Ecommerce and Shopify ·Pillar

Marketing analytics: turn scattered data into decisions

Marketing analytics connects spend, sales and retention through governed metrics and conversational AI. Find where your funnel leaks. Read the guide.

Your marketing analytics report shows strong reach and flat revenue, and nobody can tell you which step broke or what to do about it. The data exists. It is just scattered across six tools that disagree about what a conversion even is, and every one of them shows you what happened without ever saying why.

The fix is not a chat box bolted onto that raw mess. A full-funnel marketing strategy runs on three layers built in order: a data warehouse that unifies every source, a governed semantic layer that pins one definition to every metric, and only on top of those a conversational layer you can simply ask. Get the foundation right and every funnel stage, reach to retention, finally gives a straight answer.

What is marketing analytics, and what does a full-funnel strategy need?

Marketing analytics brings channel costs, on-site behavior and customer outcomes into a shared measurement framework. It helps a team compare performance, investigate where results change, and decide what to do next. A full-funnel strategy needs that view across awareness, consideration and action, rather than a separate verdict from each channel.

Marketing has a split personality. Brand building works the top of the funnel, where demand and attention are created but the impact is hard to measure. Performance marketing works the bottom, where clicks and conversions are easy to count. Because the bottom is the easy one to justify, budgets skew there and the top starves, even though the customers who arrive with an emotional connection to the brand tend to be the more loyal and valuable ones. A full-funnel marketing strategy refuses that trade-off: it unifies every marketing effort under a single, interweaved set of KPIs and measurement, and judges each stage by how it lifts the others rather than in isolation.

You cannot run that strategy, or prove it, without marketing analytics that span the whole funnel, because you cannot optimize a stage, or a hand-off, you cannot see. That is the analytics work this hub is about: measuring every stage and every transition so you can tell where the strategy leaks and act on it. Most marketing dashboards show the stages. Almost none show the transitions, which is exactly where budget leaks. That is understandable, since dashboards are typically built off ad-hoc requests covering a single marketing data source.

full-funnel marketing strategy := combining all efforts, from brand building to performance marketing under one set of inter-linked teams, KPIs and measurement, so each stage is judged by how it lifts the others, not in isolation.

the analytics behind it := a single, unified measurement system across every stage and channel, so you can see how each touchpoint moves the next and rebalance spend and effort toward what drives growth.

The data that would tell you which step broke is scattered across Google Ads, GA4, Search Console, Meta, your email marketing tool and Shopify, in six schemas that disagree about what a “conversion” even is. Maya closes that gap: it unifies the data, governs the definitions, and lets you ask for the answer in plain language. The analysis that used to eat an analyst’s afternoon (if the required infrastructure was even available, otherwise it was impossible) arrives in the time it takes to type a question.

The value pyramid: from data to outcome

There is a pyramid every analytics question has to climb, and where your stack stops on it is the whole story.

{
 "$schema": "https://vega.github.io/schema/vega-lite/v6.json",
 "description": "From data to outcome, a narrowing stack: data (the raw numbers) at the base, then insight (what happened, in context), diagnosis (why it happened), decision (what to do), and outcome (did the move work?) at the top.",
 "height": 320,
 "data": {
  "values": [
   {
    "name": "Outcome",
    "note": "did the move work?",
    "w": 8,
    "y0": 0,
    "y1": 0.5,
    "ym": 0.25,
    "yn": 0.7,
    "x0": -4.0,
    "x1": 4.0
   },
   {
    "name": "Decision",
    "note": "what to do",
    "w": 12,
    "y0": 1,
    "y1": 1.5,
    "ym": 1.25,
    "yn": 1.7,
    "x0": -6.0,
    "x1": 6.0
   },
   {
    "name": "Diagnosis",
    "note": "why it happened",
    "w": 16,
    "y0": 2,
    "y1": 2.5,
    "ym": 2.25,
    "yn": 2.7,
    "x0": -8.0,
    "x1": 8.0
   },
   {
    "name": "Insight",
    "note": "what happened, in context",
    "w": 20,
    "y0": 3,
    "y1": 3.5,
    "ym": 3.25,
    "yn": 3.7,
    "x0": -10.0,
    "x1": 10.0
   },
   {
    "name": "Data",
    "note": "the raw numbers",
    "w": 24,
    "y0": 4,
    "y1": 4.5,
    "ym": 4.25,
    "yn": 4.7,
    "x0": -12.0,
    "x1": 12.0
   }
  ]
 },
 "encoding": {
  "y": {
   "field": "y0",
   "type": "quantitative",
   "scale": {
    "domain": [
     0,
     5
    ],
    "reverse": true,
    "nice": false,
    "zero": false
   },
   "axis": null
  }
 },
 "layer": [
  {
   "mark": "bar",
   "encoding": {
    "x": {
     "type": "quantitative",
     "scale": {
      "domain": [
       -12.0,
       12.0
      ]
     },
     "axis": null,
     "field": "x0"
    },
    "x2": {
     "field": "x1"
    },
    "y2": {
     "field": "y1"
    },
    "color": {
     "field": "w",
     "type": "quantitative",
     "legend": null,
     "scale": {
      "reverse": true
     }
    }
   }
  },
  {
   "mark": {
    "type": "text",
    "style": "maya-label",
    "baseline": "middle"
   },
   "encoding": {
    "x": {
     "type": "quantitative",
     "scale": {
      "domain": [
       -12.0,
       12.0
      ]
     },
     "axis": null,
     "datum": 0
    },
    "y": {
     "field": "ym",
     "type": "quantitative"
    },
    "text": {
     "field": "name"
    }
   }
  },
  {
   "mark": {
    "type": "text",
    "style": "maya-note",
    "baseline": "middle"
   },
   "encoding": {
    "x": {
     "type": "quantitative",
     "scale": {
      "domain": [
       -12.0,
       12.0
      ]
     },
     "axis": null,
     "datum": 0
    },
    "y": {
     "field": "yn",
     "type": "quantitative"
    },
    "text": {
     "field": "note"
    }
   }
  }
 ]
}

Most analytics tools have spent years polishing the bottom two levels. A modern dashboard is very good at data and insight: it shows you what happened, cleanly and in context. That is description, and description is where almost every tool stops. It hands you a falling line and leaves the three levels that actually change a quarter (why it fell, what to do about it, and whether the fix worked) to a human with a spreadsheet and an afternoon to spare.

Maya’s whole purpose is climbing past description. Diagnosis names why the number moved. Decision turns that into what to do, ranked by what moves the most. Outcome closes the loop by checking whether the move worked. The funnel below is the map; the pyramid is the altitude. A report that tells you organic clicks fell is sitting on the data level. An answer that tells you the drop was ranking and not click-through, which queries to defend first, and whether last quarter’s fix held, has climbed to the top.

The metric on top of the pyramid is your business’s to choose

The pyramid climbs to an outcome, but which outcome is the business’s call, and that choice reorders everything beneath it. The metric a company agrees to optimize, its overall evaluation criterion, follows its business model:

  • A store sells things. Its criterion is the sale: conversion rate, order value, return on ad spend, lifetime value. Every opportunity gets ranked by what pays back.
  • A publisher sells attention. An ad-supported site earns on every pageview, because each visit is display inventory shown, so its criterion is traffic and engagement: sessions, pages per visit, click-through, return visits. A reader who opens three articles is the win, bought or not.

Same marketing analytics, a different number on top. Take the paid-to-organic keyword read: from paid search you can estimate what an organic click on each query is worth, then decide where next quarter’s content goes. A store ranks those queries by conversion rate, chasing the searches that pay back; a publisher ranks the same queries by click-through and volume, chasing the searches that fill the page. Maya reports the whole funnel either way; the business decides which number sits on top, and the ranking underneath follows. The awareness and consideration bands still matter for both: the criterion sets the priority, not the whole picture.

The three layers under the simplicity

The pyramid is what Maya climbs; the three layers are how it climbs. Maya is one product with three layers, and each does a job the layer above depends on.

flowchart LR
  DW["1 · Marketing data warehouse<br/>every platform, unified"] --> SM["2 · Semantic model<br/>one governed definition<br/>of every metric"] --> AI["3 · Conversational AI<br/>ask → routed analysis →<br/>validated answer"]
  • The marketing data warehouse lands every integration in your very own cloud database and reconciles it, so recorded Shopify orders can be read beside marketing-platform results without counting each platform’s claim as another store sale.
  • The Power BI semantic model defines each metric once (conversion rate, cost per acquisition, return on ad spend) so your CAC and finance’s CAC finally agree. This is what “governed” means here, and it is worth being precise about. Governed does not mean users are restricted. It means the layer enforces one definition of every metric, so the model cannot invent what CAC, ROAS or conversion stand for. The definitions live in typed measures, not in a prompt, which is why a rephrased question cannot quietly change what a number means.
  • The conversational layer is the part you actually interact with. It is a Model Context Protocol (MCP) server: the open standard that lets an AI assistant (Claude, ChatGPT, a Slack bot) call your data and tools directly. You ask “How is SEO performing?” and the Maya MCP routes the question to a packaged analysis that knows which measures to pull, the benchmarks to judge them against, and the traps to avoid. It is our team’s entire marketing knowledge and experience written down as software. The prompt steers which analysis runs; the semantic layer enforces what the numbers mean. That split is what makes a governed MCP different from a raw chatbot wired to a database: steering can be rephrased, while the semantic layer keeps metric definitions consistent. The assistant still needs to use the returned data correctly; a governed model does not guarantee an error-free answer.

The division of labor is the whole point. Maya does the data assembly, the statistical analysis and the matching of signals against our marketing playbooks. You do the thinking and the creative.

The marketing funnel, stage by stage

The marketing funnel is the company’s structured view of how a stranger becomes a customer. The shape is the point: it is wide at the top, where a large audience first becomes aware of you, and it narrows to the bottom, where the few who are ready actually act. In its textbook form it has just three bands.

{
 "$schema": "https://vega.github.io/schema/vega-lite/v6.json",
 "description": "The marketing funnel narrows in three bands: awareness (they learn you exist), consideration (they weigh you against the alternatives), and action (they decide and convert).",
 "height": 192,
 "data": {
  "values": [
   {
    "name": "Awareness",
    "note": "they learn you exist",
    "w": 20,
    "y0": 0,
    "y1": 0.5,
    "ym": 0.25,
    "yn": 0.7,
    "x0": -10.0,
    "x1": 10.0
   },
   {
    "name": "Consideration",
    "note": "they weigh you against the alternatives",
    "w": 14,
    "y0": 1,
    "y1": 1.5,
    "ym": 1.25,
    "yn": 1.7,
    "x0": -7.0,
    "x1": 7.0
   },
   {
    "name": "Action",
    "note": "they decide and convert",
    "w": 6,
    "y0": 2,
    "y1": 2.5,
    "ym": 2.25,
    "yn": 2.7,
    "x0": -3.0,
    "x1": 3.0
   }
  ]
 },
 "encoding": {
  "y": {
   "field": "y0",
   "type": "quantitative",
   "scale": {
    "domain": [
     0,
     3
    ],
    "reverse": true,
    "nice": false,
    "zero": false
   },
   "axis": null
  }
 },
 "layer": [
  {
   "mark": "bar",
   "encoding": {
    "x": {
     "type": "quantitative",
     "scale": {
      "domain": [
       -10.0,
       10.0
      ]
     },
     "axis": null,
     "field": "x0"
    },
    "x2": {
     "field": "x1"
    },
    "y2": {
     "field": "y1"
    },
    "color": {
     "field": "w",
     "type": "quantitative",
     "legend": null,
     "scale": {
      "reverse": true
     }
    }
   }
  },
  {
   "mark": {
    "type": "text",
    "style": "maya-label",
    "baseline": "middle"
   },
   "encoding": {
    "x": {
     "type": "quantitative",
     "scale": {
      "domain": [
       -10.0,
       10.0
      ]
     },
     "axis": null,
     "datum": 0
    },
    "y": {
     "field": "ym",
     "type": "quantitative"
    },
    "text": {
     "field": "name"
    }
   }
  },
  {
   "mark": {
    "type": "text",
    "style": "maya-note",
    "baseline": "middle"
   },
   "encoding": {
    "x": {
     "type": "quantitative",
     "scale": {
      "domain": [
       -10.0,
       10.0
      ]
     },
     "axis": null,
     "datum": 0
    },
    "y": {
     "field": "yn",
     "type": "quantitative"
    },
    "text": {
     "field": "note"
    }
   }
  }
 ]
}
marketing funnel := the company's model of how a stranger becomes a customer, in three bands: a wide top of awareness, a middle of consideration, and a narrow bottom of action, narrowing as people drop off.

A prospect moving along the funnel interacts with all marketing efforts of a team: SEO, a display ad and an influencer post all work the top, while a branded search ad and a retargeting flow work the bottom. The analytical challenge is to be able to see the whole funnel at once: whether the money going into the top actually shows up at the bottom, which hand-off is leaking, or how a dollar moved from one band changes the next.

The three bands and the hand-off between them

Across every effort, those three bands are where the money goes and where it leaks. The measurable question is never just how a band is doing, it is whether each band hands the customer cleanly to the next. Watch the level of each band and, more importantly, the transition out of it.

Band What good marketing does here What to actually measure Where it shows up across efforts The hand-off out
Awareness (top) get in front of the right people reach and impressions, plus the click-through rate that proves the message landed SEO, display, paid and organic social, PR, influencer did visibility earn a click and a visit?
Consideration (middle) earn attention and trust while they weigh you engaged visits, dwell, pages and content consumed, return visits, email opens and clicks web analytics, content, organic social, email, retargeting did engagement become a conversion?
Action (bottom) convert them, then keep them conversion rate and conversions, order value and ROAS, repeat purchase and lifetime value search ads, retargeting, checkout, lifecycle email, cohorts did the spend pay back, and does value compound?

Maya reports the hand-off, not just the band. A campaign with great reach and a dead landing page reads as “good awareness” on a vanity dashboard; across the funnel it reads as a broken awareness-to-consideration hand-off, where the click rate holds but every visit bounces.

Awareness: are the right people seeing you?

Awareness is a question of reach and its immediate implications: how many of the right people actually saw you, and whether the message was good enough to earn a click. The number that separates a reach problem from a relevance problem is the click-through rate. Plenty of impressions with a low click-through means people saw you and scrolled past, a message problem no extra budget fixes. Every top-of-funnel effort asks those same two questions in its own vocabulary, from an organic ranking to a paid impression to a social view. See SEO, search ads and paid social.

Consideration: does the visit earn engagement?

A click is only a promise. Consideration is the stretch where a prospect decides whether you are worth more of their time, and the signal is engagement: do they stay, read, go deeper, come back, open the next email. A visitor who bounces in two seconds never had a chance to convert, and no amount of extra traffic changes that. The lever is the landing page, the targeting, or the relevance of what they found, not the budget. It also matters whether the visit is a first-timer discovering you or a returning prospect weighing you again, because discovery and loyalty are different jobs. See web analytics and SEO.

Action: do they convert, and do they stay?

Action is the make-or-break band, and it has two halves. The first is the conversion rate: of the people who engaged, how many took the key action, whether that is a lead, a signup or an order. A healthy rate on thin volume says scale the channel; a fat top with a collapsing rate says fix the offer or the page before spending another euro. The second half is whether the action paid off, which comes down to the return on the spend that won the customer and how long that customer takes to repay it. The band does not end at the first sale either: repeat purchase and lifetime value decide whether you have a business or a treadmill, and a 5% lift in retention can raise profit 25-95% (Reichheld, via Harvard Business Review, 2014). See paid spend and customer lifetime value.

What marketing analytics looks like with Maya

Take the most common real question: organic traffic dropped and nobody knows why.

Without Maya: an analyst exports Search Console, joins it to GA4 in a spreadsheet, builds a pivot of clicks and average position by query, and a day later reports that a cluster of page-1 keywords slipped to page 2. With Maya: you ask “Which queries lost the most organic clicks this quarter, and was it ranking or click-through?” Maya pulls the clicks, the click-through rate and the average position by query, separates the position drops from the click-through drops, and returns the ranked list with the diagnosis attached. Seconds.

Read that against the pyramid. The spreadsheet version stops at insight: a clean chart of what happened. Maya’s version names why (ranking, not click-through), says what to defend first, and can check whether last quarter’s fix held. That is diagnosis, decision and outcome, the three levels description never reaches.

The money version of the same question: which channel quietly burned budget this month, Meta or Google?

Without Maya: someone exports Meta Ads and Google Ads into a spreadsheet, reconciles two platforms’ conversion counts that will never agree on their own, rebuilds blended cost per acquisition by hand, and the argument about whose number is right outlives the month it was about. With Maya: you ask the question as written. Maya’s packaged spend reads compare efficiency across channels on the same governed definitions, check pacing against the budget you set, and come back with a reallocation recommendation instead of a chart to interpret. The meeting starts at the decision.

The same reading discipline applies at the retention end of the funnel. Ask “Which month’s new customers came back to buy again?” and read Shopify orders by first-purchase month. Compare cohorts at the same age: an older cohort has had more time to accumulate orders and revenue. Use those recorded patterns to investigate acquisition quality, without treating them as a lifetime-value forecast or proof of acquisition payback.

One honest caveat: not every connected source carries packaged analyses yet. The launch scope distinguishes packaged analyses for Shopify, Google Ads, Meta Ads and Search Console from connected, queryable sources such as GA4, email and organic social. You can still query those numbers in plain language against the governed definitions, but the diagnosis is yours to run.

That is the shift. Your team stops being a report-assembly line and starts being what you hired it to be: people who read the signal and make the move.

The bottom line

Marketing teams do not lack data. They lack one place to see it. The funnel is easy to draw and hard to run, because the money goes out across a dozen efforts that each report on their own terms, and no single marketing analytics view shows the whole thing at once. Maya removes the assembly step (the exports, the joins, the reconciling of six definitions of “conversion”) and hands the funnel back to you as one connected picture you can question in plain language. The measurement stays rigorous, the benchmarks stay honest, and your best people get their hours back for the work no model can do for them: the judgment and the creative. When the whole funnel finally sits in one picture, which hand-off will you look at first?

Glossary

conversion rate := of the people who engaged, the share who took the key action: an order, a lead, a signup.
customer acquisition cost (CAC) := total spend to win a group of customers, divided by the customers it brought in.
cost per acquisition := what one conversion cost: spend divided by conversions.
return on ad spend (ROAS) := revenue attributed to a campaign divided by what the campaign spent.
Model Context Protocol (MCP) := the open standard that lets an AI assistant call your data and tools directly.
click-through rate := clicks divided by impressions: the share of people who saw you and acted.
lifetime value (LTV) := the total margin a customer is expected to generate over the whole relationship.
cohort := a group of customers won in the same period, tracked together as they age.
repeat rate := the share of a cohort's customers who come back and buy again.
average order value (AOV) := revenue divided by orders: what the typical basket is worth.
payback := the point where a cohort's accumulated margin repays what it cost to acquire.

Each discipline below goes deep on its slice of the funnel: its failure modes, what to measure, and the questions to ask. The money-decision pieces lead; start with the one that worries you most.



Ask this about your own account

Connect your sources, then ask Maya the question this piece just answered, on your own numbers.

Start with your store