Advanced analytics have to be nurtured with data. Intelligent insights achieved with models trained by learning techniques can only emerge from large amounts of data properly and efficiently acquired, stored, cleaned-up, pre-processed and made available to massive distributed computation.
Trained models then have to be run on live data, and they can be effective only if the data is safely and efficiently piped-up from the field, stored, pre-processed and made available to them. Often, intelligence has to be run on the edge and models have to be smoothly moved down in the field-to-cloud chain. This is Data Engineering.
Finally, devices on the field (the Things) must produce good data. This requires effective management: provisioning, versioning, security, connectivity, diagnostic, service management, planning, etc.
Of course, the business value is higher going up the pyramid, but the amount of high-quality infrastructure and software, key to the deployment of that business value, is much higher at the lower levels.