BLOG
backSo, you’ve probably seen it: the Moniepoint 2025 informal economy report is live, and along with it, an AI chatbot that can answer any questions you have about the informal economy, and a video series on invisible businesses in the informal economy. But what goes into making something like this a reality, and how could you build something like this for your industry?
Come behind the scenes with me!
Origin story
When we launched the first informal economy report in 2024, the response made one thing clear: people needed context on the businesses that make up about 90% of Nigeria’s MSMEs. But readers were limited to the analysis we provided. Sharing the raw data wasn’t possible because it’s proprietary, so the challenge was how to give people deeper access without exposing sensitive information.
On a call with Aderayo, Manager, Brand & Communications, in January 2025, the idea surfaced: “like a ChatGPT for the informal economy.” It clicked. We didn’t know how yet, but we knew it was the direction.
Building out the report
By April 2025, about six months before launch, we were officially in report mode. We started by revisiting what worked last year. Data on the scale of the informal economy, gender, location, industries, profits, and revenue were crucial, but we also expanded the scope.
Employment and ownership structures were a key new area. If informal businesses employed labour, their impact could be more far-reaching than initially estimated. Nigeria had also undergone major economic shifts, and we needed to capture how those changes affected businesses.
All of this became a four-page outline that acted as our north star as the project progressed.
Data, data, data
For the informal report, we use two types of data:
Quantitative (internal data) and
Qualitative (insights from on-the-ground surveys).
The internal data told us the “what” while the external data helped us understand the “why”.
Working with Habib, Head of Data Engineering, we reviewed what internal data could answer and what needed external validation. That helped tighten our approach and shape the questions for our surveys.
Getting the right context
With that concluded, we worked with Iwalola, Head of CX Research, to design our external survey. We widened our sample size to include businesses that didn’t use Moniepoint. Many of these informal businesses aren’t digital, so we couldn’t rely on online forms. We had people out there on the streets administering the survey to thousands of business owners.
Our total sample size was 3,200 businesses: 400 across all six geopolitical zones, plus Lagos and Abuja. Our main selection criterion: whether the business had CAC registration. The result was a much richer picture of Nigeria’s informal economy.
Analysis and writing
With all the data in hand, it was time to make sense of it. Analysis is like the skeleton of the report, the structure that holds everything together. Writing gives it flesh, turning numbers into meaning.
I reviewed the data from multiple angles: gender differences in income, regional motivations for starting businesses, and how economic shifts shaped behaviour. It required days of careful work, as any errors here would cascade across the entire report.
Then came the writer’s room. For the first time, our entire team worked together from the Lagos office for a week. We reviewed the draft line by line to ensure every conclusion held up. By the third week of July, the first complete draft of the 2025 report was ready.
Execution: bringing it to life
With the base version complete, it was time to bring it to life across all channels:
Reviewing and preparing the report for print
Producing the video series
And building M, our new informal economy chatbot
Let’s talk about M.
Meet M
Before we named it, M was simply “the ChatGPT of the informal report”. Gloria, a Senior Communications Specialist, was responsible for executing it.
There were two parts to M.
The website experience
The AI chatbot itself
The first step was reviewing the previous website to ideate what the new version could look like. This time, with M, the experience had to be different. The result was a website that told a story, not just about the data, but about the people behind it.
Meanwhile, conversations with engineers began. The vision was clear: a chatbot that felt personable, stayed grounded in our data, and provided accurate, contextually relevant answers.
Version 1 of the informal report website
Engineering M
We iterated through several versions. Early versions were slow and struggled with accuracy and context, so we went back to the drawing board.
Building the current one began with extracting information from the resources we had on the informal economy and storing it in a vector database. This vector database made it easy to retrieve relevant information based on semantic similarity.
When a user asked M anything, the system determined whether the question needed to be retrieved from the database. The next step was to take that retrieval and add a prompt. That prompt gave the LLM instructions on what to do with the retrieved information: the tone of its response, the necessary context, and guardrails to ensure the response remained relevant.
With that prompt, the LLM then gave a final response to the user.
*PS: this is a very simplified version.
Each response had to be accurate, consistent and fast. To make this happen, there were a few considerations made and a few things we learnt:
For the database, we had to find the right mix of chunking and overlapping that split the resources. This helped get the right information as quickly as possible while accounting for the likelihood that some resources in use might change.
We learnt that one thing that impacted speed was the amount of context that you pass into the LLM. Long prompts that required a lot of context made the LLM take a bit longer to respond.
Time-to-first-token varies by model, which affects speed. So we had to balance speed and quality, and we settled on GPT 4.1 through experimentation with different models.
After many iterations, we had a version we trusted. Gloria knew it had clicked the day she asked M to “explain the informal economy to a five-year-old,” and it nailed it.
Murder your darlings
About a month before launch, we had a working product. The website was put together, plugged in with the AI bot, and everything was in place, but something was wrong. We had two senior leaders test our website, and they both missed the chatbot completely. M sat quietly in the corner, unnoticed.
This was a huge problem. If people couldn’t see it, what was the point?
That night, after more user testing with the same result, Aderayo and the project's designer rebuilt the landing page to make M impossible to miss. It meant letting go of designs we loved, but it was the right decision.
The next day, we overhauled the rest of the site so everything fit the new direction. Finally, M was visible, accessible, and central.
Going live
There’s a lot more that went into the process that isn’t here. From state-to-state travel and countless edits to late-night reviews, but eventually, the pieces came together. What started as an idea on a January call became something much bigger: a living tool that opens up the informal economy for everyone to explore. That, for us, was the dream.
If you haven't checked it out yet, visit informalreport.moniepoint.com to explore Moniepoint’s 2025 report on Nigeria’s informal economy.
