You really should have noted you were using on-demand pricing at the beginning of the article.
The calculus involved with matching RIs to forecSted demand in a dynamic growing workload is NP complete. And any inefficiency delta pays AWS. There's a small part of the curve where your savings is minimal if you don't match perfectly, it it easily goes to negative on the RI change.
You have to get pretty big to get discounts from published prices, far bigger than $1M/year, more like $1M/month.
To get a rep (that's unrelated), you have to pay full support which is about 10% per month (with a high minimum fee).