Growth Marketing · Demand Gen · Lifecycle · Data

I automate the repetitive side of growth — so the focus goes to creativity and the opportunities others miss.

I'm Sara Beheyt — a growth marketer for scaling SaaS companies. Anything that has to be done twice, I turn into a system I supervise: dashboards, governance, AI automation. The machine handles the repetition; my attention goes to the creative work and the opportunities that repetition hides.

Get in touch LinkedIn GitHub

I automated a sales team's research job1 booked sales conversation for every 43 emails — researched, written and sent by a system, with zero manual prospecting timeRead the story ↓ I decided which products got built9 product bets settled without a line of code — 5 killed before they could burn engineering quarters, 3 validated, 1 shipped to buildRead the story ↓ I made the metrics honest — then beat them+24% projected engagement on 17% less volume, +5.6% CTR from AI personalization — measured on human opens, not machine-inflated onesRead the story ↓

The Journey — 2015 to today

Told in order, because each role taught the next one its method. The colors mark the growth lever each story exercises.

2015 – 2021 · Data & Measurement

First, learn to read the data

SYZYGY, London · BOSAQ, Ghent · 2015–2021

SQLMetabasePower BILookerGA4R

My career started in the data, not the creative. At SYZYGY, running paid search for Avis & Budget, I built campaign naming conventions that made per-market performance trackable — so budget followed real bookings instead of impressions. At BOSAQ I became the first marketer at a 4-person company, running three brands and three completely different growth motions on ~€100k; the SQL/Power BI/R dashboard I built decided where every euro went. Small budgets are the best teacher: when you cannot outspend a problem, you learn to out-measure it — and to spot the opportunities the spreadsheet quietly points at.

The naming conventions I designed for Avis in 2015 are the pattern I rebuilt at Gorgias in 2026: make things trackable before optimizing them.

→ The dashboard as a spec: metric hierarchy, tab structure, SQL patterns

One habit, four companies

Every stop: build the measurement layer, then scale what works

2022 – 2024 · Demand Gen & Experimentation

↳ Carried forward: read the data nobody else is reading — now inside a sales team

The prospecting tool that changed my career

Unbabel — translation AI, Series C · 2022–24

PythonWeb scrapingSalesforceGA4Looker

As a BDR selling translation software, I was prospecting manually — slow, and blind to who actually needed us. I learned the Python I needed in one week and built a tool that scraped Trustpilot reviews of enterprise prospects and checked one thing: did the company reply in the language the customer wrote in? A mismatch meant visible multilingual-support failure — a perfectly qualified prospect. Cold pitches became consultative calls, and the tool earned my promotion to Growth Engineer.

In the growth role I owned the trial-signup funnel as the company took its enterprise product self-serve: customer interviews exposed jargon as the conversion killer, a full UX rework with design followed, and A/B tests ran across acquisition and onboarding. The north star — words translated per month — grew at its +7% month-over-month target.

If a process bores you, that's the signal to automate it — a habit that has defined every role since.

How the Trustpilot tool qualified prospects

Public reviews as a demand signal

The call opened with evidence: "here is what your customers are experiencing." Data enrichment on the prospect's tool stack completed the picture.

2025 · Demand Generation

↳ Carried forward: a good signal beats a big budget — at Usercentrics, signals decided which products got built

My job was to find the company's next product

Usercentrics — incubation team · 2025

LinkedIn AdsGoogle AdsRedditQuoraGA4Looker

The incubation team's mandate: validate at least five new product concepts a year with real market demand — before any engineering spend. The biggest part of the success was the research behind each test: PESTEL-framed market analysis per concept — the audience and its culture, the local players, which channels that market actually uses — then ICP and persona definition.

The forecast came before the budget. Using Google's Keyword Planner I estimated search volume, cost, and expected impressions across each keyword landscape, and turned that into a predictive analysis of what a test should deliver — so every campaign launched with a target to beat, not a hope. I managed the budget accordingly: LinkedIn (the expensive channel) got deeply persona-segmented tests by role and seniority; Google Ads carried the broader keyword plays; Reddit and Quora reached the privacy-conscious communities.

I measured like an investor, not a media buyer — CAC, CLV, CPL, and qualitative product-need feedback — and defined the pass criterion itself: a concept passed when acquisition cost fell over the test window, because falling CAC means compounding demand rather than purchased attention. The best test drove cost per lead to €14 within one month; one validated concept, privacy-friendly analytics, went on to build.

Killing an idea with clean evidence is as valuable as validating one — and far cheaper than finding out after launch.

Nine concepts, one year — the verdicts

Each decided by live demand tests

The five kills are the quiet win: engineering quarters that were never spent on products the market didn't want. Validation signal: CAC falling over the test window — best test, €14/lead.

2025 – 2026 · Lifecycle & Systems

↳ Carried forward: ten years of building one-off systems — now build the system that rebuilds them

A lifecycle program where old campaigns create new ones

Gorgias — AI customer-experience platform, 15,000+ ecommerce brands · 2025–26

Customer.ioClaude CodeCargon8nBigQuery

I inherited ~15 campaigns running in parallel — some over a year old, carrying outdated product claims, stale designs, and naming conventions from three generations of previous owners. No registry, no renewal process: campaigns were built, launched, and forgotten.

Instead of rebuilding them one by one, I built the operating system that renews them: AI-assisted audits, rebuilds that must pull current product context, governance rails (naming and UTM conventions, mandatory eligibility segments), and the entire build pipeline — design, segment, copy, HTML, QA — codified as AI skills. Every renewal is logged in a campaign registry.

The estate now regenerates: Weekly Email reached its third generation, the winback campaign its second. The team closed Q1 2026 at 106% of its net-new ARR plan, with AI cross-sell at 167% — and the system was handed over as documentation, not tribal knowledge.

A campaign is an artifact; a renewal loop is an asset. Build the loop.

→ The system, generalized, on GitHub

The renewal loop

Every campaign is born with its next version scheduled

The pipeline behind steps 2–4 is public (generalized, synthetic data) in lifecycle-campaign-skills and lifecycle-ops-standards.

2026 · Data & Measurement

↳ Carried forward: the 2015 habit — make it trackable first — applied to an AI-era program

Making the metrics honest before optimizing them

Gorgias — Weekly Email, ~14,000 recipients · 2026

Customer.ioClickHouseSegmentationQA guardrails

A weekly newsletter looked healthy at 44.8% opens. It wasn't: the figure was inflated by machine opens, flows were broken, half the cohort silently never received emails, and one signal referenced a product that no longer existed.

Honesty came first: I separated human from machine engagement — real human opens were 22.7% — and reset the campaign's success criteria on human metrics before touching a single subject line. Then the rebuild: engagement-bucket segmentation (Active / Ignoring / Dark, each with its own flow), a two-step AI personalization pipeline with hard guardrails against invented links and product names, and a metrics migration to ClickHouse with the data intelligence team.

Results on honest numbers: the personalization test won with +5.6% CTR; the new segmentation projected +24% opens and clicks on 17% less volume. And one result that failed: an AI-tips variant QA'd at 4.0% CTOR with a 10.5% render-failure rate — I held it back, documented the negative result, and shipped the guardrail system it exposed. That guardrail system is now what the pipeline runs on.

Optimizing an inflated metric is decorating a broken house. Fix the measurement first — even when the honest number looks worse.

The same sends, two definitions of "open"

Weekly Email, February 2026

Apple's mail privacy features register opens no human made. Every success threshold was rebuilt on the green number.

2026 · AI & Automation

↳ Carried forward: the Trustpilot instinct, industrialized — automate the research, keep the human on the call

Automating a sales team's research job

Gorgias · 2026

BigQueryClauden8nCustomer.ioLangSmith

Booking sales calls required the classic SDR grind: find the right accounts, understand their situation, write a relevant email. Slow, inconsistent, and impossible to scale with headcount.

I built the whole motion end-to-end: account research on defined data traits in BigQuery, Claude-personalized sequences, orchestrated through n8n, delivered via Customer.io — with separate variants for self-serve and sales-led accounts. One signal-targeted wave of 854 sends produced 20 booked sales conversations — 8 demos and 12 CSM consultations — with every step attributed in the campaign dashboard. Alongside the data intelligence team, I then co-built a "context layer": an LLM that diagnoses 10,000+ accounts against product-usage signals and takes automated actions — +5% automated interactions in its first cohort.

AI doesn't replace the growth playbook — it removes the excuse for not running it on every account.

One campaign wave, measured end-to-end

Optimisation Signal campaign — sales-led accounts, 2026

1 : 43

one booked sales conversation per 43 emails sent

854 targeted sends → 20 booked conversations, zero manual research

8 demos 12 CSM consultations 8.0% opens 6.3% clicks

Every number attributed through the campaign dashboard — sends, opens, clicks, bookings, and the deals they created. The research and personalization ran themselves; the sales team took the calls.

How I Work

1 · Make it measurableNaming conventions, tracking, honest metrics — before any optimization. Every time.

2 · Automate what repeatsAnything done twice becomes a system I supervise — so attention is free for creative work and new opportunities.

3 · Decide with evidenceFalling CAC validates; rising CAC kills. Negative results are documented, not buried.

4 · Build for successionRegistries, documentation, governance — systems that keep working after handover.

Beyond Work

I build things from the ground up — at work and at home. I'm rebuilding my own house, I garden, and I distill and dry my own essential oils.

That interest shaped my 2024 sabbatical in Madagascar. On Nosy Be — the "island of perfumes" — ylang-ylang, vanilla, and a whole spice garden of other plants are grown and distilled for the world's perfume houses; at the Domaine de Florette distillery I got to see the craft I practice at home done at origin, at scale. From there I traveled north into the jungle, then south to Antananarivo and on to Andasibe for the wildlife — on an island where most species exist nowhere else on Earth — and spent part of the journey volunteering on the rebuilding of a school.

Based in Lisbon, Portugal, working remotely. French native, English bilingual, working Portuguese and Danish — I've lived and worked in London, Brussels, Ghent, and Lisbon.

What I'm looking for next: a company whose growth needs to be scaled and systematized — remote-first, AI-native, at the stage where the machine still needs building.

Photos of the house project, the garden, and Madagascar — coming soon.