Two projects this quarter · Tuesday & Thursday evenings

You’re about to spend real money on numbers nobody can defend.

Book fifteen minutes and I'll tell you the question you should be asking — and whether your data can answer it.

  • You describe the problem in your own words. No brief, no deck, no form to fill in.
  • I tell you what I'd do, what it would cost, and how long it takes.
  • If there's no decision underneath it, I say so — and you've spent nothing.

Gergana Tyaneva. Twelve years of product and marketing analytics — Rakuten Viber, StepStone, C&A, AKQA. I write the SQL and I present to the room, which is why the analysis survives the room.

Book me

Tuesday and Thursday evenings only. Two projects a quarter — the open slots are what's genuinely left.

Flagship

An AI your team can ask anything. That can't hand them a confident wrong number.

Today's data agents have senior execution and junior judgment. Great SQL, guessed definition. The guardrails are the product.

Point it here

  • Nobody's deciding — let people answer their own curiosity
  • Metric and decision already settled — it builds, a human signs off
  • Two defensible definitions — compute both, force the argument

Not here

  • The metric measures nothing anyone acts on
  • People will act and the definition is disputed
  • It would publish numbers unreviewed — that cascades

from €12,000 · 4–5 weeks · €1,500 assessment, credited

Talk it through →

How this works

Five steps. Stop after any of them.

  1. 1

    You send one sentence.

    Or book fifteen minutes. No brief, no RFP.

    Day 0
  2. 2

    We find the decision — and who signs.

    What changes depending on the answer, and can you approve the spend yourself? We sketch your metric tree. If there's no decision under it, I say so and you've spent nothing.

    Day 1–2
  3. 3

    Scope, price and the map — in writing.

    Fixed. If the job is bigger than the package, you hear it now.

    Day 2–3
  4. 4

    I build it.

    You get me, not a junior.

    2 days – 5 weeks
  5. 5

    Three lines you can forward.

    The number, what it costs you, what I'd do. Plus the keys.

    Delivery

The Grandmother Test

Sounds like your week?

If I can't explain it so your grandmother follows it, I haven't understood it yet. Answers in plain words, with the working underneath — twelve of them, answered.

“GA4 and Meta disagree by 40% and I don't know which to believe.”

Trust — read the answer →

“I can't tell which channel actually pays.”

Attribution

“Our numbers live in eight tools and no two agree.”

Plumbing

“Month-end reporting eats three days. Nobody reads it.”

Time

“We know people churn. Not who, not when, not what it's worth.”

Money

“Someone asked me for a dashboard and I can't work out what for.”

The real one

Prices

What it costs.

Fixed fees, never hourly. An hourly rate pays me for being slow; a fixed fee pays me for being right. Half up front, one revision round.

My rates are high and I don't pad hours. Fixed price means my speed is my problem, not your invoice — you're paying for knowing where to look, not for the time it takes me to look.

Two ways in

€400 / €900 · 3–5 days

I don't sell answers to questions — a question's cost is set by your data, not by the question. I sell bounded work with an output that always ships. Both fees credit in full against anything bigger booked within 30 days.

First · €400 · 3 days

Data Readiness Check

You send one export or read access to one source, and the question you want answered. You get back: shape, null rates by field, ID integrity, duplicate risk, date coverage and gaps — plus two lists. What this data can answer, and what it can't, with what each missing one would cost to fix.

This one can't fail. If the answer is "your data can't tell you that", that is the deliverable, and it's usually worth more than the number you wanted.

Then · €900 · 5 days

One Answer

One question, in writing, in one of five standard shapes. You get the number, the method, the caveats, and three lines you can forward: what I found, what it costs or earns you, what I'd do.

Retention curve · Cohort value & LTV · Funnel drop-off · Segment comparison · Channel contribution

Fixed at this price because the analysis is one of five known shapes — your data is the variable, the method isn't. Anything outside them is scoped as a project, not sold as a unit. One revision; a new cut is a new unit.

Also fixed scope, and nothing depends on your data being usable:

Definition sheet €250 Dashboard teardown, on video €200 Story rewrite €200

Payment on delivery. No charge if it isn't actionable.

Foundation

The Measurement Spec

€900

One week

Up to six metrics — you name them before we start — written down so the argument stops.

  • Numerator, denominator, window, in, out, why
  • Two people compute each one and agree — before you get it
  • The numbers you report today that don't survive

When nobody agrees what “active user” means.

Most booked

Tracking Rescue

€3,500

2–3 weeks

Your tracking is lying and it's moving your ad budget. Numbers never match perfectly — the job is knowing the gap, and when it moves.

  • Day 3: the readiness report — what's broken, ranked by what it costs you
  • Up to 3 platforms audited (GA4, GTM, Meta CAPI, Google Ads, consent)
  • Up to 10 prioritised fixes shipped, before/after proven
  • Ground truth established from billing first
  • A QA sheet you re-run monthly

When GA4 and your ad platforms disagree.

The money one

Where the Money Comes From

€4,500

3 weeks

Who's worth keeping, which channel earns its budget, where the leak is.

  • Day 3: the readiness verdict — what your data can and can't answer
  • Retention and LTV across up to 5 acquisition sources — never pooled
  • Up to 3 segment cuts, one agreed date range
  • What the platforms overclaim, against real orders
  • Churn: who, when, worth what in margin
  • Three recommendations, priced

When you're about to set next quarter's budget.

Quiet win

The Report Nobody Rebuilds

€3,500

2 weeks

The monthly pack, built once, updating itself.

  • Day 3: the reconciliation report — which sources disagree, and by how much
  • Up to 4 sources connected and reconciled so totals agree
  • One dashboard, up to 12 tiles, each carrying its definition
  • Scheduled to named people
  • Handover doc

When reporting eats days you don't have.

Or keep me on the bench.

A standing half-day a week — about 17 hours a month. Questions answered, numbers defended, and someone telling you when a request isn't worth doing. Monthly, cancel with two weeks' notice.

€3,200 / month

Nothing gets built until we've both signed one page.

On anything over €3,000 the first deliverable is a KPI & dimension map: your metrics down the side, your cuts across the top, a tick in every cell that has to work. Usually twenty or thirty cells, of which we tick maybe half.

Done means every ticked cell computes and reconciles. Not "the project feels finished" — that sentence has started every argument I've ever had about scope. Cells we didn't tick aren't refused, they're the next unit, quoted the same way.

Excl. VAT. Half up front, one revision round. Every package above lists its limits as numbers on purpose. Flagship AI Analytics Setup from €12,000.

Or write

Rather not get on a call?

One paragraph is plenty. I read every one myself and reply within a working day.

I read every one myself. Reply within a working day. What you write is used only to answer you — see privacy.

What happens next

  1. You send a sentence or book the slot.
  2. You get a scope and a fixed price, in writing.
  3. You say yes or no. Nobody chases you.

Or email gery.tyaneva@gmail.com

Fit

Who this works for.

Yes

  • You feel the problem and you can approve the spend — same person, same chair
  • Someone can grant access to the data within a week
  • There's a decision waiting on the answer, with a date on it
  • You'd rather hear "your data can't answer that" early than late

No

  • The budget lives with someone who won't be on the call
  • It needs three departments to agree what the metric means first — that's the Spec, not a build
  • You want someone in your tools full-time; hire, don't contract
  • The honest answer would be unwelcome

Straight talk

Things I'll talk you out of.

A dashboard nobody decided to use

Name who opens it on a Monday and what they change. Otherwise it's furniture.

A metric two people compute differently

It gets weaponised in a meeting within a month.

A percentage with no count

“Up 40%” is four out of ten often enough.

The forty-slide deck

If it needs forty slides I haven't understood it yet.

Who I am

Who you'd actually work with.

Gergana Tyaneva. I write the SQL and I present to the room — which is why the analysis survives the room.

2025 —

Rakuten Viber

Analytics & Strategy Manager. Subscription pricing, trial-to-paid, retention, campaign measurement.

2018 — 2025

StepStone

Product Analytics Manager. Seven years of product analytics and experimentation at marketplace scale.

2016 — 2017

C&A Europe

Built the brand team's first KPI framework and the tracking behind it.

2014 — 2015

AKQA

Measurement plans, KPI definition, tracking specs with dev teams.

Small on purpose. Not alone.

Tracking engineer

Server-side GTM, mobile SDKs, consent plumbing.

Data engineer

Warehouse and pipelines that still agree at 6am.

Experimentation

Test design, honest sample size, calling flat results flat.

Design

When a finding has to survive a board deck.

You hire me. I name anyone I bring in, what they cost, and you approve first. No account manager, no junior handover, nobody gets acquired out from under you — and no bench I have to keep busy, which is why the advice stays honest.

One question before you read on. I'd like to count visits with Google Analytics, which sets cookies and sends your IP to Google. Nothing has loaded yet, and the site works exactly the same either way. What's collected.