Using AI without losing the human side of the work

Artificial intelligence tools are becoming part of everyday professional work, including in international development consulting, where they raise some particular questions about confidentiality, analytical rigor, and professional responsibility. In our case, the shift toward using AI has been gradual and practical. The goal has never been to automate thinking, but to make it easier to keep up with the volume of information that complex assignments generate without losing the part of the work that depends on judgment.

Recently, working with FAI at NYU Wagner on the Small Firm Diaries study, we began asking business owners how they were using AI in their work. After a few interviews, I realized I should probably ask myself the same question. The answer was not especially tidy. Yes, we are using these tools more every month. No, they are not making the work feel easier. If anything, they are forcing us to think more carefully about what parts of the job can speed up and what parts cannot.

Our projects typically combine interviews, workshops, surveys, financial analysis, training sessions, and long technical reports. That produces a lot of material — transcripts, notes, spreadsheets, photos from field visits, draft documents, and comments from multiple stakeholders. Over time we started using AI tools simply to stay organized, reduce manual work, and make it easier to see connections across different sources of information.

Meeting capture and transcription

One of the first areas where AI proved useful was documentation. For workshops, interviews, and technical discussions, we often record sessions using Zoom or dedicated audio recorders and generate transcripts using tools such as Fathom or other transcription software. These transcripts allow us to search discussions, verify quotations, and make sure key points are not lost.

In assignments involving sensitive information, recordings may be handled manually or processed offline to respect confidentiality requirements. Transcription tools help with efficiency, but they do not replace careful note-taking and review.

Drafting, synthesis, and triangulation

The most significant use of AI in our work is in drafting, synthesis, and triangulation of information. Large consulting assignments often require consolidating inputs from interviews, surveys, financial data, workshop discussions, and technical documents into coherent findings and reports. This usually means comparing different sources, checking whether they tell the same story, and revising conclusions as new information appears.

Language models are useful for organizing notes, suggesting structure, shortening text, and improving clarity, especially when deadlines are tight. They also make it easier to bring different types of data into one place so that we can review them together instead of one file at a time.

Sometimes the material we need to analyze is not even digital to begin with. In a focus group, results may be written on paper sheets taped to a wall in a small village workshop. We may take photos, transcribe the content later, and combine it with survey results, interview notes, and financial data. Being able to merge those sources makes it easier to see patterns that would otherwise stay fragmented. The goal is not to let the tool interpret the data, but to make triangulation faster so that more time can go into discussing what the information actually means.

Data, coding, and analysis support

We also use AI to support survey analysis, qualitative coding, and financial review. Tools like Excel and SurveyMonkey remain the core of this work. AI-assisted tools can help organize categories, clean text responses, and compare results across datasets, particularly in mixed-method assignments where numbers, interviews, and field observations all need to be considered together. For formula checking and data validation, we rely on Excel’s built-in capabilities or AI tools with dedicated code execution support, rather than language models alone. The tools reduce the time spent on mechanical tasks, but the interpretation still depends on the consulting team.

Confidentiality and limits

Because much of our work involves confidential client information, financial data, and interview material, we are careful about how these tools are used. Sensitive data are not uploaded to public systems without permission, and in some cases, notes are anonymized or processed manually before being used in drafting. Where AI writing tools are used, we apply them with attention to data settings and confidentiality requirements, and we continue to review our practices as these tools evolve.

AI-generated text is always reviewed by someone who knows the project well. These tools can produce language that sounds convincing even when the underlying logic is weak, so the responsibility for conclusions remains entirely with the team.

Field work, workshops, and conversations with clients are still at the center of what we do. The tools come afterward, when it is time to organize, compare, and write.

A practical note on the tools we currently use

People often ask what “using AI” actually means in day-to-day consulting work. In our case, it is not a single system but a collection of tools that help with documentation, analysis, writing, and communication. The mix changes over time, but the current stack looks roughly like this:

Meeting capture and transcription: Zoom recording, Fathom, AI transcription tools, and manual audio recorders when confidentiality requires it

Drafting, synthesis, and coding support: ChatGPT and similar language models, used with attention to data settings and confidentiality — for organizing notes, summarizing transcripts, and helping structure reports

Data and survey tools: Excel, SurveyMonkey, and shared spreadsheets, with AI-assisted tools for coding and consistency checks where appropriate

Document storage and collaboration: Google Drive, Word, PowerPoint, PDF tools, and Dropbox for shared workspaces and version control

Visuals and training materials: Canva and PowerPoint

Web and learning platforms: Wix for websites and online training spaces

Communications: Campaign Monitor for newsletters, updates, and articles like this one

None of these tools replaces the need to read carefully, check sources, or talk things through. They just make it easier to keep up.

Mixed feelings about the shift

If I’m honest, part of me wishes these tools had never arrived. They are helping, but they are not giving time back in the way people sometimes promise. In many cases they simply raise expectations — reports are expected faster, drafts are expected sooner, and the amount of material we are supposed to process keeps growing. Instead of slowing the pace of work, the technology often pushes it harder.

There is also a more uncomfortable tension underneath. The more efficient the tools become, the easier it is to start thinking of ourselves mainly as producers of output rather than as people working with other people. In our field, that is a real risk. Much of what we do depends on listening carefully, understanding context, and building trust with clients and communities. Those parts of the job do not speed up easily, and they should not.

Susana, a Small Firm Diaries Relationship Manager in New York City, put it plainly after reflecting on what the work had asked of her: “I cannot thank you enough because I know that was really hard for me to overcome — going and talking to people I don’t know, asking them to join the study. It was hard, but I’m glad I did it.”

That kind of work — showing up, asking, building trust with strangers — is exactly what no tool accelerates.

Our approach has always been high-touch and human, and that does not always sit comfortably with tools that promise speed above everything else. Using AI does not remove the need for that work, but it does create a constant pressure to move faster, write more, and deliver sooner. Learning how to use these tools without losing the human side of the work is not straightforward, and we do not always get the balance right.

Why survival matters

For a small consulting firm, the arrival of AI tools is not an abstract debate. It is part of the daily question of how to stay viable in a field where expectations keep rising while timelines and budgets rarely do. Using these tools has been less about innovation for its own sake and more about survival. If we can document meetings faster, synthesize notes more efficiently, and prepare drafts without losing days to formatting and transcription, we have more time for the part of the work that clients actually need.

And the reason survival matters is that the work itself matters, even if it rarely looks dramatic from the outside.

Over the years, our team has been part of efforts to make financial services more client-centered, to bring more attention to women’s leadership in financial institutions, and to design insurance products for people who would never have been considered insurable not long ago — women who have never had access to a doctor, small farmers exposed to climate shocks, motorcycle drivers who risk their lives every day to earn a living. This work moves slowly, often in small steps, but it only continues if there are people willing to stay in the field long enough to understand the realities behind the data.

Daniela, who manages relationship managers across multiple U.S. cities, observed something similar about the human side of coordination: “Everybody had a different approach. Everybody had a different formula to making it a success.”

That variety is not a problem to be standardized away. It reflects the reality that the communities being studied are not uniform either.

Learning judgment the slow way

Working this way also makes it clear that the future of this field depends on whether people coming into it are able to develop judgment, not just technical skills. Many of the students and new professionals we work with are extremely capable, comfortable with data and technology, and quick to learn new tools. What they have had fewer chances to build is the kind of experience that comes from being in situations where the answers are not obvious and the context matters as much as the numbers.

It is hard to understand what household income really means until you have climbed a steep path to visit a microfinance client living on the side of a mountain and realized that the numbers on the form describe something very different in real life. It is hard to design a survey for transport workers until you have tried to interview matatu drivers in a crowded bus station in Nairobi and discovered that half your questions do not make sense in the conditions people actually work in. Moments like these are where judgment starts to form. They do not come from reading reports or prompting a model. They come from being there, getting things wrong, and trying again.

The growing use of AI makes this tension more visible. The tools can help organize information, compare sources, and even suggest ways of structuring an argument, which can make the work look finished before it is fully understood. Part of our job, when working with students, graduates, and new team members, is helping them use these tools without skipping the slower process of learning how to think through a problem from the beginning.

AI can make good analysts faster, but it cannot replace the experience that helps someone recognize when something does not add up.

Johan, a junior insurance expert who works with us in Colombia, reflected after several days visiting farmers in Colombia: “I understand what you mean about having to put the work in — going to the field and hearing things directly from clients. You understand a lot more of what is possible.”

If the field stops training people who know how to do this kind of work, it will become shallow very quickly. Used carefully, these tools can help small teams keep up. Used carelessly, they can make the work look faster while quietly weakening the skills that make it valuable in the first place. The challenge now is not whether to use AI, but how to use it without losing the habits that make this work meaningful in the first place.