Skip to content
Expertise

Forecasting & statisticsSee the peaks and shortfalls coming.

Week-by-week forecasts, so you can order, hire or arrange financing in time.

Book a call

A forecast, and the decision that goes with it

From your history, we forecast the weeks ahead with a likely range, and the forecast lands where you make decisions.

Sales of a drinks rangeunits per weekExample, fictional data
02505007502468101214161820222426WeekTodayPossible stock-out
02505007501014182226WeekTodayPossible stock-out
  • History
  • Forecast
  • Likely range (8 chances in 10)
  • Possible stock-out

The decision

Order before week 19

At the top of the range, forecast sales use up the stock of 3,060 units in week 21. The supplier needs 2 weeks to deliver.

What we do

  1. We start from your history

    Sales, workload or cash, with their seasons, promotions and bank holidays.

  2. We train a model

    On your data, every forecast comes with its likely range (Python).

  3. We build it into your tools

    ERP, Power BI, Excel or rota: the forecast lands where you make decisions.

  4. We track its quality

    Forecast and actual compared every week, model retrained if it drifts.

Examples of what we deliver

Each card starts from data you already have and ends with a concrete decision.

  • Finance

    Budget and year-end outlook

    The year end projected every month, with the variances explained.

  • Teams

    Workload and scheduling

    Upcoming volumes forecast so rotas, leave and extra staff are planned ahead.

  • Sales and stock

    Stock and replenishment

    Quantities to order, worked out from the forecast, stock levels and lead times.

  • Risks

    Anomaly detection

    Duplicate invoices, data-entry errors and unusual expenses flagged in time.

  • Risks

    Predictive maintenance

    Signs of wear spotted in machine data, before the breakdown.

  • Sales and stock

    Footfall forecasting

    Customers, covers or visitors forecast day by day, to plan purchasing and rotas.

Our approach

Your data stays with you, and we stay on after go-live.

Discover our approach

Where to start?

With a first call

30 minutes with us, by video if possible or by phone. You tell us about your situation, we tell you frankly where we would start.

Book a call contact@datafish.info

Frequently asked questions

The questions we get asked most often.

Do we have enough data?

Usually yes. For many use cases, 6 to 12 months of history already gives useful results; to capture a seasonal pattern properly, two years is better. We check this at the start, on your data, before going further.

How does a first project work?

We start with a prototype (POC): in 3 to 5 days, we train a first model on your data and compare what it would have predicted with what actually happened. You see the concrete gain before funding the full solution.

How much does a forecasting model cost?

Each project is priced based on your scope: what needs forecasting, the data available and the tools the forecast has to feed. We assess it together during an initial scoping session, and the prototype lets you see the result before you decide.

Do we need a data scientist in-house?

No. We build the model, integrate it into your tools and train the people who use it. Your teams are involved from the audit onwards: they are the ones who know the promotions, the customers and the seasons.

See our training

Is our data safe?

It stays in your environment, hosted in France by default: the model is trained and runs wherever you decide. Everyone only sees the forecasts they are allowed to see, using the accounts your teams already have.

See the technology building blocks

How often are forecasts updated?

As often as you make the decisions: daily for footfall, weekly for orders and rotas, monthly for the budget. New data is picked up automatically, with no re-keying.

See the project

Ask another question