Hi @MaheshPS, great question, if not a little abstract. There are probably as many ways to measure the ROI of using DBRX as there are use cases for it, but let's try to focus on some more concrete examples.
- If you use DBRX as part of an internal knowledge base to help employees discover information faster, you could try to estimate the average time it takes someone to find the answer to a set of representative questions with and without it, then multiply through the number of queries your DBRX system gets to approximate amount of employee time saved. Multiplying in average or median salary could help you arrive at an approximate dollar amount per unit time.
- If you use it for customer facing interactions you can similarly approximate value by estimating the number of contact center or support employee hours are freed up after the implementation of your DBRX system.
- If you're moving from a proprietary LLM to DBRX, then it's a little more straightforward. Simply calculate your average cost per token for each over a timeframe and the delta is the numerator of your ROI on using DBRX. For example, GPT-4 costs around $38 per million tokens, but at sufficient volume DBRX can be >30x lower than that.
- Other applications, such as content moderation or quality monitoring, can be harder to measure a financial ROI for, but may be no less important if your company's brand and reputation are at stake.
Tell us what you're using DBRX for! I'm sure we can come up with some ways to approximate the ROI together.
