Metrics & analytics glossary

Average Order Value is simply the average shopping cart size. If 10 shoppers visit your bakery and you ring up $200 in total sales, your AOV is $20 per purchase.

Blended CAC is the "real-world" cost of acquiring a customer. If you open your bank account and see you spent $10,000 on marketing in May, and your CRM shows 200 brand-new customers placed an order in May, your Blended CAC is $50 per customer — simple, honest, and impossible to fake with ad platform tricks.

Breakeven ROAS is your "don't lose money" threshold. If you sell a jacket for $100 and it costs you $50 to make and ship, your profit margin is 50%. You must make at least $2 in sales for every $1 you spend on ads (a 2x ROAS) just to break even on that jacket.

Think of CAC Payback as the time it takes to break even on an individual customer. If you spent $120 to acquire a subscriber and they generate $20 in profit each month, it takes 6 months before that customer stops being an expense and starts generating net profit.

Channel CAC looks at one ad network in isolation. For example, if you spent $2,000 on LinkedIn Ads and LinkedIn says it brought you 20 new customers, your LinkedIn Channel CAC is $100.

The customer says "yes, send it, I will pay when it arrives." Nothing is charged at checkout. Until the parcel is handed over and the cash collected, you have a promise rather than a sale — and promises get refused.

Group January's new customers and watch what they paid you over time; do the same for February. The month each group's running total covers what you paid to acquire them is the payback — separate stories for acquisition quality (starting depth) and retention quality (slope).

When an item sells, money immediately leaves your pocket for the product cost (COGS), shipping postage, packing boxes, and credit card swipe fees (2.9%). What is left over is your Contribution Margin — the exact money available to pay for advertising and business overhead.

CPA tells you the price tag of a specific user action. If you spent $500 on an ad campaign and got 25 webinar registrations, your CPA for a webinar lead was $20.

CPC is what the toll booth charges every time someone clicks your link and visits your website from an ad.

CPM is the industry standard price for eyeballs. If an ad network charges a $20 CPM, it means you pay $20 for every 1,000 times your ad appears on someone’s screen.

CTR measures the magnetic pull of your headline and creative. If 1,000 people see your ad and 30 people click it, your CTR is 3%.

A 20%-off sale does not cost 20% of profit — fixed-per-box costs mean it typically removes 30–40% of contribution per order, borrows next month's demand, and teaches customers to wait. Adjusted ROAS counts all three costs.

Reach is how many different people saw your billboard. Frequency is how many times each of them drove past it. If 10,000 people saw your ad and it was shown 40,000 times, each person saw it about 4 times on average.

Of every 100 customers who searched your keyword, impression share says in how many of those auctions your ad appeared. The two loss columns are the diagnosis: "lost to budget" means you could not afford your own demand; "lost to rank" means competitors beat your bid or your quality.

LTV:CAC answers: "Are my customers worth significantly more over their lifetime than what it cost me to bring them through the door?" A ratio of 3:1 is standard — meaning for every $1 spent to get a customer, you make $3 in gross value over time.

Also called "Blended ROAS", MER is the high-level financial health check of your marketing. Instead of arguing over whether Google or Facebook deserves credit for a sale, MER takes all the money your company made in a month and divides it by all the money you spent on all ads combined.

Think of ROAS like a money multiplier at an arcade: if you put $1 into an ad machine and it gives you $4 in customer orders back, your ROAS is 4x. It tells you how much revenue an ad generated compared to what you paid to run it.

You send a parcel, the courier cannot hand it over — nobody home, the customer refuses it, or the phone number is wrong — and the parcel travels all the way back to your warehouse. You paid to ship it out, you pay to ship it back, and you earned nothing.

Mobile measurement moved from a census (count every voter) to an exit poll (survey a sample, estimate the rest). The poll predicts well but you would never recount ballots with it — likewise, SKAN and Sandbox signals are directional, never row-level truth.

Think of an attribution window like an expiration date on a store coupon. If a customer clicks your Google ad on Monday and finally buys your product on Friday (4 days later), a 7-day attribution window says: "Yes, that ad caused the sale, give Google credit!" But a 1-day window says: "Too much time has passed, treat this sale as direct/unattributed."

Think of a click ID as a receipt stapled to a visitor at the door. If that receipt survives to checkout and gets written onto the order, you can later ask the platform exactly which click, campaign and creative started that order. If it gets lost along the way, the order still happens — but nobody can prove the ad caused it.

People rarely buy expensive items on their first click. If someone considers a mattress for 14 days before buying, the ad click happened on Day 1, but the conversion arrives on Day 14. If you check your ad dashboard on Day 2, the campaign looks like a failure because the sales are still in the pipeline.

The billboard you noticed on Monday becomes invisible by Friday — not because fewer people drive past, but because every driver has stopped seeing it. New creative restores attention; more budget behind the tired creative just buys more ignoring.

Imagine handing out coupons outside a pizza shop to people already walking in the door. They use the coupon, and your report claims 100% of those sales were "caused" by your coupon! Incrementality testing tests whether those customers would have bought anyway, proving true ad value.

An exit poll does not ask everyone how they voted — it asks some people and estimates the rest. When a browser blocks the tracking that would have linked a click to a purchase, the platform does the same thing: it looks at the journeys it can still see and fills in the ones it cannot.

The pixel sees the handshake; your CRM sees the wedding. Offline imports mail the wedding photos back to the platform so its bidding learns to find marriages, not handshakes.

You bid on "running shoes"; the customer typed "running shoes for flat feet under 3000". The keyword is your net; the search term is what swam into it. Reading the catch tells you which holes to close (negatives) and which fish deserve their own net (new exact keywords).

UTM tags are like luggage tags attached to web links. When someone clicks `yoursite.com?utm_source=instagram&utm_campaign=summer_sale`, your website's analytics reads the tag and knows the visitor arrived from an Instagram summer promotion.

Imagine driving past a billboard for a coffee shop. You didn't pull over immediately, but the next morning you stop by and buy a latte. View-through attribution credits the billboard for your purchase even though you never "clicked" anything.

Imagine asking a librarian: "How many customers bought red shoes?" If the librarian opens a 1,000-page book and reads every single page from cover to cover just to find three sentences, the librarian had to "scan" all 1,000 pages. In modern databases, you pay for how many pages the database had to read, not how short the final answer was.

Think of compute slots like checkout lanes at a busy supermarket. When you run a query, the database opens multiple lanes (slots) in parallel to scan and process your data faster. The more complex your calculation and joins, the more slot workers the database assigns to finish the job.

Data freshness is how "up to the minute" your reports are. If a customer places an order at 2:00 PM and your sync pipeline runs once every night at midnight, your data has a 10-hour freshness lag during the afternoon.

The same parcel scanned twice at the depot is still one parcel. Deduping is the scan that notices: one row per order_id survives, the photocopies are quarantined, and every revenue report reads the deduped view instead of the raw table.

Instead of "alert if CAC crosses ₹400" (wrong within a quarter as the business grows), the baseline asks "is today weird for a Tuesday, given the last four Tuesdays and the fact that it is Diwali week?". Only genuine surprises page anyone.

A smoke alarm for your data pipes. One query asks "is there anything newer than Tuesday?", another asks "is today half the usual size?", a third asks "does the warehouse agree with the ad platform?". Any "no" pages a human before the business review debates a phantom trend.

An exam paper for your AI with the answer key attached: 20–50 real business questions whose correct numbers a senior analyst has signed off. New model version? Sit the exam. Score up, ship it; score down, revert — no vibe-checking.

Do not hand the AI the keys to the whole warehouse and ask it nicely to behave. Give its login access to nine clean views and nothing else: fewer tables to leak, fewer wrong joins to make, cheaper queries as a side effect. Privacy and accuracy turn out to be the same project.

Imagine a filing cabinet with 365 labeled drawers, one for each day of the year. If you want yesterday's receipts, a partitioned table lets you open only yesterday's drawer, ignoring the other 364 drawers completely. This makes your queries lightning-fast and dramatically cheaper.

Your Meta account ends Tuesday 13.5 hours after your IST business does, and your US account spends dollars while India spends rupees. Without one agreed Tuesday and one agreed rupee, every daily report compares mismatched days and mixed money.

Think of schema mapping as a bilingual dictionary between human business terms and cryptic database column names. To a human, "revenue" is simple. But in a database, is revenue in `orders.total_price`, `invoices.subtotal`, or `transactions.gross_amt`? Schema mapping teaches query tools and AI models exactly what each column means and how tables connect.

It is the company dictionary for numbers. Instead of every analyst, dashboard and AI tool inventing its own definition of "active customer", there is one written-down definition, stored with the code, that everything looks up.

Text-to-SQL lets you ask your database questions in plain English, like "What was our blended ROAS last week by country?", and the AI automatically writes the necessary SQL code to fetch the answer from your CRM or warehouse.