Shrimp AI

Satellite-powered demand prediction platform

Know where demand is heading before deciding where to supply.

Project

Shrimp AI

Service category

AI / ML

Tech stack

Custom AI model · Demand prediction · Weather data

Here's the gist

A shrimp-production company needed a better way to understand where demand for its seed would grow. Until then, supply decisions relied largely on guesswork, making it difficult to know which regions were worth exploring.

We built Shrimp AI, a prediction platform that uses a custom-trained model to study regional demand patterns. It analyses historical sales, weather, pond colour and pond shape to predict and score where demand is most likely to rise.

The field team can then focus its efforts on the markets with the strongest potential.

The Challenge

Demand wasn't easy to predict. The company needed to understand what was happening across different regions before committing its supply.

Relying on gut feel made that decision harder and left opportunities difficult to identify.

So we built Shrimp AI

A custom-trained prediction platform that brings multiple data points together to identify demand opportunities.

Look beyond sales

Analyse historical sales alongside weather and pond-related data.

Read regional patterns

Study pond colour and shape to understand local conditions.

Predict demand

Score regions based on their likelihood of higher seed demand.

Help teams prioritise

Give field teams a clearer idea of where to scout first.

Did it work?

A big yes.

+30%

client revenue increase.

Custom-trained

A model built for this exact problem, not a generic tool.

Satellite-driven

Pond and regional signals read from above.

Scout-first

Field effort now goes to high-demand zones.

The client now decides where to sell using evidence rather than habit, and has a repeatable way to find new markets before anyone else does.

How we got there

From gut feeling to a ranked list.

How we got there
Step	What we did
01Listen first	What we did We sat down with the sales and scouting teams to learn how they decide where to go today, which signals they trust, and which decisions the platform needed to support.
02Get the data in shape	What we did We organised years of sales and environmental data by region, so the model could learn which conditions tend to show up before demand does. Gaps and inconsistencies got fixed before any training began.
03Teach the model	What we did We trained a custom model for this specific problem and tested its predictions against regions and periods it had never seen, so the scores could be trusted.
04Put it in their hands	What we did We built the scoring and scouting platform so the output is an action list, not just a map. The team sees where demand is predicted, why, and which regions to visit first.