Deploying natural language understanding and generative AI in consulting

Finding good use cases for generative AI is a top barrier to its deployment. In this blog post I share some use cases for Natural Language Understanding (NLU) and generative AI in the world of consulting.

TL;DR use cases:

• Scale partners and senior leaders: Help junior analysts learn and develop without needing as much of senior management’s time.
• Distill complex documents: “Chat with documents” to find information quicker and more reliably than humans can alone.
• Speed up research: “Chat with research” to find citations/references for your own analysis.
• Help with critical thinking: Generate high quality questions for the client based on a corpus of prior work and the current project context.
• Write: Generate client deliverables that adhere to writing style guidelines. And, redline client deliverables to check for internal consistency, errors, and potential risk exposure for the firm.

Why consulting?

Consulting companies already have a great reflex for human in the loop activities. There’s a productive culture of peer review. Analyst work is reviewed by associates. Associate work is reviewed by VPs/Senior Managers. And, everything ultimately gets a high level look by the Partner. Adding an AI assistant is a natural extension of this cultural DNA.

Consulting companies have great data readiness and quality. These companies have certain types of projects that they’ve done over and over again. It produces a large corpus of proprietary data for training and fine-tuning. And, because consulting firms stress quality, accuracy, and rigor the work product is generally pretty good! They wouldn’t be training a model on garbage work.

Consulting is a huge business. Consulting has a lot of verticals, but to make the point let’s just consider auditing, tax, operations, finance, and strategy. The Big 4 accounting and consulting firms alone employ ~1.5 million people and generate ~$200 billion in revenue each year. The Big 3 strategy firms employ another ~85k people and generate ~$33 billion each year. On top of this, there are many smaller firms and also other multi billion dollar verticals.

Regardless of the vertical, most types of consultants generally do similar work. They research, conduct rigorous analysis, and prepare and communicate conclusions for the client.

Based on my experience as an early-career analyst in consulting, I believe the largest opportunities of deploying NLU and generative AI are:

• improving productivity and work quality simultaneously
• improving employee satisfaction
• supercharging learning & development

A tool for Valuation and Business Modeling consulting

I started my career as an analyst in the Valuation and Business Modeling group at Ernst & Young. I think it’s now called Valuation, Modeling & Economics. My team helped corporate clients value private businesses and assets. We provided a third party assessment that the client would use for tax, financial reporting, or strategic purposes.

The tool to deploy: Firms in this niche can and should deploy an AI assistant with a user interface similar to ChatGPT that’s trained and fine tuned on (1) internally produced client deliverables, (2) regulatory (tax and reporting) guidance, and (3) publicly available literature on valuation methodologies. In addition, they can prompt the assistant to adhere to a corporate style guide. Below are some use cases for such a tool.

Scale partners and senior leaders

Knowing what analytical approach to take for a project was typically an intuition built from years of experience. That's because we valued different types of assets for various purposes across many different industries and business cycles. As a result, early career analysts like myself had to lean heavily on senior managers and partners for direction. Analysts could learn much more quickly and independently with a conversational AI assistant at their side.

Distill complex documents

For some types of analyses, we had to understand a client’s capital structure and the rights and privileges of each investor. This information was buried deep in legal documents like Stock Option Agreements and Shareholder Agreements.

You couldn’t always fetch the information by running a keyword search; you actually had to read a majority of the agreement. And, it was very possible to miss something important. Analysts could use an AI assistant to find, summarize, and cite the needed information. And, the results would be better, faster, and more comprehensive than what you could’ve done on your own.

Speed up research

There was a fair amount of research that went into our valuation analysis. For example, we had to find supporting citations / references for model inputs like growth rates and profits margins. Traditionally, an analyst would download and scour long industry reports or public SEC filings. Similar to distilling legal documents, using an AI assistant would greatly improve productivity.

Help with critical thinking

Once I built an initial valuation model, we would kick the tires as a team. What assumptions did we make? Were we unsure about the financial projections? Are our model inputs reasonable? Ultimately, is our analysis sound and defensible or not?

During this process we would ask questions to the client to clarify our thinking. You can use an AI assistant to aid in the creation of questions. An AI assistant can produce relevant questions that may have never crossed our minds. This would help create a more robust analysis and reduce risk for the firm. It would result in less instances of “why didn’t we think of that?”

Write reports

Once we built conviction in our valuation, we would finalize our analysis and conclusions into a written report. This report included an industry overview, economic overview, company summary, a rationale for our chosen valuation methodology, a rationale for our chosen market inputs, assumptions, adjustments, description of our conclusions, and exhibits of the valuation models.

There was a template and guidelines to start with, but it always took longer than expected to distill our thinking into words. One of the hard parts was conforming to writing style guidelines. And, it was critical to make sure the document was error-free and internally consistent. This is the part of the project where generative AI can really shine, especially with a prompted style-guide.

Conclusions

These are just some use cases in a niche part of consulting. But, as I mentioned a lot of different types of consultants do similar work: they research, conduct rigorous analysis, and prepare and communicate conclusions. And, consulting companies have the DNA and data for good AI assistants plus the economic incentive to deploy for greater productivity and quality.