We support agronomists and farmers with software to create VR maps. 

Some of popular practices on seeding maps are described here: 
https://docs.geopard.tech/geopard-tutorials/agronomy/variable-rate-seeding-planting-maps

Doing this since 2012 I see popular patterns and suboptimal decisions: underestimation of field potential multi-year vegetation (you can add or remove years to automated maps) and topography factors (slope, ruggedness); overestimation of grid soil data or ssurgo maps (suggestion is make good zones and then zonal sampling or soil scanning with smth like soiloptix). 

Trend of the latest years (especially on big fields) - cluster zones for seeding & fertilizer operations.
- Described cases here: 1m acres VR implemented (we tested here multiple hypotheses with company's agronomists, and still improving the process)
- 200k acres VR implemented from low start (field potential, soil brightness as proxy for OM, topography factors -> soil sampling -> VR seeding & fertilizing)

In case you're OK to share some data (in case no yield data, we can generate synthetic yield maps) - I can support you with some hypotheses (or you can just try these approaches under free trial of our software). 

PS: we're working now of automated yield limiting factors detection, so for each field you'll receive custom layers, recs, zones.