Embracing GenAI in research—let’s practice what we preach

Caveats: This is a very personal take on how GenAI has transformed my everyday work as a researcher/analyst, with the hope that it convinces a few of you to shift your perspective on personal GenAI usage. Part 1 is about what I do with GenAI, and Part 2 (coming soon) is about common myths and best practices. There’s probably a part 3 about business models and pet peeves.

This isn't about corporate metrics or broad organizational shifts, nor Accenture’s official position; I will not wax eloquent about how GenAI will solve hunger, crime, or eliminate the need for Batman. I do not own stock in any of the companies I talk about.

Why do tech analysts make for lousy tech users?

Ask any tech analyst about the merits of GenAI, and you'll hear gushing reviews. Ask them if they use GenAI to write their papers, and you'll either hear crickets or indignation (see cover photo). “Perhaps useful for people who don’t have English as their first language. I don’t have that problem, I can put words to paper,” one told me, dismissing GenAI as though it were beneath their skill level. I've heard others express similar views, describing GenAI as a tool that applies to everyone but them. A mere crutch for the less skilled. A shortcut to meh-diocrity. Ironically, the same people can be found most weekdays 9 to 5 on a rooftop near you, shouting about GenAI changing the world.

?I’m not advocating that analysts need to try every solution they recommend first-hand. Some of these are not accessible to us (Hadoop clusters), and some, like DevSecOps, are not targeted at us. That’s why we talk to actual users every single time.

?But GenAI is both accessible and intended for knowledge workers like us.

?Why the digital NIMBYism? Is it merely the age-old resistance to change, or is there deeper discomfort at play here—perhaps a fear that GenAI is encroaching on a creative domain where we consider ourselves irreplaceable? Or maybe many haven't found the right use case for GenAI in their work.

?

?Source: https://www.reddit.com/r/ChatGPT/comments/1d8j9su/i_want_ai_to_do_my_household_chores_and_i_can_do/

?Intrigued, I polled some brethren to gather anec-data on their use/attitude towards GenAI. This is what emerged.

?Anec-data: GenAI in research and thought leadership - adoption curve

Source: Mostly vox pop, unsolicited pings over Linkedin, catchups over lunch, etc. Numbers are directional (aka made up) at this stage. Image borrowed from https://www.business-to-you.com/crossing-the-chasm-technology-adoption-life-cycle/

So, here’s my personal take on the transformative impact of GenAI on research and thought leadership. Hopefully, it convinces a few of you to shift your perspective as it relates to your personal GenAI usage.

Who am I? A tech analyst/strategist. I rank among the top 50 users of ChatGPT within Accenture—yes, top 50 in a global powerhouse of over 733,000 people. #WeirdFlex #HumbleBrag

?I’ve been a research/strategy professional all my working life. Embracing GenAI has fundamentally transformed how I operate. The last time something this emphatic happened to me was when I used the internet for the first time.

Working without GenAI, at this point, feels like being asked to navigate the world without web access—a stark handicap.

?GenAI is slowly becoming key to how I process information, generate ideas, and produce content. It helps me streamline complex data analysis, refines my writing to make it more engaging, and even aids in conceptualizing innovative solutions that might take longer to crystallize in my mind alone. Here’s a rough idea:

?

My key use-cases

?

For writing:

?Managing writer’s irritable bowel syndrome (W-IBS?)

All analysts have to write; I regularly write, refine, edit, expand, and shorten my research drafts. But I have W-IBS? – a condition that somedays manifests as ‘writers block’ and other days as ‘writers diarrhea’. Sometimes nothing comes however much I try. But then I read up, think through, and a week and an epiphany later – come gushing 3000 words of pure train of thought. ChatGPT helps with both conditions. For example, I can convert my train of thought musings into polished, comprehensive documents, like a 2,000-word point of view or a 1,000-word byline. Or use it to make a terrible start on a page, which I viciously react to and make better. Together, we iterate. These are real examples.

?Humanizing Dry Technology Ideas:

I envy my research colleagues in fields like Sustainability (Babak Moussavi), Gender (Dominic King), and Disability Advocacy (Laurie Henneborn). Every paper and headline they write aims to save the planet, make our communities more equitable, and challenge our unconscious biases. People understand and actively support their work without needing a PhD in machine learning. I do too; doesn’t it just make for a better human-interest story?

?My headlines, on the other hand, mostly state that if you deploy <this> technology, you could make or save the organization money.?

In a world of Malalas and Gretas, I am but the cable guy.

?No sweat. To make complex tech concepts relatable, I try to find analogies and metaphors, humanizing them and, where possible, connecting them to societal good—not just bean-counting. This is hard. How do you make a paper about, say, microservices and APIs a human-interest story? ChatGPT makes it slightly easier to do this. Try it.

?A real example: I’ve likened a digital core to the "beating heart of an enterprise" and explained synthetic data through the lens of art, like redrawing the Mona Lisa in cubism using ChatGPT.

?Finding quirky historical insights: To spice up a paper on value of proprietary data, I asked ChatGPT to find me a counter intuitive story on proprietary data analysis. This yielded a story about how a data analyst at a large retailer, using proprietary data, found that people buy beer and pop tarts (more than water and medicines) before hurricanes.

?Streamlining Content Creation: I use ChatGPT to make PowerPoint presentations editable more quickly by converting images to text. For instance, after taking a photo of a useful table at an event, I had ChatGPT transcribe it into editable text, streamlining my content creation process.

?Organizing Information: In consulting, where acronyms are common, I use ChatGPT to bucket large lists into smaller, logical categories. A real example: “Organize <this> into memorable acronyms such as ABC (Assess, Begin, Capitalize) or CCC (Create, Capture, Contain).”

Identifying Applicable Models: I use ChatGPT to identify existing economic or business models that can structurally test my hypotheses. For example, if I hypothesize that a particular technology will disrupt market equilibrium, I ask GenAI to suggest established models that have been used in similar contexts. This helps ensure that my analysis is grounded in proven methodologies, enhancing credibility and relevance.?

For Numbers/Analysis:

?This is a recent development (Sep 2023). Ever since ChatGPT introduced multi-modality and a data analysis mode, I have been able to work with survey data and perform sophisticated analyses in real-time. It's important to exercise caution here, as this domain is typically for specialists, and misinterpretation can occur. Our quant researchers vet all my work. But there is immense value in generating initial insights even before it gets to them. It is empowering.

?With survey data, I use ChatGPT to both construct indexes to study emergent properties and/or study interesting correlations using English as code. For example, consider the following prompts:?

For index building: “I’ve uploaded survey data in excel. Create an index with questions Q1 to Q5, which are 3-point Likert scales. Assign 100 points for "exceeded expectations," 50 points for "met expectations," and 0 points for "below expectations." Total and average the scores across these questions. Store this data in a new column called “index score” at the right.”        

?Next iteration: “Tell me something unique about rows which score the highest on this index across other questions”?        

For fishing in the dark: “Analyze all data in the uploaded Excel file and identify the top 10 interesting observations. These could include anomalies, a very high proportion of unexpected responses, or vice-versa.”        

?This is not meant to be exhaustive, but to show that leveraging ChatGPT in this way is possible.

?Beyond ChatGPT—other GenAI tools I Use:

In addition to Enterprise ChatGPT, I use a few other GenAI tools (can’t name them for ‘reasons’) for complementary tasks. I use a foundation model-based visual creation tool for creating visuals for reports and presentations, which helps clarify complex ideas. This is what I made for my paper on synthetic data.

I’ve recently started using a discovery and search platform, which is amazing for searching through detailed transcripts and expert opinions. A writing assistant integrated into my Office365 knows the Accenture style guide fully and ensures I meet editorial guidelines—such as no Oxford commas, no UK spellings. There go my colo(u)rs, labo(u)rs, and hono(u)rs.

I’ve used ChatGPT replicas. I'm warming up to an AI productivity assistant to refine call summaries and adjust report text in place. Additionally, I regularly use Accenture’s internal GenAI tools to manage and summarize large volumes of proprietary data.

To end part 1: I firmly believe that a researcher with GenAI is better than one without. Why? Augmentation. I sometimes wish to be as eloquent as some colleagues (read this by Somak Roy), to have the technology depth of others (read anything by Tony Baer) or to have the statistical chops of others (look up Tomas Castagnino’s work). But I can’t, and they aren’t available to me in my hour of need. However, having a helping hand to get halfway there is easier today with GenAI.

?I leave you with this quote:

“Be not afraid of any task, no matter what its size. When self-doubt threatens, call on me, and I will equalize.”?– Unknown LLM, 2024

?To be continued in part 2 – GenAI for Research: Common Myths and Best Practices