AI smart charging: what it adds beyond load balancing

Most CSMS platforms advertise smart charging, and in practice it often means static load balancing: sharing a fixed amount of power among the vehicles plugged in. That is a useful baseline, and increasingly a legal one. This article explains what smart charging asks for beyond it, what the standards provide, and where AI can help without overpromising.

What the regulation requires

The AFIR defines smart recharging as the real-time adjustment of charging intensity based on information received through electronic communication. Public points built after 13 April 2024, or renovated after 14 October 2024, must be capable of it. All public points must also be digitally connected, meaning they can be monitored and controlled remotely (AFIR, Regulation (EU) 2023/1804).

The text sets a capability, not a method. It does not say how to optimise, and it contains no performance threshold. Load balancing satisfies the definition. What goes beyond it is your choice.

Why static balancing reaches its limits

A fixed rule such as an even split works at one site with a stable connection. It ignores three things.

The grid changes over time. The AFIR notes that smart charging helps integrate charging into the electricity system through aggregation and demand response, and that bidirectional charging helps further.

Prices change. Under the Electricity Directive, a customer with a smart meter can ask for a dynamic price contract from suppliers above 200 000 customers (Directive (EU) 2019/944). The same directive calls demand response pivotal to smart charging.

Vehicles and drivers differ. Not every driver needs a full battery, and not every session has the same departure time.

What smart charging uses beyond a fixed rule

Think of it as more inputs to the same decision: how much power each vehicle gets, and when.

Site and grid limits. The connection capacity, and any signal sent by the network operator or an energy manager.

Energy price signals, where the operator buys energy on a dynamic contract.

Session information, such as the requested energy and expected departure.

Carbon content of the electricity. The renewable energy directive requires network operators to publish the renewable share and greenhouse gas content of electricity, in time steps of one hour at most, in a form readable by charging points (Directive (EU) 2023/2413, RED III).

The protocol side is moving too. The OCPP 2.0.1 certification programme includes a Smart Charging profile, and OCPP 2.1 adds improved smart charging tools, support for ISO 15118-20 with bidirectional power transfer, and functional blocks for V2X and distributed energy resources (OCPP 2.0.1 certification, OCPP 2.1 announcement). Check which of these your chargers can actually execute: the software can only command what the hardware accepts.

Where AI can help, and where it cannot

AI is a method for making the decision better. It is not a source of capacity. A fair description of what it brings:

Forecasting. Estimating when vehicles will arrive and leave and how much energy they will ask for, from the history of a site. Forecasts are only as good as the data, and they are wrong sometimes. The system needs a safe fallback when they are.

Optimisation across constraints. Weighing connection limits, prices and driver needs together, which a simple rule cannot do.

Coordination across sites, for operators with several locations sharing contracts or limits.

What it does not replace: a correct connection limit, reliable charger communication, and clear rules about driver priority. Without these, a sophisticated algorithm only produces wrong decisions faster. This article gives no savings figure: results depend on the site, the tariff and the usage, and we have not found a published figure that would apply to your network. Measure on your own sites.

What to do this week

Check what your CSMS does today when a site is short of power: fixed split, priority rules, or something dynamic. Confirm that your chargers support the smart charging functions of the OCPP version they run. Pick one site, record energy delivered and charging power over a month, and use that baseline before testing any more advanced approach. Ask any vendor selling AI optimisation to show you the baseline it compares against.

GFO Omni is a CSMS that supervises charging points over OCPP 1.6, 2.0.1 and 2.1.