make my already not so great

margins even worse. But now that I have

a little bit more breathing room, I

decided let's just go ahead and add it.

Now, when you're at a restaurant,

there's a new option where you can take

a picture of the menu. And we're using

Gemini 2.5 Flashlight to extract all the

items for the menu and kind of

reconstruct it in the app with the

description and even broken down by

section. And you can just tap the

relevant items and we'll add them into

Amy and calculate the calories. I think

this is one of the coolest features I

have added into the app. It's definitely

a magic moment for me. But I think when

people use this a couple times, that's

when they'll go from, okay, this is a

cool feature to I can't use an app that

doesn't have this. And again, because of

the cost, I don't think a lot of

competitors are going to be adding this

feature anytime soon. I am so happy with

how this feature turned out. By the way,

the way that I have been tracking all of

my costs, as you've seen, is through

Post Hog, and a huge shout out to them

for actually sponsoring this video. If

you've been following along, you know

that I've been recommending them way

before they sponsored the channel, and I

will continue to do so. But they are my

favorite tool for app analytics. I've

also been tracking all of my AI costs

with their LLM analytics feature. And

it's amazing because you can see total

cost, you can see the cost per model,

latency, everything is laid out here for

you. You can even see a breakdown here

that Perplexity Sonar is costing me $217

and Gemini is only $3, which is amazing

because now I have a benchmark of what

these models typically cost me in a

month. And if I see something like the

Gemini cost rising, for example, I can

go figure out, okay, is there something

wrong? It's also how I track things like

retention. So, I have a bunch of these

different graphs set up for all of the

important metrics like the download to

sign up rate, how many people are

turning on notifications, a bunch of

other metrics that I care about. I

seriously cannot imagine trying to build

an app and make decisions without Post

Hog. I'll leave a link in the

description if you want to check them

out and if you do sign up, please tell

them that I'm the one that sent you.

Since we're talking about metrics, let's

actually talk about the most important

metric that I've been tracking and that

is week one retention. If you've been

following me for a while, you know that

this is my northstar metric. It's the

percentage of people who after

downloading your app are still using it

one week later. And it's the best signal

that I have found to answer the

question, have I built something truly

valuable? When I first launched, week

one retention was about 3%. That means

that for every 100 people that signed

up, only three people were still using

it 1 week later. After about a month of

work, we did bump that number up to 8%.

Which is good. But the number that I'm

really trying to target is 30 40%.

That's where the really successful apps

tend to live. So a lot of the changes

that I've been making, the features I'm

adding, the bugs that I'm fixing are all

in service to try to move the needle on

that 8% week 1 retention. And I'm proud

to report that in the last month, we did

move the needle on that and went from an

8% week 1 retention to about a 10% week

1 retention. That might not seem like a

big change, but these percentages are

really hard to get. So, let's go over

some of the things that I changed to

move the needle on this. The first is

that there were a ton of small UX

improvements that I made. Now, it sounds

really boring, but it really does add

up. The less friction that an app has,

the more likely people are to stick with

it. I'll give you one example. I was

talking to users to try to figure out

what kind of pain points they were

facing, where the friction was.

Something that kept popping up was that

they love the photo feature where they

can take a picture of the food and it'll

go calculate the calories, but they wish

that they could edit the description

after the photo was taken. They said

that they would use the feature way more

if they had that ability, which kind of

surprised me because I was like, "Wait,

I definitely built it where you can edit

that description afterwards." Like, this

is an input box. When I told them, they

were like, "Oh, I had no idea. Thought

you couldn't really edit that." I'm

pretty sure they weren't alone. So, I

went ahead and added an explicit button

that says, "Is something wrong? Do you

want to make an edit here?" And now they

can type what edit they want to make in

plain English, and we'll update

everything for them. It's a small

change, but it really did reduce the

friction and made the photo feature just

that much better. I probably made about

10 UX changes that were similar to that

to try to improve the UX of the app.

Now, the second thing I added, which I

think is going to make a pretty big

difference to retention in the long run,

it's that I added proper weight tracking

with progress photos. This was hugely

requested. People wanted to be able to

track their weight over time, but they

also wanted to attach a progress photo

so they can see the changes physically

in front of them. Here's the strategic

reason of why I think this will make a

big difference and I took the time to

implement it. When people start storing

data in your app, especially data like

this, kind of like a journal, their

switching cost of changing to a

different app goes up dramatically. They

really don't want to lose all that

history when they move to a competing

app. It does really help improve

retention. The third thing that I added

was a better way to export data and also

better syncing with Apple Health. So,

we'll send even more data to Apple

Health. We actually have this Apple

Health integration where we can pull in

your steps, your workout data into the

app and display it. But Amy also has the

ability to send information to Apple

Health. So we send in the calories

consumed data to Apple Health so other

apps can use it. At the request of some

users, I made some improvements. So we

also send in how much protein you

consumed, how much sugar you consumed.

But changes like this and that ability

to export your data. This was made to

really build trust with users that

they're the ones that own their data

here. If users trust that they can

export and get their data out of the app

really easily, they'll be way more

willing to input data and invest the

time into building their data footprint

into your app, which again increases

retention. Now, the fourth thing I added

on the retention side was smarter

notifications. This one was actually

driving me crazy as a user of my own

app. Amy has notifications where we'll

send you reminders throughout the day

reminding you to log your food. But

something I hated was they did not take

into account if you already logged food

for that day. So let's say you log

dinner, then 30 minutes later you get a

notification saying, "Oh, reminder to

log dinner." So now it does take into

account where if you logged food for

breakfast or dinner, we're not going to

send you a notification for that. And

this matters a lot because notifications

are a huge, huge, huge lever when it

comes to improving retention. If you

bombard users and annoy them with

notifications and they turn that off,

that could have a dramatic effect on

retention. In the future, I actually

really want to make the notifications

even smarter where we'll take into

account your eating pattern so that we

can send you the notifications at the

right time. I was actually hoping to do

it this week, but then I realized I made

a huge mistake and I have not been

storing timestamps properly this entire

time. I did store the date, so I know

exactly what day someone ate their lunch

at, but I didn't store the time. So, I

have no idea when that actually

happened. So, I can't even build these

smart notifications. So, I fixed that

and now we are collecting way better

data. So, I'll give it some time and

then hopefully in the future we can

start building these smarter

notifications. But after all these

changes, the week 1 retention went from

8% to about 10%. 10.2% to be exact. I

know this doesn't sound huge, but

percentage points at this stage are very

hard to move. So, I'm actually really

happy with this number. And more

importantly, the revenue is going up,

the cost is going down, and I have a

pretty clear road map of what I need to

build next to move these numbers in the

right direction. The profit margin on

the AI stuff went from 50% to about 85%.

And when you factor everything else like

hosting costs and all the other stuff,

we're at a very sustainable and healthy

profit margin. Now, that's where we're

at 2 months in. Is it a massive success?

Not really, but it is working. I'm

learning stuff every single week and I'm

having way too much fun doing all of

this. If you like this kind of content,

check out my Instagram and Tik Tok. I

post every single day about building

productivity apps. And obviously, if you

like this content, don't forget to

subscribe. But thank you guys so much

for watching and I will see you guys in

the next video.