new research in International Journal of Research and Innovation in the Social Sciences A survey of 200 women micro-entrepreneurs in Bangalore found that 32.5% of them reported “actively using at least one AI-enabled tool” in their business. The paper said the tools include KhataBook (a digital ledger), WhatsApp Business auto-replies, Canva, and Facebook’s “best time to post” suggestions.
As the women in the study experienced, none of this is artificial intelligence.
The underlying research doesn’t matter. The Bangalore paper makes some truly useful discoveries. Women adopt digital tools when trusted NGO intermediaries provide ongoing peer-based training, and abandon them when training is one-time or absent. This finding has been the consensus in ICT4D since Heeks established it in 2010.
What’s new is the label. The paper described a digital literacy intervention and called it AI empowerment, and peer-reviewed journals accepted the framework. This is not an isolated drafting error. It’s the fairy dust of the development department.
If you call something “artificial intelligence,” donations will fall from the sky.
Subscribe now to find AI Pixie Dust in your inbox!
Inflation is now commonplace
The Bangalore study defines “AI tools” to include things like digital bookkeeping apps, WhatsApp auto-replies, social media post suggestions, and SMS inventory reminders. For readers in 2026, who are promised generative AI, agent systems, and machine learning models trained on billions of parameters, that list is a category error.
KhataBook runs machine learning on the backend for fraud detection and user profiling. None of these were used by the women in the study.
- They record credits and debits on a digital ledger and send payment notices that are digital versions of the paper notebooks their grandmothers kept.
- WhatsApp Business gets auto-responses from scripted templates.
- Canva’s time-of-post suggestions are recommendation heuristics that have been around long before the current wave of AI.
Calling these “AI tools” extends their rule-based and template-based capabilities to something other than that.
Authors aren’t the only ones who do this. Tool vendors sell the same combinations, grant applications copy the vendor’s language, and researchers writing for fast-turning journals adopt the grant’s language. The Bengaluru paper was received on February 26, 2026 and accepted on March 3, 2026, resulting in a six-day review cycle.
Any review that happens at that pace is unlikely to ask definitional questions, and once a paper is labeled “peer-reviewed,” the next grant proposal can cite it as evidence that AI tools improve women’s business confidence by 73%.
SEC Fine for AI cleaning
In March 2024, the SEC settled charges against Delphia and Global Predictions for telling investors they used AI and machine learning when they did not. Together, the companies paid $400,000 in civil penalties. SEC Chairman Gary Gensler used the term “AI washing” and said, “Investment advisers should not mislead the public by saying they are using AI models when they are not.”
Since then, academic researchers have defined AI washing as the act of exaggerating, misrepresenting, or falsely claiming the capabilities of an AI for reputational or financial gain.
The donor is not the SEC. For development implementers that promise AI-driven decision support and offer templated chatbots, there is no form ADV. The financial sector was fined for precision. Inflation is accelerating because the development sector does not have an equivalent enforcement function.
Cost is not rhetorical
When the next funding cycle calls for an “AI for Women’s Empowerment” program, three things will happen.
1.The same work will be repackaged under a new label.
This is the lowest-cost path for implementers who already know that the model works, a path made easier by the Bangalore paper.
Donors who read an article that says, “73% of women using AI tools supported by NGOs reported feeling more confident in their decision-making,” have no way of knowing that the underlying tool is a digital ledger. Donors fund AI. Implementers provide digital literacy.
Both sides get what they want, but the evidence is never built on actual AI tools.
2. AI tools will be deployed based on this inflated evidence.
When credit scoring vendors, generative chatbots for health advice, or AI-driven fraud detection systems enter the same market, they will cite “established evidence” that AI tools empower women micro-entrepreneurs in LMICs.
The evidence cited is studies like the Bengaluru paper that don’t test such things. We have seen similar movements in mobile money. There, early evidence of basic SMS payments was used to justify the deployment of complex algorithmic lending products that produced very different outcomes.
3. Real discoveries are lost.
The real contribution of the Bangalore paper is that 35% of non-adopters cite fear and lack of confidence as the main barrier, rather than cost or infrastructure. This is a finding worth building on, as it tells donors that handing out smartphones or subsidizing data plans will not address binding constraints.
But the discoveries are buried under an “AI” framework, drawing the reader’s attention to the technology and away from the social infrastructure that records the paper in action.
Who is responsible?
If “AI” can stand for a digital ledger or a Facebook hashtag proposer, then the term has no meaning in development studies and we should stop using it.
But machine learning systems are now entering the fields of health decision support, credit scoring, and humanitarian aid, and we need a practical terminology to describe them and the specific risks they create.
But the responsibility to keep the term meaningful is distributed, and no one is resisting it at this point.
1. Researchers need to define terms.
Papers that call KhataBook an AI tool will need to justify their claim against a specific technical definition, or use a more precise label. “Digital tools,” “mobile business apps,” and “smartphone-based business support tools” are all available, and all are more accurate. Reviewers should return manuscripts with mixed categories.
2. Journaling should slow down.
A six-day peer review cycle is incompatible with capturing the shifting definitions of a rapidly changing technology. Editors at IJRISS and similar venues should require authors to specify model classes and technical mechanisms for tools that are described as AI. This is what SEC enforcement standards already require of regulated entities. Academic publishing should not be a soft target.
3. Donors must request specifications.
All grant proposals that use “AI” must include the model class, the source of training data, and the location of decision-making. If the proposed intervention turns out to be rule-based template messaging, it’s fine to fund it as such. Funding could be provided for fine-tuned language models for providing health advice to community health workers. But donors need to know which one they’re buying.
4. Implementers must resist marketing.
If a vendor offers an “AI-powered” version of a tool you already use, ask what exactly the AI is in that tool. If the answer is a pre-transformer recommendation heuristic, you can write your proposal more honestly and your evaluation will be more useful.
broader pattern
There is a huge gap between what women micro-entrepreneurs need and what donor programs can provide. The pattern is consistent. The department will reach for any technology currently available for funding, deliver it to people who need different types of sustained support, and create evidence that the technology has worked when it has worked.
The Bangalore newspaper is honest about this in its own research.
The strongest predictor of research confidence is training time, not tools. Women who received more than 5 hours of NGO support had confidence scores between 7 and 9. Women with less than 2 hours scored between 2 and 4. Technology is constant. The variable that explains the results is relationships.
As the paper states, AI tools are not the only ones empowering, humans are the ones empowering.
We’ve known it for 15 years. No new technology was needed to learn it. What we are doing now is recycling old discoveries into new funding cycles using new technological vocabularies, at the cost of losing the ability to distinguish between rebranded digital literacies and real algorithmic interventions.
If algorithmic interventions fail, and they will fail. Having burned both of our rebranding ledger apps, we no longer have the conceptual vocabulary or evidence base to explore them. The SEC showed us the way. Researchers, journals, donors, and implementers in the field of development should not wait for equivalent enforcement mechanisms to be established to advance the field.
