All Drafts

I Gave My Customers an AI Personal Trainer, and They Sign In With the Same Account They Shop With

A training app inside the online store: ready-made or AI-built routines, every set logged on the phone, and an AI coach that knows the customer's history. No second account, no second password.

Daniel Flores Osses
Daniel Flores Osses
Nov 18, 2026 · 7 min read

A supplement store has a calendar problem. A customer buys a protein and doesn't come back until it runs out, four to six weeks later. But they train three, four or five times a week. The store shows up in their life once a month; the gym, almost every day.

In the Chilean eCommerce business I built this for (a retail store and a wholesale one, with two warehouses of their own), the question was simple: what if the store came along for the training, not just the purchase? The answer was an AI personal trainer that lives inside the same online store and that any customer can use for free with their account.


One account, one password

The decision that mattered most was not building another app. Nothing to download from an app store, nothing to sign up for twice. The coach is a web app that runs inside the online store, on the same domain, and recognizes the session the customer already has open.

1. You sign in to the storeThe password is typed only there••••2. You open the coachSame session, no second passwordSame account3. Every set lands in your accountFirst on the phone; seconds later, in the accountSaving to your account…All savedAnother phone, same account: the same history
There is no second account and no second password: the coach lives inside the store and uses its session.
  • The password is typed only in the store. The app never asks for it or sees it; if there's no session, it sends the customer to the store's sign-in page and brings them straight back to the coach.
  • Creating an account is free, and it's the same account they shop with.
  • The history lives in the account, not on the phone. Every change is saved on the device first and, a few seconds later, to the account. If the connection drops, the data stays on the phone and syncs when it's back.

That last point sounds like a technical detail, and it isn't. Signal in a gym is bad. An app that loses the set you just logged never gets opened again.


Ready-made routines, or built with AI

People who don't want to think just pick one. The app comes with more than twenty ready-made routines (upper/lower, push/pull/legs, full body, a 30-minute express session, among others) and a catalog of more than a thousand exercises, each one animated with step-by-step instructions.

People who want something tailored have three AI paths:

  • Build a routine from scratch based on their profile: goal, experience, available days, current and target weight, plus a free note ("more focus on glutes").
  • Paste a routine they already had written in another app, a PDF or a message.
  • Ask for a change to an existing routine: the AI applies only that change and leaves the rest as it was.

The second path needed the most care. Many people arrive with the routine their gym trainer gave them, and the last thing they want is an AI "improving" it.

Pasted textMonday legssquat 4x8leg press 3x12lunges 3x10 60sfrom another app or a PDFParsed routineBarbell squat4 × 8 · rest 90 sLeg press3 × 12 · rest 90 sLunges3 × 10 · rest 60 sMissing data: the standard is used (90 s)Same orderAdds or drops nothingYou review and save
The AI transcribes what the customer already had; it doesn't design a new routine or “improve” it.

The rules are strict: the order is kept, sets, reps and rest are respected, and no exercise is added or dropped. If the text is missing a detail, a standard is used (3 sets of 10 to 12 reps and 90 seconds of rest) instead of making up something specific. Names are matched against the catalog so every exercise gets its animation, and nothing is saved until the customer reviews the result.


Four questions and a week already planned

The first time, the coach asks four quick questions: what the customer is after (building muscle, getting stronger, losing fat, or health and energy), how long they've been training, how many days a week they can go, and their height and weight. With that it plans the week and puts it on the today screen.

The weekly plan assigns a routine to each day. If the customer misses a day, they move it to another one without breaking the rest, and days without a routine show up as what they are: rest. "Recovering is training too" isn't a decorative line; it's the rule everything else was designed around. An app that punishes people for missing a day is an app that gets abandoned.


The session, set by set

On the gym floor, the app has to be faster than a notebook. Every set starts with the weight the customer used last time on that same exercise, so most of the time all it takes is a tap.

Barbell bench pressLast time: 80 kg × 8WWarm-up40 kg × 101Set 180 kg × 8 · RPE 8Rest1:302Set 282.5 kg × 8 · RPE 8RecordEstimated 1RM (Epley formula)weight × (1 + reps ÷ 30)82.5 × (1 + 8 ÷ 30)≈ 105 kg
Every set starts from the previous session's weight; records and 1RM are calculated automatically.
  • Warm-up sets kept apart from working sets, so they don't inflate the volume.
  • Weight, reps and RPE (rate of perceived exertion, from 1 to 10) on every set.
  • A rest timer that starts on its own and plays a sound when it's done.
  • Supersets, and the screen stays on for the whole session.
  • Records are detected automatically and celebrated when the session ends.

The 1RM (the most weight you could lift for a single rep) is estimated with the Epley formula, so nobody has to risk an all-out set. With that, the progress section shows weekly volume, 1RM per exercise, body weight over time and a yearly consistency map.


A coach that knows the history

The most visible AI piece is the chat. Customers ask what they'd ask a trainer on the floor, and the AI coach answers with their data in front of it.

What the coach sees before answeringProfile and goalRecords (1RM)Recent sessionsActive routinesExercise catalogLast 10 messagesHow do I add weight on the bench press?Last time you did 80 kg × 8 at RPE 8.Try 82.5 kg on your first setand keep all 8 reps.Uses your real numbers. Never promises discounts.
The answer comes from the customer's history, not from a generic template.

Before every answer, the assistant receives the profile and goal, the estimated records, the latest sessions, the active routines, the exercise catalog with its exact names and the latest messages in the conversation. That's why it can review a session and suggest how much to add to the load, or a deload week if it sees a plateau, using the customer's real kilos rather than a generic chart.

I applied the same principle as with the store's assistant: work with real data and know what not to say.

  • It recommends exercises from the catalog by their exact name, so the customer can find them.
  • It keeps answers short and casual, and never guilt-trips anyone for skipping a day: rest is part of the plan too.
  • It doesn't promise discounts or mention features that aren't available yet.
  • The AI model's key lives on the server, never in the browser, and there's a query limit so AI spending doesn't spiral.

The app also has a supplements section that calculates protein (1.8 to 2.2 g per kilo) and water (35 to 45 ml per kilo) from the customer's weight, with links to the store's categories. It's there, but it isn't the point. The point is that the customer trains.


What the business gains

Frequency. The store goes from being opened once a month to buy something to being opened at every training session, with the same account and the same brand. Every visit to the coach is a store visit that didn't exist before.

And first-party data. What goal each customer trains for, how many days a week, whether they're a beginner or advanced: information no ad platform hands the store, and that customers share because it's useful to them. It all stays in their account, inside the store, without depending on a third-party app. It's the same principle as the points program: the customer relationship is built at home.

The purchase happens once a month. The training, several times a week.

If you want your store to stay with your customers beyond the purchase, start with an AI diagnosis for your company.

Contact me