I'm aware it's been said before, but I just want to add my voice to those who are requesting a better party-based dialogue system as written in the quote below. That's really the one thing that is preventing this game from being truly great. D&D is about party cooperation.
The beauty of the table top is that it lends itself to interpretation and improvisation. The DM can wave a role if the player proves sufficiently clever and no video game yet possesses such flexibility. Computer games should play to their strengths, instead of rolling for persuasion or knowledge, reduce it to pass fail. If someone in the party possesses the required proficiency let it ride. If it is an NPC rather than the player, let their chance of assisting be dependent upon how invested they are in the outcome (is it their own quest? is it one they care even care about?) and their relationship with the player. That is sort of how real people interact anyway. Besides, on the table top, a failed roll stops nothing. Someone with a similar skill, or another skill but similar aims, can try again. That isn't an option in games like these, and shouldn't be because it would make the challenge feel redundant if one simply failed and NPCs always salvaged the situation for them (victory is meaningless without the possibility of failure). DnD is a deterministic game, help make the player's choices meaningful.
Second, the dice which are used in DnD were chosen, I imagine, according to two basic criteria: How well they roll on a physical surface and second how greatly they allow for variation. This studio should work with the strengths that digital systems have. Reduce everything to equivalent fractions. A 65% chance to hit doesn't need to be processed on a 1-100 scale. Reduce it to three parts, two of which result in success and then players won't run into nearly as many unlikely outcomes when they attempt an action. Round up where you need to in order to make it work, keeping it in the player's favor since the developers control literally every other aspect of the game except for the player's actions (and even those are heavily influenced by what that developers allow). Many developers will boast how their random number generators are accurate in 3,000, or 10,000 rolls but the truth of the matter is no player is ever going to track more than 100, and more than likely will make their judgment on 20. If you are going to use percentiles to help guide player decisions you should do everything you can to make them dead accurate. We have all played X Com right? We all have our own (numerous) experiences of such egregious bullshit it made us set down our controllers in disgust however much we enjoyed the game. One can accept losing do to poor coordination or inferior tactics, but losing when you have done everything right except getting a few rolls you needed is always bitter.