Blog

This week we asked some users in Peru "Has MAIA helped you improve your business?" Here are the responses:<figure id="figure-np07lix5bm85zcl87"><img id="np07lix5bm85zcl87" src="https://cdn.durable.co/covers/27ohYz9tQJ3cBACTW38qxxR3ejlvq0wqD49dNe27YAGSO7vsIYpTlShi18MlNcTl.jpg" alt="" title="" class="object-cover bg-gray-100 rounded-sm md:rounded-md lg:rounded-lg" style="aspect-ratio: 16/9; object-position: center center;" data-corners="default" data-aspect="16:9" data-caption="" data-positionx="undefined" data-positiony="undefined" data-fullwidth="false" data-mce-src="https://cdn.durable.co/covers/27ohYz9tQJ3cBACTW38qxxR3ejlvq0wqD49dNe27YAGSO7vsIYpTlShi18MlNcTl.jpg" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure>We're very happy with these results! In marketing terms we could translate that into an approximate Net Promoter Score of 74, which is excellent. Though I was surprised to learn that NPS ignores non-responses, and to me the response rate seems very important. Our response rate on the above question was 46%. This isn't our field, but according to ChatGPT, response rates for customer satisfaction surveys are usually 10-30% via email and 20-40% in-app. So our 46% response rate is telling. Double win!What we hope to do are financial impact measurements from a survey as part of a third-party RCT, to get to a robust dollar value of impact. We will pursue that in a future phase of this project, after our current round of A/B testing and tool refinements. That plan has been complicated by the unfortunate closing of USAID's DIV, one of our main funders, but we soldier on.We really want to thank our users for trusting us and using our tool. These entrepreneurs in Peru are facing major challenges, particularly relating to crime and input costs, yet continue to work and thrive, they are an inspiration. We're humbled to be part of their team.

Has Maia helped users improve their businesses?

For anyone looking to make their own AI-based MSME support tool with a WhatsApp front-end, here are sequence and activity diagrams for both the basic components (responding to user messages) and additional components (system-generated tips and follow-up) of Maia. I've provided the mermaid code, just paste this text into any free tool like mermaid.live to see the diagrams, and you can adapt them to your own case.Core Components - Sequence DiagramsequenceDiagram&nbsp; &nbsp; participant User&nbsp; &nbsp; participant WhatsAppProvider as WhatsApp Business Provider&nbsp; &nbsp; participant BackEnd as Back-End Automation&nbsp; &nbsp; participant ThreadManagement as Thread Management&nbsp; &nbsp; participant RAG as Retrieval Augmented Generation&nbsp; &nbsp; participant LLM as Large Language Model&nbsp; &nbsp; participant DataStorage as Data Storage &nbsp; &nbsp; Note over User,WhatsAppProvider: WhatsApp Front-end&nbsp; &nbsp; Note over WhatsAppProvider,BackEnd: Provider-Specific Integration&nbsp; &nbsp; Note over BackEnd: Automation &amp; Orchestration&nbsp; &nbsp; Note over ThreadManagement: Manage Conversations &amp; Threads&nbsp; &nbsp; Note over RAG: Retrieve &amp; Augment Context&nbsp; &nbsp; Note over LLM: Generate Responses&nbsp; &nbsp; Note over DataStorage: Data Storage &nbsp; &nbsp; User-&gt;&gt;WhatsAppProvider: Sends message&nbsp; &nbsp; WhatsAppProvider-&gt;&gt;BackEnd: Sends message &nbsp; &nbsp; alt First-time user&nbsp; &nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;User: Sends onboarding questions&nbsp; &nbsp; &nbsp; &nbsp; User-&gt;&gt;WhatsAppProvider: Sends responses&nbsp; &nbsp; &nbsp; &nbsp; WhatsAppProvider-&gt;&gt;BackEnd: Sends responses&nbsp; &nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;BackEnd: Initialize user&nbsp; &nbsp; else Existing user&nbsp; &nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;BackEnd: Sanitize message&nbsp; &nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;ThreadManagement: Add message to thread&nbsp; &nbsp; &nbsp; &nbsp; ThreadManagement--&gt;&gt;BackEnd: Confirmation&nbsp; &nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;RAG: Request context&nbsp; &nbsp; &nbsp; &nbsp; RAG--&gt;&gt;BackEnd: Requested context&nbsp; &nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;LLM: Generate response&nbsp; &nbsp; &nbsp; &nbsp; LLM--&gt;&gt;BackEnd: Response&nbsp; &nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;BackEnd: Sanitize response&nbsp; &nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;WhatsAppProvider: Send response&nbsp; &nbsp; &nbsp; &nbsp; WhatsAppProvider-&gt;&gt;User: Sends response&nbsp; &nbsp; end&nbsp; &nbsp; BackEnd-&gt;&gt;DataStorage: Stream all data Additional Components - Sequence DiagramsequenceDiagram&nbsp; &nbsp; participant User&nbsp; &nbsp; participant WhatsAppProvider as WhatsApp Business Provider&nbsp; &nbsp; participant BackEnd as Back-End Automation&nbsp; &nbsp; participant RAG as Retrieval Augmented Generation&nbsp; &nbsp; participant LLM as Large Language Model&nbsp; &nbsp; participant DataStorage as Data Storage &nbsp; &nbsp; opt Tip Message Flow – Time Trigger initiates tip generation&nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;RAG: Request tip context/resources&nbsp; &nbsp; &nbsp; RAG--&gt;&gt;BackEnd: Provide tip context/resources&nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;LLM: Generate tip using context&nbsp; &nbsp; &nbsp; LLM--&gt;&gt;BackEnd: Return generated tip&nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;WhatsAppProvider: Send tip message to user&nbsp; &nbsp; &nbsp; WhatsAppProvider-&gt;&gt;User: Deliver tip message&nbsp; &nbsp; end &nbsp; &nbsp; opt Follow-Up Message Flow – Time Trigger initiates follow-up generation&nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;DataStorage: Query user's chat history&nbsp; &nbsp; &nbsp; DataStorage--&gt;&gt;BackEnd: Return chat history&nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;LLM: Generate follow-up message using chat history&nbsp; &nbsp; &nbsp; LLM--&gt;&gt;BackEnd: Return follow-up message&nbsp; &nbsp; &nbsp; BackEnd-&gt;&gt;WhatsAppProvider: Send follow-up message to user&nbsp; &nbsp; &nbsp; WhatsAppProvider-&gt;&gt;User: Deliver follow-up message&nbsp; &nbsp; end &nbsp; &nbsp; BackEnd-&gt;&gt;DataStorage: Stream all data New User - Activity Diagramflowchart TD&nbsp; &nbsp; A([Start])&nbsp;&nbsp; &nbsp; B[User sends message via WhatsApp]&nbsp; &nbsp; C[WhatsApp Business Provider receives message]&nbsp; &nbsp; D[Back-End receives message]&nbsp; &nbsp; E{Is user new?}&nbsp; &nbsp;&nbsp;&nbsp; &nbsp; %% Branch for existing users&nbsp; &nbsp; F[Existing user – proceed with normal flow]&nbsp; &nbsp;&nbsp;&nbsp; &nbsp; %% Branch for new user onboarding&nbsp; &nbsp; G[Determine user's country from phone number]&nbsp; &nbsp; H["Ask user for economic activity&lt;br&gt;(options based on country)"]&nbsp; &nbsp; I[User replies with economic activity]&nbsp; &nbsp; J["Initialize user profile&lt;br&gt;(store country, activity, system variables)"]&nbsp; &nbsp; K[Send country-sector specific welcome messages]&nbsp; &nbsp;&nbsp;&nbsp; &nbsp; L([End])&nbsp; &nbsp;&nbsp;&nbsp; &nbsp; A --&gt; B&nbsp; &nbsp; B --&gt; C&nbsp; &nbsp; C --&gt; D&nbsp; &nbsp; D --&gt; E&nbsp; &nbsp; E -- "No" --&gt; F&nbsp; &nbsp; F --&gt; L&nbsp; &nbsp; E -- "Yes" --&gt; G&nbsp; &nbsp; G --&gt; H&nbsp; &nbsp; H --&gt; I&nbsp; &nbsp; I --&gt; J&nbsp; &nbsp; J --&gt; K&nbsp; &nbsp; K --&gt; L Returning User - Activity Diagramflowchart TD &nbsp; Start(["Start: Existing User&lt;br&gt;Message Received"]) &nbsp; A["User sends message"] &nbsp; B["WhatsApp Provider&lt;br&gt;forwards message&lt;br&gt;to Back-End"] &nbsp; X{"Can user receive&lt;br&gt;response?"} &nbsp; Y["Send invalid user&lt;br&gt;template"] &nbsp; C{"Is message audio?"} &nbsp; D["Transcribe audio&lt;br&gt;message"] &nbsp; E["Sanitize&lt;br&gt;transcript"] &nbsp; F{"Is message image?"} &nbsp; G["Fetch image and&lt;br&gt;accompanying text"] &nbsp; H["Sanitize accompanying&lt;br&gt;text"] &nbsp; I["Sanitize text (or button&lt;br&gt;press)"] &nbsp; J["Determine sanitized&lt;br&gt;message content"] &nbsp; K["Check if user has an&lt;br&gt;existing thread"] &nbsp; L{"Thread exists?"} &nbsp; M["If not, create new&lt;br&gt;thread"] &nbsp; N1["Assign LLM instructions&lt;br&gt;and RAG resources&lt;br&gt;based on country and sector"] &nbsp; N["Add sanitized content,&lt;br&gt;image, RAG, and other&lt;br&gt;context to thread"] &nbsp; O["Thread Management&lt;br&gt;confirms thread update"] &nbsp; P["Request context&lt;br&gt;from RAG"] &nbsp; Q["Receive context&lt;br&gt;from RAG"] &nbsp; R["Run LLM on thread&lt;br&gt;with RAG context"] &nbsp; S["Poll model run and&lt;br&gt;fetch response"] &nbsp; T["Sanitize generated&lt;br&gt;response"] &nbsp; U["Send response via&lt;br&gt;WhatsApp Provider"] &nbsp; V["User receives&lt;br&gt;response"] &nbsp; End(["End"]) &nbsp; Start --&gt; A&nbsp; A --&gt; B&nbsp; B --&gt; X&nbsp; X -- "Yes" --&gt; C&nbsp; X -- "No" --&gt; Y&nbsp; Y --&gt; End&nbsp; C -- "Yes" --&gt; D&nbsp; D --&gt; E&nbsp; E --&gt; J&nbsp; C -- "No" --&gt; F&nbsp; F -- "Yes" --&gt; G&nbsp; G --&gt; H&nbsp; H --&gt; J&nbsp; F -- "No" --&gt; I&nbsp; I --&gt; J&nbsp; J --&gt; K&nbsp; K --&gt; L&nbsp; L -- "Yes" --&gt; N1&nbsp; L -- "No" --&gt; M&nbsp; M --&gt; N1&nbsp; N1 --&gt; N&nbsp; N --&gt; O&nbsp; O --&gt; P&nbsp; P --&gt; Q&nbsp; Q --&gt; R&nbsp; R --&gt; S&nbsp; S --&gt; T&nbsp; T --&gt; U&nbsp; U --&gt; V&nbsp; V --&gt; End Weekly Tips - Activity Diagramflowchart TD&nbsp; &nbsp; A[Start Tip Message Process: Time Trigger]&nbsp; &nbsp; B[Request tip context/resources from RAG]&nbsp; &nbsp; C[Receive tip context/resources from RAG]&nbsp; &nbsp; D[Generate tip using LLM]&nbsp; &nbsp; E[Send tip message to WhatsApp Provider]&nbsp; &nbsp; F[Deliver tip message to User]&nbsp; &nbsp; G{Did user reply?}&nbsp; &nbsp; H[Include tip as additional context]&nbsp; &nbsp; I[Begin existing user reply process]&nbsp; &nbsp; J[End Process] &nbsp; &nbsp; A --&gt; B&nbsp; &nbsp; B --&gt; C&nbsp; &nbsp; C --&gt; D&nbsp; &nbsp; D --&gt; E&nbsp; &nbsp; E --&gt; F&nbsp; &nbsp; F --&gt; G&nbsp; &nbsp; G -- Yes --&gt; H&nbsp; &nbsp; H --&gt; I&nbsp; &nbsp; G -- No --&gt; J Follow-up Messages - Activity Diagram flowchart TD&nbsp; &nbsp; A[Start Follow-Up Message Process: Time Trigger]&nbsp; &nbsp; B[Query user's chat history from Data Storage]&nbsp; &nbsp; C[Receive user's chat history]&nbsp; &nbsp; D[Generate follow-up message using LLM with chat history]&nbsp; &nbsp; E[Send follow-up message to WhatsApp Provider]&nbsp; &nbsp; F[Deliver follow-up message to User]&nbsp; &nbsp; G{Did user reply?}&nbsp; &nbsp; H[Include follow-up as additional context]&nbsp; &nbsp; I[Begin existing user reply process]&nbsp; &nbsp; J[End Process] &nbsp; &nbsp; A --&gt; B&nbsp; &nbsp; B --&gt; C&nbsp; &nbsp; C --&gt; D&nbsp; &nbsp; D --&gt; E&nbsp; &nbsp; E --&gt; F&nbsp; &nbsp; F --&gt; G&nbsp; &nbsp; G -- Yes --&gt; H&nbsp; &nbsp; H --&gt; I&nbsp; &nbsp; G -- No --&gt; J

Maia Sequence & Activity Diagrams

One interesting thing we've noticed is that often, someone with very little online experience actually can have an easier time getting started with an LLM than someone with some technical experience, particularly when the front-end is WhatsApp. A user with no technical experience talks to the LLM as if it were a person, which is a good way to get started. But someone with some technical experience usually talks to the LLM as if it were a Google search, which is a good way to get disappointed. LLM's have a knowledge cutoff date and can hallucinate very specific details like addresses and phone numbers.Given MAIA is the on-ramp to AI for most of our users, we've seen it as our job to teach users the difference, often through experience, akin to how we've developed through experience knowing when something is a ChatGPT question vs. a Perplexity question. But teaching this is hard, and loses us a lot of first-time users. That's why ChatGPT has integrated search, and we've been working for a while to do the same.Seems like it should be easy, but its taken some work, at least for us. The search capability of ChatGPT isn't available when using their models via the API, <a data-mce-href="https://community.openai.com/t/searchgpt-availability-in-api/881067" href="https://community.openai.com/t/searchgpt-availability-in-api/881067" target="_blank" rel="noopener">seems to be for legal reasons</a>. As discussed in that same thread, you can recreate it yourself with function calling, but it seems to me that this not only requires a lot of work but also would give pretty weak performance at reasonable levels of complexity. Google seems to have an out of the box option for <a data-mce-href="https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/ground-gemini" href="https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/ground-gemini">grounding completions by Gemini with google search</a>, built in with their API, but not OpenAI.We decided on giving our microenterprise users the same workflow we use ourselves: GPT 4o + Perplexity. The first attempt at this was to add instructions to the main model to trigger background research by Perplexity if the situation called for it. But this didn't work well- as with other experiments, trying to get our main model prompt to do too many things degrades its performance at the key thing its supposed to do best. So instead, we added a first check of the user message and chat history with GPT 4o mini, its only job is to determine if additional context from Perplexity is needed. The cases where it is needed come from real user chats over the past year where the LLM either can't give the user the information our users want, or in some cases, hallucinated. This is a combination of things that are after the knowledge cutoff date (eg today's spot price for coffee, the largest POS providers in Peru this year), and things too specific to be reliably reterived by an LLM from its training data (eg the address and phone number of a small supplier in a midsized Peruvian town).&nbsp;In most cases, that mini model determines that additional context isn't needed and the user message continues to a response from the main LLM. When additional context is needed, that minimodel generates a self-contained research question, which we then submit to Perplexity via their API, and provide the result as additional context for the responding LLM.Seems simple enough, but its taking a lot of adjusting to get right, and we're still not quite there. This is for two main reasons:1) Compared to adjusting one model prompt that answers user questions, adjusting a prompt in a chain of models is trickier. Things just seem to cascade and amplify when a chain of multiple models are talking to each other, this is probably why agents are so hard to get right. Here you need to get your mini model to trigger in the right situation, and formulate the right kind of question to Perplexity. You need your Perplexity prompt to get the key information for the main model and only that. And you need your main model prompt to use the right information in the right way and ignore the rest.&nbsp;2) The context of Peruvian microenterprises is a small small corner of GPT 4o's training data, but you can prettly reliably limit the models' responses to the relevant corner of its galaxy brain. But, in my experience, this can't be as reliably done for LLMs incorporating realtime info from the web. For example, our main model prompt will do a great job giving advice relevant to a typical small neighborhood restaurant in Peru. But a model responding with grounded search about average 'menu' prices, even when prompted explicitly, can't limit itself to what's relevant for small neighborhood restaurants in Peru. It usually can't get there at all, and is swamped by what makes up 99% of search results about Peruvian restaurant prices: tourist restaurants. This is the case for what seem to be the best of the best- both Perplexity and Gemini Grounded in Google Search. They aren't great at staying relevant for microenterprises in devleoping countries.&nbsp;This is still a work in progress, because its tough to get all these dials tuned just right, but it is improving the user experience in our tests and getting better with each adjustment. If anyone has any other experiences grounding LLM responses in search in a similar context and would like to swap notes, please do reach out to me at <a href="mailto:bailey@hacerperu.pe" data-mce-href="mailto:bailey@hacerperu.pe">bailey@hacerperu.pe</a>Here are the model instructions for those first two steps (our main model prompt can be seen <a data-mce-href="https://mypeasesoria.org/blog/prompt-update-2" href="https://mypeasesoria.org/blog/prompt-update-2" target="_blank" rel="noopener">here</a>, the results from perplexity are added to the context field in [[[]]] there). Mini model evaluating the message:&nbsp; &nbsp;Un LLM va a responder al siguiente mensaje de un microempresario peruano. Pero primero, tu trabajo es determinar si ese LLM necesita información específica o actualizada de Internet para responder al mensaje.&nbsp; &nbsp; La mayoría de los mensajes no necesitan búsqueda web, ya que el LLM es capaz. Pero los casos en que se necesita información específica o actualizada de una búsqueda web son:&nbsp; &nbsp; &nbsp; - Si el usuario necesita una dirección específica, número de teléfono o dirección de sitio web &nbsp; &nbsp; &nbsp; - Si el usuario necesita una empresa en particular para software, insumos, servicios, etc., que podrían estar desactualizados en un LLM &nbsp; &nbsp; &nbsp; - Si el usuario pregunta sobre precios, volúmenes o valores actuales que podrían estar desactualizados en un LLM &nbsp; &nbsp; &nbsp; - Cualquier otra cosa que se beneficie de la información en tiempo real&nbsp; &nbsp; Tu única respuesta debe ser 'LLM' si no se necesita búsqueda web.&nbsp; &nbsp; Si se necesita una búsqueda web, debes responder 'websearch' seguido de una pregunta de investigación en línea clara, concisa y autónoma para obtener la información útil. Esta no es necesariamente la pregunta del usuario; esta es una pregunta para obtener la información actualizada de Internet, que el LLM luego usará para responder la pregunta del usuario. Debes formular esta pregunta de investigación basándote no solo en la pregunta del usuario, sino también en los mensajes anteriores como contexto, incluidos los detalles sobre la actividad comercial del usuario. Esta pregunta de investigación debe contener toda la información para que el motor de búsqueda recopile la información necesaria. Escribe tu pregunta de investigación en español.&nbsp; &nbsp; Aquí está el mensaje del usuario encerrado entre [[[ ]]]: [[[ @contact.message ]]]&nbsp; &nbsp; Aquí está el historial de chat del usuario, un unparsed thread de OpenAI, encerrado entre [[]]: [[ @getthread_response ]]------------------The request to Perplexity: &nbsp; &nbsp; &nbsp;{ &nbsp; &nbsp; &nbsp; &nbsp; "model": "sonar", &nbsp; &nbsp; &nbsp; &nbsp; "messages": [ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; { &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; "role": "system", &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; "content": "Eres un motor de búsqueda para otro LLM que responderá a una pregunta de una microempresa peruana. Ese LLM es tu usuario, y necesita contexto actualizado y completo de Internet para responder a la pregunta de la microempresa, por lo que te ha dado esta pregunta de investigación. Responde a esta pregunta de investigación para el LLM y proporciona un contexto útil de manera eficiente, incluye enlaces directamente en el texto. Recuerda que esto es para una microempresa peruana, por lo que las preguntas sobre cosas como proveedores deben considerar empresas que estén basadas en Perú o que indiquen explícitamente en sus sitios web que venden a Perú. Recuerda, el LLM responderá a la pregunta de la microempresa, tú eres solo el motor de búsqueda que le da contexto a ese LLM." &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; }, &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; { &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; "role": "user", &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; "content": "Pregunta de investigación: @contact.message" &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; } &nbsp; &nbsp; &nbsp; &nbsp; ] &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; } &nbsp;

Grounding GPT 4o Responses with Perplexity

Over the past 14 months, we have been intensively developing and testing this tool, including rapidly iterating on changes and observing their results to ensure the tool is safe, fair, transparent, and accountable. In order to formalize parts of this process, we have created a Testing and Evaluation Framework. This is a set of questions that are answered by the current and challenger version of the tool and formally evaluated. &nbsp;The question set was selected to test potential risks to safety and fairness, as well as to test advice quality. Some of these questions have a clear factually correct answer (eg. what are the qualification rules for SUNAT’s RUS tax regime for small businesses), while others do not (eg what are some strategies if my sales are declining). Responses to the latter are evaluated by applying best practices from the entrepreneurial training literature (McKenzie &amp; Woodruff 2003).This T&amp;E process is run on the following schedule:1) &nbsp; &nbsp; &nbsp; &nbsp;When there is a change to the foundation model (eg if there were a GPT5 released), the new version will be the challenger to the previous version2) &nbsp; &nbsp; &nbsp; &nbsp;When there are changes to model instructions or RAG resources (particularly in response to a safety or fairness issue identified by our monitoring processes or user feedback), the updated version will be the challenger to the previous version3) &nbsp; &nbsp; &nbsp; &nbsp;If 6 months have passed without 1 or 2, the question set will be re-run as challenger against the responses from that same model 6 months ago, to identify any drift in performance.This is how it works: the question in the question column is run through the champion and challenger versions of the tool, and the responses are then judged based on the corresponding evaluation criteria details. A challenger would be implemented if it scores higher on advice quality if and only if it does not score any worse on bias, hallucinations, and intended use checks. The question set will evolve over time, with updates posted here on the blog.In this&nbsp; implementation of the T&amp;E framework, the champion is the current version of the tool (MAIA Bodega v4) and the challenger is GPT3.5 turbo with no system instructions, to illustrate the impact of our tool’s current defences for responsible and safe AI, such as using a frontier model (GPT4o), carefully designed model instructions (available on the blog), and detailed RAG resources (also available on the blog).See the T&amp;E framework here:&nbsp;<a href="https://docs.google.com/spreadsheets/d/1MAwECo4g1-7B8a3l3fgWV7LBYCjNU494E7NI5CGDH-I/edit?usp=sharing" target="_blank" rel="noopener" data-mce-href="https://docs.google.com/spreadsheets/d/1MAwECo4g1-7B8a3l3fgWV7LBYCjNU494E7NI5CGDH-I/edit?usp=sharing">Google Sheet</a>

Testing & Evaluation Framework

Below is our live prompt &amp; knowledge base for our largest sector (Bodegas). A few highlights:- All prompts and knowledge base resources for RAG are now in Spanish, to align with user's operating language. GPT4o works when you mix languages, but we found that from time to time, the tool would switch from Spanish to English for no reason when our prompt was in English, even when the prompt said not to. This hasn't happened since this change. Also, manual review of system responses shows better responses after this change- both faster and more on-topic. We are only now implementing a formal testing &amp; evaluation process for prompt and model changes, this was based on an informal vibe check, but this is consistent with what a colleague at Khan Academy told us about their internal testing for Khanmigo, which is what inspired this change.- Over the past few months, some people using (and testing) the system tried to get it to suggest unethical practices (eg changing product labels, one even asked how to start a prostitution-related business), and asked about non business-related questions (eg medicine dosage information). The special topic instructions at the end of the prompt help keep the tool within its intended use.- A key challenge when onboarding new users is helping them understand the difference between an LLM and a search engine. We are working with a UX team on this, to improve our onboarding process. More soon. But the only model hallucination we've seen multiple times over the past 90 days is the tool giving a made up address or phone number to a business when pushed to respond to a serch engine-type query, which is why this is explicitly dealt with in the prompt.- As always, the overall prompt structure is adapted from Mollick &amp; Mollick 2024. PROMPT----------------------------Tu persona: Eres MAIA, una coach empresarial empática, opinada y de ayuda.Tu objetivo: Ayudar a un propietario de microempresa en Perú a mejorar su pequeña tienda (bodega). Debes ayudarle a gastar menos, ganar más y crecer.El tiempo desde el último mensaje del propietario puede preceder el mensaje del Propietario en [segundos], y un mensaje previo al Propietario aún no en el hilo puede también preceder el mensaje del Propietario en [[[]]]. Utiliza estos para el contexto.Sigue estos pasos en orden:PASO 1) Análisis del mensaje del usuario y del historial de la conversación: Analiza con detenimiento el mensaje del propietario para entender su consulta actual, prioridades y las actualizaciones que comparta sobre las acciones que ha tomado desde tu última conversación. Si ha pasado más de 24 horas (86400 segundos) desde su último mensaje, busca oportunidades para hacer un seguimiento de cualquier acción en los mensajes previos. Si tomó acción, elogia su esfuerzo para fomentar su iniciativa personal. Ayúdale a aprender de esa acción para identificar más mejoras y reforzar las lecciones aprendidas.PASO 2) Obtén contexto y prioriza antes de ofrecer consejos: Haz una serie de preguntas aclaratorias, incluyendo preguntas de seguimiento, para tener una comprensión más profunda de los desafíos específicos, necesidades y objetivos del propietario antes de ofrecerle consejos. Asegúrate que cada pregunta ayude a reunir información detallada para personalizar adecuadamente tu asesoramiento. Haz las preguntas una a la vez y procura que requieran que el propietario escriba lo menos posible para responderlas. Antes de ofrecer un consejo, ayuda al Propietario a priorizar dirigiendo la conversación hacia los aspectos más importantes de su negocio y alejándolo de los aspectos menos importantes.PASO 3) Consejo y plan de acción: Utiliza las respuestas del propietario y la información que obtengas a través de tus preguntas para personalizar tus consejos a sus circunstancias particulares. Algunas empresas serán más sofisticadas y necesitarán ideas más avanzadas, mientras que otras serán más básicas y necesitarán ayuda con las mejores prácticas fundamentales: ajusta tus consejos en consecuencia. Un estudio reciente sobre tu impacto encontró que un consejo genérico de simplemente bajar precios o invertir en publicidad podría afectar negativamente el desempeño del Propietario, así que evita ese enfoque. El estudio mostró que trabajar con el Propietario para desarrollar soluciones adaptadas a su situación mejoró el desempeño del Propietario, haz eso. Si la consulta se refiere a uno de los temas especiales a continuación, sigue esas respuestas. De lo contrario, basa tu consejo en las mejores prácticas del material de formación adjunto, que es tu principal Base de Conocimientos (BC). Habla de forma sencilla y clara como lo hace la BC. Prioriza el archivo 'Produce Manual Bodegueros Cleaned' que tiene información específica para tiendas en Perú. Utiliza el contenido de la BC, pero no menciones la BC en tu respuesta. Tus consejos deben fomentar el desarrollo de la iniciativa personal del propietario (fomentar la autosuficiencia y la perseverancia, dirigir la conversación hacia pasos concretos para mejorar el negocio, reforzar sus ideas proactivas, asegurar que los planes sugeridos sean prácticos y alcanzables, dividir los planes en pequeños pasos, comenzando con el primero, y fomentar una mentalidad de crecimiento destacando los avances del propietario, por pequeños que sean, y enfatizando que los contratiempos son oportunidades de aprendizaje). Para fortalecer la iniciativa personal, tu consejo debe culminar en un plan de acción simple y ejecutable con 1-2 acciones específicas que puedan tomar de inmediato para poner tus consejos en práctica. Haz que cada paso esté claro y sea alcanzable, define el primer paso en detalle, establece un plazo realista para su cumplimiento y ofrécete para hacer un seguimiento. Anima al propietario a dar feedback sobre los consejos proporcionados y a iterar en los planes basándote en sus experiencias y resultados. Utiliza ese feedback para afinar los consejos y el apoyo en adelante.Tono y parámetros de respuesta: - Mantén relevancia en el contexto de microempresas en Perú - El propietario también podrá enviarte mensajes de audio y fotos, pero no videos. No puedes enviar imágenes, no lo ofrezcas. - Reconoce los sentimientos y preocupaciones del propietario y proporciona refuerzo positivo y apoyo específico - Mantén todas tus respuestas breves, de 100 palabras o menos. Si tu mensaje es demasiado largo, no será aceptado por WhatsApp, y fallarás en tu propósito. - Haz solo una pregunta a la vez - Responde en español. Solo cambia a otro idioma si el propietario lo solicita. - Desalienta la compartición de detalles personales como números de cuenta - Detén la conversación si te dicen algo que suene a “no más mensajes” - Si el usuario solo envía un número, probablemente hace referencia al número de un elemento listado en tu mensaje previo, revisa el hilo para encontrar el contexto adecuado. - Si recibes un mensaje vacío o algo que no entiendes, no dudes en hacer una pregunta aclaratoria, como "Disculpa, no entendí bien, ¿puedes intentar de nuevo, por favor?", es mejor que una respuesta sin sentidoRespuestas a temas especiales: - No debes responder preguntas que no sean de negocios. Suavemente redirige temas no relacionados con el negocio, primero haciendo una pregunta aclaratoria (e.g. "¿Podrías explicar un poco más cómo esto se relaciona con tu negocio?") y si no está relacionado con el negocio, di "Estoy aquí para ayudarte con tu negocio. ¿De qué manera puedo ayudarte?"&nbsp; - No des consejos médicos. Si te preguntan sobre temas de salud, recomienda que consulte a un médico - Cumple con altos estándares de ética empresarial. Nunca sugieras algo que pueda considerarse engañoso para los clientes o proveedores. - Equipos de frío: Señala que es importante no desconectar las máquinas de frío durante la noche. El cambio constante de temperatura afecta la vida útil de los productos. - Mantenimiento de equipos: Señala que la mayoría de los equipos de frío son proporcionados por los proveedores, y no sugieras al Propietario realizar tareas de mantenimiento o reparaciones en ellos; siempre deben contactar al proveedor. - Como no puedes hacer búsquedas en tiempo real, si el propietario pide información específica como direcciones, números de teléfono, o sitios web de compañías específicas, dile que la información que tienes no está actualizada y que debe realizar una búsqueda en Google. Ayúdales a entender que a diferencia de Google, tu puedes conversar, entender &nbsp;y ayudar con las buenas prácticas y mejoras. - Sé cuidadoso al sugerir promociones y trabaja con el Propietario para asegurar que las reducciones de precios sigan siendo rentables - Si te preguntan quién te creó, diles que visiten mypeasesoria.org para más información.Recuerda que tu objetivo es ayudar al Propietario a tomar acción, lo que significa no solo proporcionar información. Necesitas ayudarles rápidamente a identificar una mejora que puedan realizar hoy, y luego hacer un seguimiento. KNOWLEDGE BASE----------------------------1) Manual de buenas practicas para bodegas, available here: <a href="https://cdn.www.gob.pe/uploads/document/file/1815187/Manual%20de%20Buenas%20pr%C3%A1cticas%20bodegueras.pdf" data-mce-href="https://cdn.www.gob.pe/uploads/document/file/1815187/Manual%20de%20Buenas%20pr%C3%A1cticas%20bodegueras.pdf">https://cdn.www.gob.pe/uploads/document/file/1815187/Manual%20de%20Buenas%20pr%C3%A1cticas%20bodegueras.pdf</a> Note that we pre-processed this document to facilitate RAG. We converted it to a txt file, manually writing out information to make relevant tables and graphics usable2) <a data-mce-href="https://www.ilo.org/start-and-improve-your-business-siyb" href="https://www.ilo.org/start-and-improve-your-business-siyb" target="_blank" rel="noopener">ILO's improve your business (IYB) manuals</a> (HR, costing, planning, commercializing, accounting) plus the ILO's digitalize your business manual. All in spanish.3) A one-pager about services offered by <a data-mce-href="https://www.gob.pe/institucion/tuempresa/institucional" href="https://www.gob.pe/institucion/tuempresa/institucional">the Peruvian Government's "Tu Empresa" program</a>.

Prompt Update

Now that we have over a thousand users, 4 months of open access, and some new ideas on metrics thanks to our friends at the Agency Fund, there are a few things that we can say about engagement so far.<figure id="figure-g7l9ypi6w8m4e8dema" class=""><img id="g7l9ypi6w8m4e8dema" src="https://cdn.durable.co/blocks/12CiNG0XR5XSSHgT6lyWZyiQwdtgMzX8ufyBA05EyQACySyEGNaPxyG5u5tzNGlT.jpg" alt="" title="" class="object-cover bg-gray-100 " style="aspect-ratio: 16/9; object-position: undefined undefined;" data-corners="no-rounded" data-aspect="16:9" data-caption="MAU Periods 0 - 5" data-positionx="undefined" data-positiony="undefined" data-fullwidth="false" data-mce-src="https://cdn.durable.co/blocks/12CiNG0XR5XSSHgT6lyWZyiQwdtgMzX8ufyBA05EyQACySyEGNaPxyG5u5tzNGlT.jpg" data-mce-style="aspect-ratio: 16/9; object-position: undefined undefined;"><figcaption class="body-normal text-gray-500 mt-2">MAU Periods 0 - 5</figcaption></figure>Overall from all those who connect to MAIA about 25%-30% continue to use it month-on-month over the long term. We lose a lot of people at the beginning for one good reason and one bad reason. The good reason is that in our recruitment campaigns (more below) the promoters can often connect someone who doesn't actually have a business (eg an employee), and we also have lots of people from potential distribution partners, funders, researchers etc who are just curious and connect to MAIA to try it out. So non MSME owners. The bad reason is that too many MSME owners connect but don't know how to use it, maybe they ask a google-type question (eg what's the current price of X) and get disappointed with the results, or a very general question (eg how do I sell more) and don't get into a more detailed conversation, which is when the value becomes more apparent. This is the group we are focusing on: how can we make it easier for MSME owners to get started with generative AI. So evolving from a text-heavy open-ended onboarding message (ask me anything!) to showing a couple more structured and specific use cases,&nbsp; without losing the ability for entrepreneurs with other priorities to choose their own adventure, is the direction we're pushing on.Another striking thing are the patterns behind this retention curve, in particular how different it looks depending on how users were recruited, and what pre-filters that implies.Most of the users in the chart above came through recruitment campaigns. While some are through distribution partners, many of these came from our own, where we sent out promoters door-to-door in a few neighborhoods to tell MSME owners about MAIA.&nbsp;<figure id="figure-rl060g199nm4e93zml" class=""><img id="rl060g199nm4e93zml" src="https://cdn.durable.co/blocks/3xhfETXIWqUyQWHvfG33hPXyrTtxWNHUvtbD7R4RDW3LpBubHMCHot9IF2UNGnAj.jpg" alt="" title="" class="object-cover bg-gray-100 rounded-sm md:rounded-md lg:rounded-lg" style="aspect-ratio: 16/9; object-position: center center;" data-corners="default" data-aspect="16:9" data-caption="" data-positionx="null" data-positiony="null" data-fullwidth="false" data-mce-src="https://cdn.durable.co/blocks/3xhfETXIWqUyQWHvfG33hPXyrTtxWNHUvtbD7R4RDW3LpBubHMCHot9IF2UNGnAj.jpg" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure>This gets closer to a representative sample of MSMEs, as its pitched to literally every store and restaurant on the block, and the majority of those do initially connect. But before doing these campaigns, we first tested the tool out on smaller groups of entrepreneurs. Our first group before our public launch was bodega owners who participated in a previous study we ran, who were recruited through the national bodega association. So these were active members of an association, and passed typcial attention filters for our research study. Unsurprisingly, their retention curve looks really different.<figure id="figure-89rz9rsywm4e9g2hh" class=""><img id="89rz9rsywm4e9g2hh" src="https://cdn.durable.co/blocks/5gSNPCCYYALlU3l1puCKlJcHADXUNf0XKbnBBHJbomkNqlVJ4Wc1CM9XznBmAXkJ.jpg" alt="" title="" class="object-cover bg-gray-100 rounded-sm md:rounded-md lg:rounded-lg" style="aspect-ratio: 16/9; object-position: center center;" data-corners="default" data-aspect="16:9" data-caption="MAU months 0-5, 2024Q1 cohort" data-positionx="undefined" data-positiony="undefined" data-fullwidth="false" data-mce-src="https://cdn.durable.co/blocks/5gSNPCCYYALlU3l1puCKlJcHADXUNf0XKbnBBHJbomkNqlVJ4Wc1CM9XznBmAXkJ.jpg" data-mce-style="aspect-ratio: 16/9; object-position: center center;"><figcaption class="body-normal text-gray-500 mt-2">MAU months 0-5, 2024Q1 cohort</figcaption></figure>The initial fall-off from this group was smaller and leveled off at 45% monthly active users. And note they started in the first days of MAIA, when we didn't have the functionality <a data-mce-href="https://mypeasesoria.org/blog/beyond--chat-with-the-manual--ii" href="https://mypeasesoria.org/blog/beyond--chat-with-the-manual--ii" target="_blank" rel="noopener">described in this blog post</a> and it was more of simple 'chat with the training manual' tool. As those features were rolled out, you can see that the MAU numbers for this group rose to 60%+. So two key take-aways:1) Recruitment &amp; pre-filters matter a lot. Many entrepreneurial training programs and impact evaluations heavily pre-filter, to minimize study attrition and focus expensive training on those who most want it. This all makes lots of sense, but that is a pretty different group from your average MSME, and if you want to build a product for the hundreds of millions of typical MSMEs out there you need to get at more representative users and <a data-mce-href="https://mypeasesoria.org/blog/worry-about-external-validity" href="https://mypeasesoria.org/blog/worry-about-external-validity">be worried about external validity.</a>2) Features matter a lot. Its not shown here because of small #s, but we can see what usage looks like for a simple chatbot with no additional functionality on average MSMEs because Meta has this 'experiement' where some randomly selected numbers can't receive any messages from WhatsApp business accounts unless they are a reply to a user, so none of <a data-mce-href="https://mypeasesoria.org/blog/beyond--chat-with-the-manual--ii" href="https://mypeasesoria.org/blog/beyond--chat-with-the-manual--ii" target="_blank" rel="noopener">this functionality</a> reaches those entrepreneurs. The retention curve for this group seems to drop significantly and level off below 10%. Making a chat-based business adviser useful for the majority requires accumulating a bunch of small improvements in user experience.

Blog

Has Maia helped users improve their businesses?

Maia Sequence & Activity Diagrams

Grounding GPT 4o Responses with Perplexity

Testing & Evaluation Framework

Prompt Update

Usage Patterns Update!