Businesses assume they need to hire expensive data analysts to make sense of their data. Conversational analytics are showing that there’s a better way.
Two paths for business analytics
After years of “talent wars” talk and warnings about a shortage of data professionals, the market for data jobs is in a downswing. Open data positions have declined 35% year over year as of September 2024.
Whether or not hiring picks back up, we think companies will try to keep talented people and then upskill their data abilities.
This will start with leaders pausing to ask, “Wait, before we go to invest $500,000 to hire and train five entry-level data analysts, is there a way we can empower non-analysts to do what new hires would do? Can we free up the capacity of our experienced data people?”
In other words, these leaders will question a central tenet of data-driven business: that the company with the most and best analysts digs up the best insights and makes the best decisions. Rather than make analytics dependent on macroeconomic hiring trends, they will turn to conversational analytics, a class of data tools that enable non-coding “business users” (i.e., people in various desk jobs) to explore, crunch, and use data on their own.
What are conversational analytics?
With conversational analytics, users ask questions about data in plain English and get answers in seconds. While that sounds like ChatGPT, it is not—no executive in their right mind would share sensitive data with an LLM. They hallucinate, they’re terrible at keeping secrets, and they don’t sign NDAs.
Normally, when data lives in a secure warehouse, like BigQuery or Snowflake, someone has to write code in SQL, Python, or R to analyze it. That process is slow and costly.
Conversational analytics take a different approach—at the least the way we do it at DataChat. We keep data safely in your warehouse. We don’t require any coding skills or advanced training in statistics. While we can leverage LLMs to generate answers, we do so without sharing any of your data. Plus, our conversational analytics explain how answers were reached, so you can validate them.
To be clear, conversational analytics don’t replace data scientists or data engineers doing advanced, specialized work. They do, however, give anyone in your organization the ability to explore and interrogate data that could inform their work, boost their performance, and lead to quantifiably better decisions.
And the ROI of that?
We won’t pretend like we can calculate the impact of better decision-making without knowing anything about your business. Still, we can make some conservative claims about the ROI of conversational analytics:
Speed to Insights
At most companies, “business analytics” are just dashboards—assemblages of graphs and metrics that update somewhat regularly. To make a dashboard, it takes anywhere from 3-5 days on the lower end, 2-3 weeks in the mid-range, and 4-6 weeks on the higher end. Using conversational analytics, you could create the components of a dashboard and share them with your team in five minutes flat. You’d get the same information, plus an opportunity to act on it days or weeks sooner than you would normally.
Labor Costs
365 Data Science, an educational platform, crunched 1,000 data analyst job postings from 2024. They find that entry-level data analysts make $61,000 to $101,000 per year. Over half the job postings expect applicants to know SQL, and a third expect Python skills. Conservatively, it would cost $900 over three days of work for an analyst paid $80,000 annually (working 260 days per year) to make a simple dashboard. That figure hits $6,300 for the 3-week dashboard. For comparison, the five-minute dashboard would cost $3.20 in labor.
Technical Debt
Some time ago, one of our colleagues consulted for a company that relied on dashboards. That company laid off the people who coded the dashboards, which soon broke—something that happens without maintenance. There was no record of how those dashboards were made. Turns out, the dashboards were not only broken but had been generating inaccurate metrics. A difference with conversational analytics is that it documents exactly how each visualization or metric is derived. It’s easy to vet, easy to change, and easy to replicate. You don’t have to pay a consultant to fix or reverse engineer your business analytics.
Although you could use conversational analytics to build dashboards, we think that undersells the real potential, which is to think, explore, and test assumptions iteratively.
Iterating with conversational analytics
At most companies, any analytics beyond the cookie-cutter dashboards are made-to-order. Someone has to request them, and someone with data science skills has to create them. By the time the analytics are ready for use, they’re late. Conversational analytics upends that flawed process by giving domain experts, with or without coding skills, the ability to explore data at the speed of thought. It’s an iterative process in which basic questions about data lead to more poignant questions, which lead to strategic questions and decisions.
Say you lead a direct-to-consumer (D2C) ecommerce brand getting ready for Black Friday. You have a trove of data from last year. You start iterating:
- Which Black Friday deals yielded the highest volume of sales in 2023? What about the highest average order value?
- Of customers who used a Black Friday deal, which were not already customers? Which have made subsequent purchases? Which deal was mostly likely to drive two or more purchases between November 2023 and October 2024?
- Which deal was associated with the highest predicted customer lifetime value, based on average order value and purchase frequency?
- For shoppers with a predicted lifetime value above $XX, what SKUs were they most likely to have purchased?
That’s five minutes or less of iterative analytics that could enable a D2C brand to optimize its Black Friday deals for customer acquisition, order value, and lifetime value. Those moments of exploration could be part of every work day, for anyone with curiosity and decisions to make.
A tale of two ROIs
We laid out two potential paths for data-driven businesses. Option A: hire entry-level analysts, pay them to code other people’s questions, and get answers long after you wanted them. Option B: implement DataChat’s conversational analytics, let experienced team members ask questions in plain English, and make data-driven decisions minutes later.
A business can understand itself without hiring armies of analysts. You already have enough talent and knowledge. You just need the right tool.