Full disclosure: I work for a company which offers these kinds of services.
Your question is complicated and there are multiple topics and considerations, so lets break it down:
1. Technical feasibility/methodology.
Change detection for Grassland -> arable or tree-covered -> arable using Sentinel data is possible and is done operationally in all sorts of contexts. Methods wise, the simplest is to find an index (for optical data - Sentinel 2) which identifies bare soil (should happen sometimes if arable, but not in permanent grassland or forest), and identify times when that index rises where it hasn't before. There are all sorts of other approaches too, including using machine learning.
This link about deforestation monitoring may be interesting:
https://medium.com/sentinel-hub/mapping-deforestation-from-sentinel-hub-de6aae67f817
In terms of existing open products which may be relevant, if you are based in Europe, the Copernicus Land Monitoring Service provides a tree cover product including change detection at 20m resolution. See here: https://land.copernicus.eu/pan-european/high-resolution-layers/forests
Reliability/accuracy/ground observations:
To be confident of your method, it is important to still have some ground observations. Use of satellite data can reduce the burden, and if confidence is not critical, even replace such observations. Required reliability/accuracy depends on your own use-case. For example, if the aim is to narrow down areas for further investigation by other means, a lower accuracy is required than if the goal is to provide direct high quality evidence for legal reasons. There is also a component linked to the complexity of the methodology (cost to you in human resources) and required data (data processing or access costs). More complex methods may give higher accuracies, but will cost you more. So therefore it will be quite a challenge to answer this question in your context.
Financial and human resources costs:
This depends as I say above, on what approach you take, but also other factors such as what area you wish to monitor, and how frequently. A once-a-year analysis of a few hectares is fundamentally different to continuous near-real-time monitoring of countries and continents. So again, this is not straightforward or something for which there are easy resources to point you towards.
One data access and processing option which I have found to greatly speed up development and streamline production is https://www.sentinel-hub.com/ (see also the first link), but it comes at a cost. It does however significantly simplify the process of getting hold-of Sentinel data and processing it, reducing human resource costs (which are likely to be biggest as long as your organisation can bite the bullet of paying for "free" data). You could also consider google earth engine: https://developers.google.com/earth-engine/tutorial_forest_01.
Lastly, and I don't mean this as an advertisement, but you may wish to reach out to a remote sensing company who have the resources, expertise and infrastructure to do this (if you need more than just analysing a small area e.g. once a year). They should be able to quote you a price for development and/or provision of a service.
