I Had Nine Apps Full of Metrics and No Answers. Here's How I Brought Them Into One Dashboard

A metrics hub that brings together, in one place, the numbers of a Chilean eCommerce: new customers, repeat purchases, points, every app in the business, what people look at but don't buy, and live campaigns. It only reads, and it's built to anticipate.

Daniel Flores Osses

Oct 20, 2026 · 7 min read

Monday, nine in the morning. The owner of a Chilean eCommerce store wants to know something simple: how last week went. They open the CRM, the mailing tool, inventory, imports and Google Analytics. Five screens, five ways of counting, five different date ranges. Half an hour later they have plenty of numbers and no answer. And the question that really matters, who is going to buy again, isn't answered by any of them.

That business has two stores (retail and wholesale) and I had already built nine internal apps for it, each with its own metrics. What was missing was the place where they'd come together. That's how the metrics hub.

was born. A mirror that only reads

The idea is to look at today's data to anticipate what's coming: who will buy again, how much stock will be left over, which shipment is about to arrive. And the first design rule was that it only reads. It doesn't modify orders, customers, stock or campaigns. If a number looks off, you fix it in the source app. It's a mirror, not another source of truth that can get messy.

Each app sends its metrics to the hub, which builds a card per app and per store on the front page.

It has three customer reports (new customers, repeat purchases and the points program), one tab for each app in the business, an Overview that pulls together the essentials, and two tabs that are among the most used today: Visits and Cyber.

One period for everything

What changed the day-to-day the most was something small: you pick the date range once and it applies to all the tabs. You can jump from the CRM to the mailing tool to inventory, comparing exactly the same days.

A single date picker for all tabs; above 62 days the charts switch from days to months.

Above 62 days the data is grouped by month; below that, by day.
Store selector where the app works with more than one.
At the bottom of each tab, Sources and method explains where each figure comes from.
Permissions per report and per app: the warehouse can see inventory without seeing sales by channel.

The three customer reports

New customers

Cross-references store orders with Google Analytics 4 and regenerates itself every night.

Where they come from: new customers by channel (Google, Meta, direct, even AI assistants) and how many buy on their first visit.
Who they are: age, device, city and interests.
When they buy: heat map by day and hour.
Which channel brings better customers: entry ticket versus repeat purchases.

The channel that brings the most people isn't always the one that brings customers who return.

That last section is the one that has moved the most investment decisions: a cheap channel with no repeat purchases can be worth less than an expensive one with customers who come back.

Repeat purchases

Answers how often customers come back, what percentage returns and after how many months. It uses orders from the store and the ERP.

The cumulative second-purchase curve, and how the time between purchases shortens with each purchase.

Time between purchases and how it shortens with each purchase (the first repeat purchase is always the slowest).
Month-by-month second-purchase curve and ready-to-use phrases: "by month X, this percentage had already come back."
Cohorts: groups of customers who bought for the first time in the same month, followed over time.
Excel lists of recurring and loyal customers.

Points program

A percentage of each order accumulates as a balance for future purchases. The report shows how much is issued, how much is redeemed, how much balance is still live, how long it takes to use it, and whether they buy again after redeeming. A health section detects months when accrual was switched off.

One tab for each app

Each app answers a specific question, with the key numbers on top:

Store CRM: how much the AI assistant answers, how much it escalates to people and how much it sells.
Wholesale CRM: lead funnel, which source works and a ranking of salespeople.
Pricing: a snapshot of the latest sweep from the price comparison tool.
Mailing: opens, clicks, unsubscribes and sales attributed per campaign.
Inventory and purchasing: margin with reliable costs, overstock and the stockouts that cost the most sales.
Turnover and expiry: the highest-value batches at risk.
Imports: shipments, upcoming payments and arrivals.
Affiliates: sales by code from the ambassador program.
Single product record: catalog progress and health paperwork.

Visits: what nobody sees and what gets seen but not bought

This tab was born from an uncomfortable question: how many products with stock does nobody look at? And how many get lots of visits, but almost nobody buys them? The system cross-references visits to each product page with orders and puts a diagnosis on every product.

Each product falls into a quadrant based on how many people see it and how well it converts relative to its category.

Nobody sees it: so few visits that even at normal conversion it wouldn't sell three orders.
Seen but not bought: sells less than half of what's normal for its category.
Out of stock with demand: no stock, but people are still looking for it.

On top of that, a language model reviews each product's figures and assigns it an action: Star, Hidden gem (needs a showcase), Review price, Review page, Push or clear out or Restock.

How a product's sales at stake are calculated and what happens when someone acts on the suggestion.

What makes it actionable is the sales at stake: the orders a product is missing to sell what's normal, multiplied by its ticket. The list is sorted by money. Each suggestion is marked as done or skipped, and a follow-up shows the before and after. It also shows what people search for most in the store and can't buy.

Cyber: the live campaign

During a Cyber event nobody has time to open five screens. This tab is designed to be left on full screen in the office.

Cumulative sales are compared with the previous campaign at the same point, and the close is projected.

Live orders, with district and channel.
Sales compared against previous campaigns at the same hour of the event.
Live margin and profit using the real cost from the ERP, and closing projection.
Shipping by warehouse, people on the site and live conversion.
New, recurring and reactivated customers, with the profit of each group.
Abandoned carts and failed payments.

At close, a final snapshot is saved: a campaign can be reviewed months later exactly as it looked.

What the business gains

One conversation about the numbers, with the same dates and the method written down.
Marketing decided with repeat purchases, not just clicks.
A work list for the catalog, sorted by money.
Campaigns that get corrected live, not the following Monday.

Nine apps knew what had happened. The metrics hub dares to say what's coming.

Back to Drafts Back to Home

Want something like this in your company? Check out the AI consulting.

Contact me