Predict growth. Automate decisions. Turn your data into momentum.
I use data science to drive real growth, not vanity dashboards. From predictive LTV and churn modelling to product led onboarding flows, pricing automation and audience targeting, I build systems that connect your data directly to the levers that grow your business. These are not just models, they're decision engines: they power dynamic pricing, trigger lifecycle campaigns, score leads, cluster behaviours and forecast revenue in ways no static report can.
Every signal is actionable, every insight is tied to outcome. Whether you're scaling a SaaS platform or optimising an eCommerce funnel, I turn your raw data into a compounding advantage, owned, production ready, and deeply integrated with your stack.
No guesswork. No lag. Just data that acts.
competitor_scraper.py
⇣ Predicted LTV from real usage data
Predicted LTV Actions
Based on behaviour, recency and product usage
| Customer | Segment | Predicted LTV | Recommended Action |
|---|---|---|---|
| C. Lopez | Power user | £1,850 | Offer annual plan |
| R. Singh | At risk | £120 | Trigger win-back email |
| T. Anders | New user | £480 | Wait and observe |
Decisions on a hunch
Your data is sitting on money you cannot see
Most businesses collect far more data than they use. It lands in spreadsheets and dashboards nobody opens, so the big calls, what to charge, who to keep, where to spend, still come down to gut feel and last quarter's guess.
- Data trapped in spreadsheets
- Numbers scattered across tools and tabs, never joined up, so the answer you need is technically in there and practically out of reach.
- Dashboards nobody reads
- Pretty charts that report the past and change nothing, while the decisions that matter still get made on instinct.
- Flying blind on the future
- No forecast worth trusting, so you over order or run dry, over hire or burn out, and cash flow is a monthly surprise.
- Churn and value, unmeasured
- You cannot see who is about to leave or who is worth keeping, so you spend the same on everyone and lose the ones that mattered.
Why me
A data scientist who reads a balance sheet
I am Georg. I pair machine learning with a finance background from the LSE, so the models I build are aimed at profit, not novelty: forecasting, churn and lifetime value, pricing, lead scoring, all wired straight into the tools you already run. Real, production ready systems you own, from the one person who builds them and stands behind the numbers.
What your data allows
Everyone promises growth. Here is what your numbers can actually carry.
Pick your industry, your traffic, your customers, your history and your budget. The planner runs the same sample size, budget and forecasting arithmetic I use before any engagement and tells you which stage you are in, what is statistically possible there, and what I would actually do with you. No email, no form, just the maths.
Your numbers
Industry
Clean historical data
Past paid campaign data
Monthly ad budget you could commit
Touchpoints you already have
Your stage
Foundations
No statistical tests yet. Best practice, clean tracking and channels that cost time rather than money carry the growth, and a small paid budget is for learning with kill rules, not for deciding.
Website tests
- Conversion rate
- 2 %
- measured from your numbers
- Visitors per variant to detect a 20% lift
- 21,110
- One A/B test on your traffic
- not in any useful time
- Tests a month at 20% lift
- 0
- 2 by 2 multivariate test in four weeks
- not yet
Nothing statistical is possible on your traffic yet. A change that doubles your conversion rate would still be invisible in a test, so this stage is about applying what is known to work and growing the traffic, not about testing.
Paid campaigns
- Assumed cost per conversion
- £30
- Conversions a month at your budget
- 10
- What that count can decide
- a signal, with a margin of error above 40%
- Days to 30 conversions
- 90
- Days to 100 conversions
- 300
- Learning budget before a channel is judged
- £1,500
- Kill rule for a losing ad
- 150 clicks without a conversion, about £120, and the ad goes
Your budget is below the learning budget for this channel. Spend it to learn about creative and audiences with kill rules, and treat every cost per acquisition you see as a rough signal, never as a decision.
Forecasting
Next month, roughly
Under six months of data supports a level forecast with a wide band and no seasonality. Expect errors of 25 to 40% or worse. A direction, re forecast monthly.
Growth curve
With no tests possible the curve is flat until the traffic grows. Best practice still moves the number, it just cannot be measured as a test yet.
What I actually do with you at this stage
- Tracking that counts each order once and attributes it sensibly, so the data is clean when the traffic arrives.
- The conversion fundamentals from 15+ years of accounts, applied without waiting for a test to confirm them for you.
- Channels that cost time rather than money, and the direct routes above where they fit your industry.
- A small paid budget run on brackets and kill rules so it teaches something without pretending to decide anything.
15+ years of growth, performance marketing and data science work, from well before the AI hype, are the prior that makes the small stages productive and the big ones honest. Nothing here assumes you already have the data. It assumes you want to know what you can do with the data you have.
Assumptions: 95% confidence and 80% power for tests, a two week to four week window, one win in three at a 6% real lift per completed test, and illustrative late 2026 UK and DACH costs per conversion by industry. Your competition, product, geography and season move those costs, which is exactly what the workshop pins down.
From Insight to Action
Core Data Science Benefits That Move the Needle
These are not just models or metrics. They are growth levers, deeply embedded into your product, marketing and pricing. From better segmentation and real time decisioning to churn prediction and LTV forecasting, I use data science to automate what used to be guesswork and expose what used to be invisible. Every insight is engineered to drive action, every model tuned for business impact.
- Smarter segmentation
Unlock more valuable audiences with intelligent clustering and scoring. Know who is ready to upgrade, who will convert with the right nudge, and who is worth reactivating.
- Customer journey analytics
See the complete path to conversion, and where it breaks. I track and model drop off points, time to value, activation friction and funnel bottlenecks with clarity.
- Automated decision making
Data that sits in a dashboard is wasted. I build systems that act, triggering campaigns, pricing changes or product nudges the moment the data says they should.
- Attribution clarity
Stop guessing what worked. I use probabilistic models and behavioural data to reveal which touchpoints actually drive conversions and retention, across channels.
- Dynamic bidding and offer logic
React in real time to market signals. Whether it's Google Ads, product feeds or on site offers, I optimise your decisions based on competition, demand and performance.
- Churn prevention and recovery
Spot users at risk of leaving before they do. My models trigger personalised retention flows, reactivation sequences or even product interventions to win them back.
- Product led insights
Understand what really keeps users around. I map feature usage to retention and upgrades, giving your product team direction grounded in real behaviour.
- Revenue forecasting
No more spreadsheets. I use cohort based models to predict MRR growth, CAC payback, LTV by segment and other critical growth metrics, with confidence intervals, not gut feeling.
Data Science for Growth
From Raw Signals to Revenue
This is the pipeline that turns your product and marketing data into measurable, automated outcomes. Every signal becomes a trigger, every model drives an action, and every part of your stack learns as it grows.
Marketing Personalisation
Triggered offers, behavioural flows, adaptive messaging
Campaign Automation
Pricing changes, ads, CRM sync, smart scheduling
The Whole Journey
From first click to confident decision
Data science is not a report at the end of the quarter. It is a pipeline that runs every day, and it only works if every stage is built properly. This is the arc I build for ecommerce stores, for Regios, for every client, designed on day one so the figures are simply there the moment you need them.
Track
A tracking strategy, not a pixel dropped on a page. Every touchpoint planned.
Collect
Events from shop, ads, emails and CRM land in one place, structured.
Clean
Duplicates merged, gaps filled, formats fixed. Rubbish in stays out.
Understand
Insights in plain language, not a wall of numbers.
Decide
Statistics with error margins, so you know when to act.
Forecast
Models that look ahead, so you plan instead of react.
Because the architecture exists from day one, no time is ever wasted rebuilding history. Whatever question comes next, the data to answer it is already flowing.
Collect & Clean
Your data is messier than you think. That is fixable.
Every real dataset arrives with duplicates, typos, missing fields and seventeen ways of writing the same city. Models built on mess produce confident nonsense. So before anything clever happens, the data gets cleaned, automatically, every single day.
What arrives
What the models see
Merged, completed, validated. Automatically, nightly, with an audit trail.
Understand
Insights, not table dumps
Most agency reporting is impressions, clicks and cost per click: about as useful as chopsticks with a steak. A proper analyst gets you to real revenue numbers, but still leaves you alone with a table. Real analysis ends in a sentence you can act on. Watch the same story climb all three steps.
What most agencies send
| Channel | Impr. | Clicks | CPC | Spend |
|---|---|---|---|---|
| Google Ads | 412k | 12,480 | £0.62 | £7.7k |
| 986k | 9,870 | £0.48 | £4.7k | |
| TikTok | 1.2m | 5,660 | £0.35 | £2.0k |
| Display | 2.3m | 3,140 | £0.19 | £597 |
| YouTube | 764k | 1,890 | £0.41 | £775 |
Millions of impressions. Not one number about money coming back.
What a good analyst sends
| Channel | Orders | CR | Revenue |
|---|---|---|---|
| Google Ads | 312 | 2.5% | £18,140 |
| Comparison sites | 158 | 4.9% | £21,020 |
| SEO | 224 | 2.8% | £15,930 |
| Direct | 201 | 2.9% | £14,880 |
| Newsletter | 171 | 3.8% | £12,270 |
| 118 | 1.2% | £6,410 | |
| Referral | 64 | 3.2% | £5,470 |
| TikTok | 45 | 0.8% | £2,210 |
412 rows closer to the truth. Still no decision.
What you get from me
The same data, understood
Customers arriving through comparison sites spend 2.3x more over their first year than social traffic, but get only 12% of the ad budget.
Recommendation: shift £1,800 per month and measure for six weeks.
Decide
Decisions with an error margin, not a gut feeling
Two campaign variants, one looks better. Is it real or is it luck? Statistics answers that question with a number. Early on, the uncertainty ranges overlap: acting now would be gambling. With enough data they separate, and the decision makes itself.
After 500 visitors
Variant A: 3.6% ± 2.3
Variant B: 4.4% ± 2.5
Ranges overlap. Could be luck. Keep testing.
After 9,600 visitors
Variant A: 3.7% ± 0.6
Variant B: 5.1% ± 0.6
99% confident B wins. Act.
Every Number Has A Range
A conversion rate is never just 4.2%. It is 4.2% plus or minus something, and that something decides whether you should care.
Known Error Margins
Decisions come with their uncertainty attached: act at 95% confidence, wait below it. No premature champagne, no missed winners.
Tests That End
Defined sample sizes and stopping rules, so experiments produce verdicts instead of running forever.
Forecast
Forecasting: the part that changes how you run the business
A good forecast turns management from reactive to deliberate. Different industries need different models: retail lives on seasonality, SaaS compounds, energy follows the weather. Pick your world below and see what the model has to understand.
Retail forecasting is seasonality first: weekly rhythm, promotion spikes and the December wave. Stock, staff and ad budget get planned against the curve, not against hope.
Models Per Industry
Seasonality, trend, promotions, churn, weather: whatever actually drives your numbers is what the model is built around. No one size fits all.
Plan With Confidence
Management sees an honest range months ahead: stock, hiring, cash flow and campaigns planned against the curve instead of last quarter.
Budgets From Forecasts
When you know what a month will bring, ad budgets stop being a guess. Spend rises into strong periods and pulls back before soft ones.
Marketing, Finally Measured
Against a modelled baseline, performance marketing shows its true added value week by week, instead of claiming credit for sales that would have happened anyway.
Automation & Dashboards
The right numbers for every level, updated by themselves
Nobody should export a spreadsheet to know how the business is doing. Dashboards run automatically and every level sees exactly what it needs: the board gets the big picture, management steers, operations acts. Everything is live around the clock, nothing is compiled by hand. Built exactly this way for Stint in London and for Regios.
Three numbers and a direction. Enough to govern, nothing to drown in.
Revenue vs forecast
+7.2%
Health index
78 / 100
Runway effect
+2.1 months
Real Time, Any Time
No month end wait, no PDF arriving two weeks late. Open the dashboard at 7am or 11pm and the numbers are current, because they update themselves.
Not 30% Of Your Retainer
Classic agency reporting is put together by hand and quietly eats up to a third of the retainer. Here reporting is software: it builds itself every night, and that budget goes into actual work.
The Right Story Per Role
C level, management, executives and analysts each get their own view of the same data, delivered at the altitude that role actually works at.
The Engineering Behind It
Real code, not spreadsheet magic
All of this is built in Python with the standard scientific stack, engineered like software because it is software: versioned, tested and running on a schedule, not living in a notebook on somebody’s laptop.
python · the stack
✓ pandas # data wrangling and cleaning
✓ scikit-learn # models and validation
✓ statsmodels & Prophet # forecasting and seasonality
✓ NumPy & SciPy # the numerical engine
✓ Matplotlib & Plotly # visuals humans understand
Versioned & Tested
Analysis code lives in git with tests, like every other part of your product. Results are reproducible, not one off miracles.
Pipelines, Not Rituals
Collection, cleaning, modelling and dashboards run automatically, nightly. Nobody has to remember anything.
Wired Into Your Product
Models talk to your Rails app, your shop and your dashboards through clean APIs, the same architecture GrowCentric.ai runs on.
Marketing Analytics 1:1
Want to know what proper marketing analytics looks like?
Most agencies report from the tables they can download out of an ad backend: clicks, impressions, likes, a cost per click. Those numbers describe the platform, not your business, and none of them tells you what to do next. I work from insights: a model of your demand, your customers, your channels and your risk, with an error margin on every number, so each decision has a reason. This series is the whole toolkit, chapter by chapter, exactly as I use it for clients.
What an ad backend gives you
- Clicks and impressions
- Likes and followers
- Cost per click
- Last click conversions
- A monthly PDF
What an insight gives you
- Which price pays, per segment
- When a customer will leave, not just whether
- Where the next pound should go, with a margin of error
- Whether the campaign caused anything at all
- A decision, its value and its risk on one page
Part one: how analytics helps
Part two: dependent variable techniques
- 4Modelling demand and price elasticity with regressionRead the chapter
- 5Polynomial distributed lags: how long does marketing keep working?Read the chapter
- 6Poisson regression: modelling how many times customers do thingsRead the chapter
- 7Logistic regression, lift charts and market basket analysisRead the chapter
- 8Survival analysis, churn and customer lifetime valueRead the chapter
- 9Panel regression and same store salesRead the chapter
- 10Forecasting demand: autocorrelation, seasonality and honest error bandsRead the chapter
Part three: interrelationship techniques
- 11Simultaneous equations: when marketing causes sales and sales cause marketingRead the chapter
- 12Principal components and factor analysis for marketersRead the chapter
- 13Segmentation: strategy before algorithmsRead the chapter
- 14Tools of segmentation: k means, latent class analysis and going beyond RFMRead the chapter
Part four: media and loyalty
- 15Modelling the value of marketing communicationsRead the chapter
- 16Media mix modelling: adstock, saturation and where the next pound goesRead the chapter
- 17Loyalty: the three Rs, the spectrum, and designing earn and burnRead the chapter
- 18Loyalty with structural equation modellingRead the chapter
- 19The customer loyalty journey: from segments to experiencesRead the chapter
Part five: testing and big data
- 20Statistical testing: sample size, lift and full factorial designsfrom 4 November
- 21Big data for marketers: what actually changes and what does notfrom 2 December
Twenty one chapters, English and German, with the formulas, the charts and worked examples on real looking numbers. Inspired by Mike Grigsby's Marketing Analytics; the explanations, examples and numbers are my own.
Your Stack, Versioned, Validated and Battle Tested
Behind every personalised experience or automated campaign is a serious stack, pipelines, models, infrastructure, and tracking layers that need to be clean, reproducible and production safe. I build with the same tools and practices used in high performing teams: notebooks that ship to production, models that self monitor, and pipelines that hold up to scrutiny. Whether you're a solo founder or have your own data team, this work integrates seamlessly into your workflow.
- Data pipeline orchestration
- Airflow, Dagster, dbt or fully custom Python jobs, orchestrated for reliability, observability and long term maintainability.
- Modelling and evaluation
- LTV, churn, segmentation and regression models using XGBoost, CatBoost, sklearn, scored with proper validation and tracked over time.
- Tracking and identity stitching
- Server side events, cookie less identifiers and anonymous to known user resolution built directly into your backend stack.
- ETL/ELT infrastructure
- BigQuery, Snowflake, Postgres, Fivetran, Segment. I integrate the tools that work best for your team and make sure the data flows cleanly.
- Reproducibility and versioning
- Model version control, MLflow tracking, git based pipelines and structured feature stores for consistent and auditable results.
- Notebook to production flow
- Convert exploratory Jupyter notebooks into production safe API endpoints used in your app, CRM or internal tools.
- Privacy aware workflows
- All data flows are consent first and GDPR compliant. From hashed IDs to compliant logging and pseudonymisation, privacy is built in.
Where are you right now?
Point me at your situation and I will take you to the right next step.
Serious Growth Starts with Serious Data Infrastructure.
If you're tired of hacks, dashboards and duct tape, I'll help you build a durable foundation that scales cleanly.
AI writes the code. Who is responsible for it?
A confident wrong number is worse than no number
AI will produce a dashboard in an hour. Whether the events behind it fire once, whether consent was respected and whether the model answers the question you asked is where experience earns its keep.
Measurement that lies
Events that fire twice, test orders counted as revenue, attribution to the wrong channel. Tracking is designed, tested and reconciled against the source of truth before anyone makes a decision on it.
Personal data in the wrong place
Consent enforced in the pipeline, not only in the banner. Server side events renamed when consent is missing so they can never feed an audience. GDPR and PECR as code, with the evidence to show it.
A model that fits the demo and nothing else
Sample sizes, baselines and uncertainty are stated before the forecast, and the forecast is tested against what actually happened. Evidence, not hope.
Where AI builds, where I decide
Low risk: AI builds, I glance
- Layouts and styling
- Forms and validation
- Boilerplate off the critical path
- Seed data, scripts, docs
Medium risk: AI drafts, I edit
- Edge cases for the tests
- Review as a second reader
- Refactoring under green tests
- Release notes from the tests
Critical path: I decide, AI advises
- Architecture and data model
- Money, tax, payments
- Authentication and permissions
- Every verified critical path
The lower the blast radius of a mistake, the more AI builds alone. The higher it is, the more it advises while a person with a name on the contract decides, reads and tests.
Why that costs you less, not more
- Account manager
- Project manager
- Designer
- Developer
- QA
- One accountable engineer
- AI on the low risk work
Agency team
Me, with AI
The hours for layouts, forms, scaffolding and documentation have collapsed, and that saving is in my package prices. The hours for judgment, architecture and proof have not, and that is what you pay for. One person, no hand-offs, no coordination overhead, and more tests than before because writing them is no longer the expensive part.
What you can check, not just read
- Tests you can read in plain language in your client hub
- Critical paths verified and frozen behind approval
- A release register and an audit log
- Code, data and infrastructure in your name
AI changed who writes the code. It did not change who answers for it.
How every engagement runs
Enterprise discipline, one person
Anyone can promise quality. I run every engagement through systems that prove it: a real client portal, legal clarity before work begins, and evidence for every release. This is what working with me actually looks like.
Your own client portal
Every client gets a seat in the Keferboeck Hub: onboarding, documents, dashboards and live project links in one place. Structured from day one, so nothing important ever lives in a lost email thread.
Legal clarity before work begins
Contracts, terms and insurance documents live in a portal vault. Signed versions are locked forever, every view and download is logged, and onboarding includes a formal review and acceptance step, with the option to hand it straight to your legal team.
Every release documented
Production releases are versioned, timestamped and written up: what was added, what changed, what was removed. You can always answer the question "what exactly went live, and when", because the record exists.
Tests gate every release
The full TDD and BDD suites must pass before anything reaches production, and the runs are recorded as evidence, with coverage tracked. A release that has not proven itself does not ship.
Verifiable by third parties
Critical business logic, how a calculation works, which legal documents a signup binds, is written up in plain language, approved by you, frozen at that state, and exportable as a PDF in English and German for your auditor. Changes reopen the approval, so the record never drifts from reality.
Security tested
Major production releases get web application security testing on top of the automated suites, grounded in formal offensive security training. Systems that hold customer and payment data deserve nothing less.
Insured, properly
A comprehensive insurance package is in place through the broker JMG: professional indemnity, public liability, employers liability and cyber cover. The insurer knows explicitly that my systems handle accounting grade workloads for SaaS products, loan interest and repayment calculations included, so exactly that work is covered.
And if I fall under a bus?
Everything is documented: architecture, decisions, processes, tests and the code itself, all written for the next person to pick up cold. If anyone else ever needs to take over, for any reason, the handover is instant and smooth. No key person risk, no vendor lock in, no hostage code. Very few agencies can honestly offer that; almost no freelancer can.
The other side of that coin: I play a central role in every project and I take responsibility for it. I do not vanish the moment a better offer turns up. The setup is simply built to change as you grow. When the workload gets massive, parts of what I do can move in house, your team owns the day to day, and I focus on whatever challenge matters at that stage of the growth cycle. That is the plan working, not a breakup.
How working with me runs, in full