The questions worth asking any CEA software vendor

Controlled-environment agriculture attracts strong software claims and thin evidence. That is rarely dishonesty. Words like open, autonomous and AI have no agreed meaning, so “open” can mean a documented API or a CSV download, and “autonomous” can mean a closed control loop or a timer.

Scepticism on its own just slows a purchase down. What helps is asking questions specific enough that a vague answer is visibly vague, and bringing your own site to the demonstration: a crop plan, a hardware list and a problem your team deals with every week.

On integration

“Is there an API?” is too easy to pass. Better: do your own apps use it? A supplier whose web and mobile apps run on the same authenticated API they sell you has a reason to keep it working. Then ask whether a third-party system can write as well as read.

List your makes and models and ask which are supported today, which need a gateway or adapter, and which would be new work. A supplier who names the ones that need a gateway is more believable than one who says everything works. Sharing a protocol such as MQTT does not mean two systems understand each other’s messages.

On control

Ask which features predict, which recommend and which actually send commands to equipment, and expect different answers for different features. Then walk through the awkward cases: a missing sensor reading, a lost connection, a manual override. Which software limits apply on each path, what does the equipment do when the software goes quiet, and how does an operator know a command was carried out?

On evidence

Ask what independently verified yield, energy or quality improvements exist, and treat any figure quoted before a deployment with care, ours included. The more useful question is about method: what would we measure, over what period, against what baseline?

That leads to a question for yourself. Do you know your kWh per kilo, water per kilo, labour minutes per kilo, cycle time and rejection rate today? If not, no supplier can prove they improved them. A clean baseline over a full cycle is the most useful work in any evaluation, and it is worth starting before you choose.

On leaving

Data ownership, export format, timing and retention belong in the agreement, not in a sales conversation. Ask for a sample export with the fields you would actually need: timestamps, units, zone and batch identifiers, observations and outcomes. Ask, too, what happens if the supplier fails, is bought or retires the product. Small suppliers get asked this and large ones often do not, which is the wrong way round.

The full checklist, with our own answers, including the ones where we still have limits, is on the evaluation page.