Azure Document Intelligence was the best of the three big cloud document services on our handwriting sample, at 8.67% WER with reading order preserved. The errors were scattered, isolated word substitutions rather than structural failures, which makes the output repairable on a slow read.
Its real strength is printed documents and structured forms, where it is both accurate and inexpensive: raw OCR runs around $1.50 per 1,000 pages. For handwriting it is workable but not specialist, and its form-extraction features inherit whatever errors the handwriting recognition makes underneath.
If you are already invested in Azure and your documents are mostly typed, this is the natural choice. If handwriting is the main event, its error rate is roughly ten times a purpose-built engine, which turns into real correction time at volume.