[PREDICTIVE MODELS]

Your bespoke predictive model, trained on your data

Galadrim develops and trains predictive models using your historical data, then integrates them into your business tools, to anticipate demand, a breakdown, or customer behaviour and improve your upstream decision-making.
+50 predictive models deployed and profitable for our clients
Up to 98% accuracy forecasting with our models
+140 senior tech profiles from Polytechnique, Centrale or ENS
Chat with an expertDiscover 5 predictive models
Neural network diagram - the predictive model trained on your data.
+800 organisations entrust us with their tech and AI projects
[OUR EXPERTISE]

Examples of Applications

We train models using your data, integrate them into your business tools, and maintain them in production, with a high level of precision
Demand and sales forecasting

Demand and sales forecasting

Time series models project your future activity based on your history and the factors that make it vary, such as weather, school calendar, or key commercial periods
Financial and cash flow forecasting

Financial and cash flow forecasting

Forecasting models project your turnover, receipts, and cash position over your financial year, with each projection accompanied by its confidence interval
Predictive maintenance and breakdown forecasting

Predictive maintenance and breakdown forecasting

Maintenance models learn the signals that precede an equipment failure and trigger intervention before production stops
Inventory and capacity optimisation

Inventory and capacity optimisation

Forecasting models size your supplies, staffing levels, and capacities based on the actual expected demand, day by day and site by site
Missing data estimation and indicator reliability

Missing data estimation and indicator reliability

Estimation models reconstruct measurements that your sensors or readings have not reported, so that your contractual and regulatory indicators hold up on a partially instrumented fleet.
Lead time and duration forecasting

Lead time and duration forecasting

Forecasting models estimate a delivery time or production duration from your historical data and current conditions, and alert you as soon as a deadline becomes untenable
Fraud and anomaly detection

Fraud and anomaly detection

Classification models learn what's out of the ordinary in your transactions, flows, or metrics, and flag suspicious cases in real-time
Scoring, churn, and customer segmentation

Scoring, churn, and customer segmentation

Scoring models assign to each customer a probability of leaving, defaulting on payment, or purchasing, so your teams know which accounts to prioritise
Recommendation and dynamic pricing

Recommendation and dynamic pricing

Recommendation models calculate the most relevant product and price for each customer and every moment of your commercial season
Marketing performance and attribution

Marketing performance and attribution

Regression models link your sales to the marketing actions that generated them, and estimate the expected return on a budget before you commit to it
[Projects]

Five predictive models in service with our clients

SCHOOL CATERINGTurboself

Predicting canteen attendance to reduce food waste

Turboself equips schools with payment and catering management solutions. Its clients ordered their food without knowing how many pupils would turn up for the service, and threw away the surplus. Galadrim developed the attendance forecasting algorithm integrated into its software, trained on historical attendance data, Météo France data, teacher attendance, and menu analysis by a language model. Each canteen adjusts its orders based on the expected daily attendance, and the model retrains as new data arrives.
5 500client sites
2Mdaily users
÷5waste costs
Turboself Logo
DISTRIBUTION NETWORKSSuez

Estimating water meter consumption to obtain a reliable network yield every day

Suez provides its client communities daily with the yield of their water network, an indicator that requires knowing the consumption of each meter. Part of the fleet remains without remote reading or provides irregular data, which undermined the accuracy of the indicator. Galadrim developed the estimation engine that fills these gaps: an XGBoost model predicts the consumption of meters with historical data by incorporating weather and local correlations, a statistical method builds typical profiles for poorly equipped fleets, and Monte Carlo simulations consolidate the whole into a yield with its confidence interval.
600 000inhabitants covered
59 000reliable meters
95 %of accuracy
Suez Logo
FinanceBNP Paribas

Making an algorithm reliable that builds the investment portfolios for two banks

BNP Paribas and Hello bank offer their clients a portfolio of stocks and money market funds built from a risk appetite questionnaire. Galadrim maintains and evolves the algorithm that produces these recommendations: input data control, calculation of input variables, error case handling, and optimisation of response times to meet expected thresholds. An advisor obtains a proposal consistent with the profile declared by their client, and monitoring tools signal a deviation before it reaches the clientele.
+200 000portfolios built /year
<2 sto build a portfolio
BNP Paribas Logo
INTERNATIONAL TRANSPORTLogfret

Assigning a product's customs code among 5,000 nomenclatures

Logfret's clients, specialists in international transport, classify each exported product under the HS customs nomenclature, a code chosen from 5,000 references on which the applied tariff depends. Galadrim developed the classification engine that assigns this code based solely on the product's description, drawing on all already processed declarations. Teams get an immediate proposal which they validate, on each declaration line.
5 000+customs categories
65countries covered
300employees
Logfret Logo
PUBLIC PROCUREMENTOrdiges

Pre-filling public procurement files through prediction

Ordiges publishes Lia, the software with which administrations, local authorities, and hospitals prepare their public procurement contracts: dozens of regulatory fields to fill in and a European nomenclature to choose from, even though very similar consultations already exist. Galadrim built the software's two prediction algorithms: the first finds neighbouring consultations and suggests the content of fields to reuse, the second suggests the nomenclature code corresponding to the subject of the contract. Each buyer's data remains segregated, and reused files are anonymised before being proposed.
+80 000indexed references
+40 areas of expertisein public procurement
Ordiges Logo
[Our method]

Our method for putting a predictive model into production

  • Framing
    Step 01

    Decision to be supported, target indicator, and accuracy threshold

    We start from the decision that the forecast must serve, and agree with you on the indicator to predict and the acceptable error level. This threshold becomes the criterion that will decide on the production launch, set before the first development.
  • DATA
    Step 02

    Audit and preparation of your historical data

    We audit your existing data, measure its depth and gaps, then prepare and enrich it with external sources useful for forecasting, such as weather, calendar, or market data.
  • MODELLING
    Step 03

    Model selection and training on your data

    We choose the approach adapted to your problem, whether time series, classification, or regression, then we compare the models on a test set that you validate. These models are deterministic: with identical data, the forecast is identical, and the calculation remains reproducible at any time.
  • INTEGRATION
    Step 04

    Forecasts delivered where decisions are made

    Our engineers display the forecasts in your business tools rather than in a separate dashboard, and integrate them securely and robustly into your existing information system.
  • DEPLOYMENT
    Step 05

    Production launch and deviation monitoring

    We industrialise the calculation chain, automate its execution, and implement monitoring that compares each forecast to the actual outcome. The deviation becomes visible as soon as it appears.
  • Sustained operation
    Step 06

    Retraining and continuous improvement

    We retrain the model on new data and adjust it when your activity changes, so that it maintains its accuracy year after year.
01/06
[Why Galadrim?]

Why do companies entrust us with their predictive models?

Proven expertise

800+ clients supported, 170+ experts, and 9 years of experience in AI product development, with predictive models in service in industry, public services, and banking.

Models trained on your own historical data

We audit your data, fill in its gaps, and enrich it with external sources that truly explain your variations. Accuracy is measured on your own data before any service launch.

Auditable and deterministic forecasts

With identical data, a predictive model always yields the same forecast, and each forecast is traceable back to the variables that explain it. Your decisions remain justifiable before a management committee, an auditor, or a regulator.

An integrated forecast where decisions are made

A model delivered alone remains a demonstration. We develop the tool that puts the forecast before your teams' eyes at the moment they order, plan, or make decisions.

Monitored accuracy after service launch

We compare each forecast to the actual outcome, retrain the model on new data, and adjust it when your activity changes, so that it remains reliable over time.

Let's bring your project to life together

Contact Us
[Our team]

A team that advises you before building

We recruit and train passionate engineers and experts from top schools, who combine Artificial Intelligence expertise with software engineering skills. This hybrid positioning allows us to design your bespoke AI systems and industrialise them in production.
Benjamin Drighès
Benjamin Drighès CTO Data & AI
Advises our clients on their product strategy, their long-term AI transformation, and their structuring choices of technologies and approaches, and defines the quality standards for our AI teams.
Quentin Massonnat
Quentin Massonnat AI Tech Lead
Designs the architecture for predictive computation chains, including data preparation, training, and automated execution, and oversees developments until production deployment.
Marc-César Garcia-Grenet
Marc-César Garcia-Grenet AI Engineer
He deploys features into production that expose a forecast within your teams' tool, from the model to the application that surrounds it.
Félix Monnier
Félix Monnier AI Engineer
Trains and compares forecasting models adapted to your data, then industrialises the one that meets the selected accuracy threshold.
Lucien Maillard
Lucien Maillard Principal AI Strategy Consultant
Frames predictive projects with client management, identifies the decision to be supported and qualifies feasibility based on available data.
Pierre-Antoine Dornic
Pierre-Antoine Dornic Principal AI Strategy Consultant
Maps AI use cases, prioritises them by ROI and feasibility, frames the executive boards' roadmaps and manages their deployment until team adoption.
Eva-Garance Tison
Eva-Garance Tison Lead AI Product Manager
Transforms business needs into adopted AI products: frames projects with clients, prioritises the roadmap, manages team Delivery and ensures the impact of delivered solutions.
Naïs Schietecatte
Naïs Schietecatte Senior AI Product Manager
Manages the product cycle of AI solutions, from functional specifications to the control interfaces used daily, in close collaboration with engineers and business users.
[FAQ]

Some frequently asked questions from our clients

Your historical data on the subject to be predicted, in its existing format, including exports from your ERP or business software. We audit it to establish what it allows us to predict, then we take charge of its preparation and enrichment using external sources that explain your variations.
We aim for the highest achievable level with your data and verify it on a test set before any commissioning. Our models in production exceed 95% accuracy for their use cases.
You retain the decision. Each forecast is delivered with its confidence interval, and your teams remain in control to adjust it when they have information that the model ignores.
Yes. A predictive model is deterministic and auditable: the same input data always produces the same output, and we keep track of the variables that influenced each forecast. You can therefore reference a decision before a committee, an auditor, or a regulator.
We continuously compare each forecast with actual results and make this discrepancy visible in your tool. As soon as it deteriorates, the model is retrained on recent data.
Yours. Code, trained models, prepared datasets, and documentation belong to you with each delivery. The transfer of rights is contractual from the outset.
Each project is scoped according to your requirements and priorities. Describe your context to us, and we will get back to you within 24 hours with an initial assessment.
[Contact us]

Let's bring your project to life together

We work with all types of clients, across all sectors. Whether you're an entrepreneur or managing a large organisation, a tailored team will meet your needs.

Over 800 companies have trusted us to create their web, mobile, and AI products

Your enquiry

We'll get back to you within 1 hour.