Data Science Consultancy

Data science measured in euros saved, not dashboards shipped.

We build and ship production ML for e-commerce, logistics, and fintech teams: pricing, forecasting, routing, fraud, and retention. We get measured against the metric we agreed on.

One-page plan in 5 working days NDA on request No call needed
From our track record
Logistics · Amazon
Parcel forecasting feeding short-notice dispatch automation across the European network.
$5.3M
In annual savings, delivered
6 FTE
Manual dispatch
work eliminated
24/7
Production system
still running

Where margin lives or dies on pricing and retention.

For online retailers and marketplaces where margin lives or dies on pricing, demand, and whether customers come back.

Price every SKU at its right number, automatically. A pricing engine that reacts to demand, competitors, and stock instead of a quarterly spreadsheet review.
See demand before it arrives. Sales and demand forecasting per product and location, so buying and placement follow what is actually coming.
Know who is about to leave, and reach them first. Churn and lifetime-value models that flag at-risk customers while there is still time to act.
Match the right offer to the right shopper. Recommender and offer-matching systems that lift basket size and conversion instead of guessing.
Hear what your reviews are actually saying. Topic mining and satisfaction analysis across reviews and tickets, surfacing the recurring complaint and the quiet win.

Where a few percent of route cost is real money.

For operators moving physical goods, where a few percent of route cost or warehouse space is real money.

Cut the miles you do not need to drive. Route optimization that folds in traffic, weather, and delivery windows, and re-plans when reality changes.
Get more out of the space you already have. Warehouse slotting and storage optimization, so fast-moving stock sits where the pickers are.
Stop running out and stop over-ordering. Replenishment and demand forecasting that holds the line between stockouts and dead capital on the shelf.
Staff to the day you are actually going to have. Labor-demand forecasting for warehouse and dispatch, so shifts match volume.
Fix trucks before they break. Predictive maintenance that flags the vehicle heading for failure instead of waiting on the service calendar.

Where models decide risk, price, and trust in real time.

For lending, exchange, and trading teams where models decide risk, price, and trust in real time.

Catch the bad transaction as it happens. Real-time fraud and anomaly detection tuned to your risk tolerance, not a static rules table.
Price and match in real time. Matching and pricing engines for lending and trading that hold up under live volume.
Score risk on your data, not a generic curve. Credit and counterparty risk models built on the signals you actually have.
Keep the customers you paid to acquire. Churn and lifetime-value modeling for financial products, with the trigger to intervene before they go quiet.
Prove the model made money. Trade-success and execution analytics that tie the system to the profit and loss.

Every support conversation is data you are not using.

Why they are really calling. Contact-driver analysis that names the top reasons, so you fix the cause, not the ticket.
Who to call before they call you. Proactive outreach models that reach the account heading for trouble.
What the whole queue is saying. Topic mining and satisfaction analysis across calls and tickets at scale.
The routine ones, handled. Intelligent automation that resolves predictable contacts and escalates to a human only on low confidence.

Not on the list? Good.

Most companies that could use this don't call themselves tech companies. If the same decisions come round every week in your business, and you suspect they could be made better, send us the problem. We'll tell you on one page whether it's worth solving.

The teams behind Delta Labs have delivered production systems at Amazon, Mercedes-Benz, Crypto.com, Morpho, and Bluefin, some under NDA.

Experience of the founders, not a claim of client endorsement.

$5.3M+
In annual savings from systems we built
6 FTE
Manual work
automated away
10+
Production systems
shipped and running
~12 wks
Median time from
kickoff to live system

Is this for you?

This is for you if:

It is probably not for you if you want a dashboard handed over with nobody to own the outcome, or a six-month discovery deck. We embed, ship to production, and get measured on the number we agreed on.
You are an online retailer or marketplace and pricing, demand, or retention is run on gut and spreadsheets.
You move physical goods and a few percent off routing, space, or stockouts is real money.
You are a fintech or DeFi team and a model decides risk, price, or fraud in real time.
You run something else entirely, but staff, stock, prices or routes are planned on experience and spreadsheets. You don't need a data team to have a data problem, and that's often where we help most.

Four shapes of work, sized to the problem.

Most engagements start in one of these shapes and adjust if the work calls for it. Timelines and budget signal are starting points - we quote against the actual problem, not a template.

Entry engagement

Pilot

2–3 weeks · Fixed €6,000

One question, answered on your data. You get a working prototype, an accuracy and business-case readout, and a straight recommendation on whether to go further. It's how we start with new clients, not what our work costs.

Continue within 60 days and the full pilot fee is credited to the follow-on project. We run a limited number of pilots per quarter. The clock starts once data access is in place.

Start with a one-page plan →

Work we can talk about.

Work by our team, including in-house roles before Delta Labs. Some specifics are under NDA. Where we can name numbers, we do - and where we can't, we describe the shape of the work honestly.

De Graaf Logistics · Logistics · University collaborationNDA
Routing
Optimal routes under real constraints
Route optimization under operational constraints. A routing problem set by a Dutch logistics company through a university collaboration: find the optimal routes within the constraints the company actually operates under. Data and specifics are under NDA.
Crypto.com · ExchangeNDA
24/7
Liquid markets, 100s of instruments
Data science on the trading floor. Helping ensure liquid markets across hundreds of instruments, 24/7, for millions of active customers. Downtime and slippage have direct, immediate P&L consequences.
Morpho · DeFi · Consulting
Custom dataset
Built from on-chain primitives
On-chain analytics and protocol models. Built novel datasets directly from on-chain data to model rewards efficiency, lending-vs-borrowing rate spreads, and pool dynamics. The dataset didn't exist - we built it, then built the models on top.
Bluefin · DerivativesNDA
Decision support
On-chain derivatives venue
Decision-support analytics in DeFi. Decision-support and analytics for one of the major on-chain derivatives platforms. In DeFi everything is public - including mistakes - which raises the bar on what you ship.
Mercedes-Benz · Production ML at scale
10k+
Operators steered by our dashboards
Production AI engineering at scale. Large ML/AI systems deployed across production clusters - training pipelines, deployment, monitoring, retraining. Tens of thousands of operators were steered day-to-day by the dashboards and decision support we built on top.

How a project actually goes.

Five stages, regardless of size or complexity. Skipping any of them is, in our experience, how data science work quietly fails - usually because the modeling started before the problem was understood.

01
Your problem, on one page
Send a paragraph. Within 5 working days you get our approach, the data it needs, rough phases and a price range. A call follows only if you want one.
02
Deep understanding
We map the real business problem behind the proposed solution - often different - and the logic upstream of the data.
03
Solution
A simple dashboard or a state-of-the-art model - whichever fits. Deployed, monitored, with a client-side app if appropriate.
04
Measurement
Did we move the metric we agreed on day one? Against the original target, not a moved goalpost.
05
Ongoing support
Complex systems need maintenance. Many clients extend - the work naturally spills into adjacent use cases.

The questions you'd ask us anyway.

How long until we see something working? +

A pilot shows a working prototype within 2–3 weeks of data access. For a fixed-scope project, a usable v1 in 4–6 weeks is normal. Complex systems take longer - 3 months to a deployed system, more to wring out the operational edges. Your one-page plan says which one your problem is.

What does this typically cost? +

New work usually starts with a fixed-price pilot: €6,000 for two to three weeks, credited to the follow-on project if you continue. Fixed-scope projects start around €28k. Embedded engagements run €6k–€16k per week depending on time commitment. Complex systems are quoted against scope and usually land between €90k and €350k. Your one-page plan includes a price range before any formal proposal.

Do you sign NDAs? +

Routinely. Three of the case studies above are under NDA. We can sign yours before you send us any details.

Who owns the IP we build together? +

You do. Default contract: all work product, models, and code are assigned to the client on payment. We retain the right to describe the engagement at a high level - and even that we'll skip if you ask.

Can you work in our stack? +

Python and SQL by default. We've shipped on AWS, GCP, and bare-metal corporate environments. For greenfield: Postgres + Python + AWS unless there's a reason not to. For your stack: whatever you've already got, we'll adapt.

Can we talk to someone you've worked with? +

Yes, once we both think there's a fit. Much of our track record comes from in-house roles at larger companies, so references are former colleagues and managers rather than consultancy clients. We'll tell you exactly who you're speaking to.

What happens if we want to stop? +

Fixed-scope projects bill in two stages: a deposit and a delivery payment. Embedded engagements bill weekly with a two-week notice clause. We don't believe in lock-in - if we're not earning the next invoice, you shouldn't be paying it.

Get in touch

Tell us what you're stuck on.

If your problem is ambiguous, complex, or sits somewhere between the data and the code - that's our zone. Send a paragraph; you'll get one page back.

Prefer to talk first? Book a 30-min call

01
Send us the problem.
A paragraph is enough. No call, no questionnaire.
02
Get one page back.
Within 5 working days: approach, data needed, rough phases, price range.
03
Honest answer, even when it's no.
If we don't see a business case, we say so in two lines, before anyone spends money.