It's a great idea, in theory: Put the analog/digital converter (ADC) and its complement, the digital/analog converter (DAC) on the FPGA die.
This is technically possible due to converter architectures such as the sigma-delta design, which rely primarily on digital gates and substantial processing and filtering rather than precision analog components to achieve adequate mixed-signal performance.
So if your FPGA vendor has embedded that converter for you, you may have one less major problem: no worries about interconnect between converter and FPGA, much less about any formatting or protocol handling between the two, to cite two key points. But is that converter what you really need?
The answer is the same as for most engineering questions: It depends. If your converter requirements are modest, the integral converter may do the job. If your requirements are more stringent, however, the integral converter may not be up to the task.
But what defines your converter's "goodness"? Certainly, the first parameter pair you'll look at is resolution and sampling (conversion) rate. However, there's much more to a suitable converter than these two top-level factors.
If you're performing instrumentation and data acquisition tasks (such as seismic data), you'll look at linearity, offset, and drift with temperature. If you're focused on signal processing for things like software defined radio (SDR), then the specifications for the converter's signal-to-noise ratio (SNR), spur free dynamic range (SFDR), and noise floor will matter to you.
For example, Xilinx offers 12 bit/1 MSPS blocks, which are a possible fit with motor control, EKG/EEG medical, security, and monitoring applications. (See: Programmable Analog in Digital FPGAs?)
It's nice to know you can get a basic converter in your FPGA, but be careful: It's those second- and third-tier specifications that make a converter the right one for your application. You have to study those numbers carefully, and then decide if you really need a discrete, stand-alone converter instead, with more-comprehensive specs and likely better (and better-defined) performance.
Have you ever selected a converter that had adequate top-level specs, but fell short in the secondary ones and so had to be designed out, or needed a major software effort to try to fix its shortcomings?