Blog

Part 3 in our series on MAIA's architecture---With AI tools becoming more accessible, more organizations are exploring AI-powered services like coaching bots, advisory tools, training apps. The technology holds a lot of promise. But what does it actually take to go from "we'll use AI" to a service that works reliably for real users?Based on our experience building MAIA across five countries for the past 2.5 years, here are the core capabilities that a well-functioning AI coaching service needs, and a rough sense of how much of the work each one represents (based on lines of code, which is an interesting but imperfect measure).1. Knowing What to Say — and How to Say It (~26%)The largest piece of the puzzle is instructing the model. To be useful across a variety of use cases and contexts, you actually need a library of prompts, not just one, because a coaching service needs to behave differently depending on who it's talking to, what stage they're at, and what they need. A first-time user in Panama exploring formal registration needs different guidance than a returning user in Kenya refining their pricing strategy. This includes decision logic that routes a user's question to the right capability and prompt, if your chatbot will do more than just generic 'talk do your documents' advice. Should this trigger a web search? Generate a marketing image generation? Offer a diagnostic? Simply answer? This library is never "done." It evolves constantly as you learn what works.2. Reliability and Delivery (~22%)The gap between a demo and a product is reliability. Messages need to arrive once and only onc, even when services go down. When an upstream provider has an outage, the system needs to gracefully switch to another provider rather than go silent. When a user sends a voice note, it needs to be transcribed and processed. When two messages arrive at the same instant, they can't corrupt each other's state, and when users multi-message (common in WhatsApp) you need your tool to behave like a human and combine them, rather than behave like a chatbot and respond one at a atime. This is the plumbing. No one sees it, but you definitely notice when its not right, and it significantly impacts usage.3. Proactive Coaching (~19%)A service that only responds when spoken to isn't really coaching. Nearly a fifth of MAIA's codebase is dedicated to reaching out: personalized weekly outreach, smart follow-ups that review the recent conversation to decide whether a check-in is useful, engagement logic that adjusts frequency based on how active each user is, structured multi-week sprints, and onboarding sequences tailored to each country program. This entire subsystem exists independently of the regular response flow, and has to be built around the unique economics of WhatsApp.4. Memory and Understanding (~11%)AI models don't inherently remember users between conversations. Building persistent memory means extracting structured information from each interaction, particularly when chats span months and years as some of ours do. This includes things like what the user's business does, their financial situation, their marketing channels, their goals, and maintaining an evolving profile that gets better over time. Unlike generic chatbots which can struggle with coherent focused relevant memory, MAIA's business focus allows for a much smarter and more targeted memory. Moreover, MAIA goes beyond recall to a structured diagnostic that assesses users across seven business pillars (like sales, finance, operations, and digital tools), so that advice is calibrated to where someone actually is and what is their key next priority.5. Context Assembly (~7%)Even with a good model and good instructions, the quality of an AI response depends on what information it can see. Before every interaction, the system assembles a context window: the user's business profile, their diagnostic results, recent conversation history, any active coaching program, and relevant knowledge base references. Think of it like briefing a human coach before each session. The better the briefing, the better the session.6. Conversation Handling (~10%)Real users don't behave like demos. They send three messages in a row before the bot can respond. They reply to a messages out of order, from last week. They forward something from another chat. They go quiet for two months and come back. Handling all of this gracefully, like batching rapid messages, transcribing audio, threading quoted replies, managing session state, preventing the bot from talking to itself in edge cases, is another significant layer of work.7. The AI Model Call (~7%)Here is the part that most people picture when they think "AI service": sending a request to a language model and getting a response. It includes connecting to the model provider, parsing the response, and having fallback options when one provider is unavailable.8. Localization and Cultural Fit (~5%)Operating across countries isn't just translation. Coaching advice about formal business registration is completely different in Peru versus Kenya. The relevant government agencies, legal frameworks, financial products, and market conditions are all different. The system needs to resolve the right language, the right country context, and the right institutional references for each user, and allow per-country overrides without requiring a new deployment.ConclusionEach of these capabilities is essential, and they do overlap. Your proactive coaching engine needs your memory system to know who to reach out to. Your context assembly depends on your reliability layer to ensure profiles are current. Your instruction library depends on your localization system to serve the right version. The LLM is the engine, but a well-functioning service is the whole car: navigation, fuel system, chassis, dashboard. Each one is a product design decision that impacts user satisfaction, tool usage and ultimate impact.

Building an AI Coaching Service: The Key Ingredients

Two data sources are starting to paint a picture of Maia's early impact on entrepreneurs.<h3>ILO Colombia Study: Behavior Change in Coffee Country</h3>The International Labour Organization conducted an independent pre-post survey of coffee farmers in Colombia's Huila region—a baseline and follow-up two months later. The results were:Business practices: &nbsp; - Farmers keeping financial records: 56% → 78% &nbsp; - Formalized businesses: 2% → 17% &nbsp; - 55% of users reported directly implementing changes based on MAIA interactionsAgricultural practices: &nbsp; - Harvesting only mature beans: 80% → 85% &nbsp; - Minimizing waste/residues: 72% → 87% &nbsp; - Proper fertilizer application: 88% → 91% &nbsp; - Using climate-appropriate seeds: 83% → 87%The formalization jump—from 2% to 17%—is particularly notable. Formalization is notoriously hard to move. One farmer shared: "MAIA helped me formalize my coffee brand and create the company with all the legal requirements."What users valued most was having expert guidance available on demand. As one producer put it: "It's like having the agronomist right there on your phone." Another noted: "I used to mix fertilizers and composts together. Now I learned there are products that shouldn't be combined—MAIA explained that to me."<h3>Peru User Surveys: Self-Reported ROI</h3>Separately, we recently began surveying MAIA users in Peru and Panama at months 2 and 3 about changes in their business finances. The questions are simple: have your costs gone down due to Maia? Have your sales gone up due to Maia? Users choose between no change, up to S/150/month (~10% of average MSME profits), or more than S/150/month.From the first 165 responses: Cost Reduction (Month 2): 44% reported lower costs, Sales Increase (Month 3): 58% reported higher sales.&nbsp;These numbers are a floor. At months 2-3, MAIA's coaching relationship is just beginning. User profiles are still taking shape, and most haven't completed full coaching cycles. The progression from 44% (month 2) to 58% (month 3) illustrates momentum building over time. But still just using these early results, the average user reports S/47/month in cost savings and S/60/month in sales gains (at the bin midpoints). At a ballpark program cost of $1/user/month, that's a 12-17x ROI.According to <a data-mce-href="https://voxdev.org/voxdevlit/training-entrepreneurs" href="https://voxdev.org/voxdevlit/training-entrepreneurs" target="_blank" rel="noopener">https://voxdev.org/voxdevlit/training-entrepreneurs</a>, standard entrepreneurship training programs (typically 5-7 days in a classroom at ~$177/participant) produce around 5% sales gains and 10% profit improvement. The most effective "personal initiative" programs cost ~$750/participant and achieve 30% profit gains over two years. MAIA is showing comparable behavior change at a fraction of the cost: $12/year versus $177-750 for traditional training—and delivered continuously rather than as a one-time event, meaning its impact should rise over time while the impact of one-time training fades.<h3>What We're Learning</h3>The ILO study surfaced an important insight: "MAIA facilitates progressive adoption of digital technologies in rural areas, especially when combined with technical support and contextualized content." Users valued personalized, locally-relevant guidance. This is what enabled adoption of an advanced AI tool by users with very low tech comfort.These are early numbers. The surveys are self-reported, sample sizes are modest, and respondents still in the first months of the coaching relationship. But the direction is clear: users are changing behavior, and they're attributing it to MAIA.We'll keep measuring and sharing what we learn.

Impact Numbers

Context is a bit of a buzzword in the AI space, but it is central to making tools like MAIA useful. Context is the difference between the general answer to a general question any entrepreneur can get out of ChatGPT, and a targeted next step to solve the entrepreneur's next most pressing problem, along with accompanying assets to make it easy to implement.&nbsp;One approach is for tools like Maia is to be integrated with, and even come from, an MSMEs large supplier or customer. So for example a small retailer may buy many of their products from a large distributor. If that distributor creates an AI advisor plugged into all of the shop's historical orders, it could immediately give targeted advice on day 1 with all this rich context. There are many tools being developed along these lines today. But beyond the technical challenges in building them, there are two fundamental problems that keep coming up.First is the organization's desire to put their fingers on the scales and direct responses towards their own ends. This can start innocently enough- let's make sure the tool has our entire price list and features loaded so it can give accurate information, but we won't spend time doing that for other distributors, that's not our job. And it progresses from there. It makes sense: why would a distributor spend money building a tool that suggests' its competitors products and services over their own? Or that encourages small suppliers to only sell to them instead of diversifing their distribution. In the end that just means the tool is not a real coach, it is not operating in the best interests of the entrepreneur.Second, the entrepreneurs themselves behave strategically. If that retail shop is having some financial problems, would they really seek help from their AI coach about it? Couldn't the supplier use that early information to proactively limit trade credit?So the rich context comes at the cost of some very messy unaligned incentives which undermine how useful the tool is for the entrepreneur, and if/how they use it.The other approach is to build that context over time, which is how Maia works. This takes longer and isn't as elegant as the other short cuts, but has the benefits of being completely aligned with our users. We've found that business profiles become quite useful for giving grounded advice for a user after only a few weeks, and we spend a lot of time finding high-value low-effort first uses of the tool, to give the business owners a quick win while building context (eg making logos, a user favorite).&nbsp;

Context is King

MAIA started in Peru, which remains our largest country, but one year ago we started working in Panama, in partnership with <a data-mce-href="https://ampyme.gob.pa/" href="https://ampyme.gob.pa/" target="_blank" rel="noopener">AMYPME </a>and the ILO. As in all countries where we operate, the first stage is a detailed proecss of adaptation with local experts. In the case of Panama, our initial focus was small-scale producers of coffee and vegetables, and the tool was adapted and validated with sector experts. With them we collected a series of key country-specific manuals, guides, and rules, ending with country-sector specialized prompts and knowledge bases for RAG. MAIA @AMPYME was born.After multiple rounds of testing, in the face of very positive user feedback the tool was made generally available. <a data-mce-href="https://www.instagram.com/reel/DP4L1NeDlZ3/" href="https://www.instagram.com/reel/DP4L1NeDlZ3/" target="_blank" rel="noopener">Here you can see Jackeline Munoz, Panama's minister of labor and workforce development, along with leadership from AMPYME and the ILO at their official launch of Maia</a> <figure id="figure-of710qks13mi7olx5l"><img id="of710qks13mi7olx5l" src="https://cdn.durable.co/blocks/3aPOnZBjZgu8MNUusiZVgKUcnXeuzusOp8wZP8ZkxLJqrhz4He5GYJDOuatGTpFU.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/blocks/3aPOnZBjZgu8MNUusiZVgKUcnXeuzusOp8wZP8ZkxLJqrhz4He5GYJDOuatGTpFU.jpg" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure> Panama has quickly grown to become MAIA's second largest country with thousands of users, and we are entring new rounds of validation and expansion with the ILO and AMPYME!

MAIA @AMPYME Panama

Here is the latest MAIA overview, all data as of mid-oct<figure id="figure-5b2bfb0ylmhdn7qol" class=""><img id="5b2bfb0ylmhdn7qol" src="https://cdn.durable.co/blocks/3achi9WTMClfHY6KtCzRmPi0XvHixpDeVKKD1GDEDrwNWQnjRLdWpMFTNvif61xi.png" 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/blocks/3achi9WTMClfHY6KtCzRmPi0XvHixpDeVKKD1GDEDrwNWQnjRLdWpMFTNvif61xi.png" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure><figure id="figure-6prnpn70njmhdn9ylr" class=""><img id="6prnpn70njmhdn9ylr" src="https://cdn.durable.co/blocks/20AcUW0ag6AIPXvXY8nPE7tMnnFKsGApYsEK9zY2P3cG6nPMqyIZuTfYlN7sexIp.png" 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/blocks/20AcUW0ag6AIPXvXY8nPE7tMnnFKsGApYsEK9zY2P3cG6nPMqyIZuTfYlN7sexIp.png" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure><figure id="figure-hvg4xh6tkdmhdnal63" class=""><img id="hvg4xh6tkdmhdnal63" src="https://cdn.durable.co/blocks/5Ytvm0C6nO2q1tXTJ7QoKmgAhOkgfmXhCamLgax9qX3AyVT6n09ZAisKmDTMUsap.png" 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/blocks/5Ytvm0C6nO2q1tXTJ7QoKmgAhOkgfmXhCamLgax9qX3AyVT6n09ZAisKmDTMUsap.png" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure><figure id="figure-n86b8xmaxmhdnaypo" class=""><img id="n86b8xmaxmhdnaypo" src="https://cdn.durable.co/blocks/7DStsQ0OXEMMobEG5n9RttsfTe2UsOpHvclNWSEEsMnYlx8TYIl00ZxL5FjYISiD.png" 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/blocks/7DStsQ0OXEMMobEG5n9RttsfTe2UsOpHvclNWSEEsMnYlx8TYIl00ZxL5FjYISiD.png" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure><figure id="figure-7upurcrhw7mhdnba1p" class=""><img id="7upurcrhw7mhdnba1p" src="https://cdn.durable.co/blocks/8EAKEWkQ6ztkDmWEDWfjWnbJPptqyehEZP7VIGAa7sC1ZgNrtLUAzL4Rq44ruPf9.png" 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/blocks/8EAKEWkQ6ztkDmWEDWfjWnbJPptqyehEZP7VIGAa7sC1ZgNrtLUAzL4Rq44ruPf9.png" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure><figure id="figure-irmovvfuqomhdnbohd" class=""><img id="irmovvfuqomhdnbohd" src="https://cdn.durable.co/blocks/aIcKJKXZYTuqpVxlf0lZlMjtsAOpdCwCRJa8VQkvCKeOiDcXAe1bH3fAQ9eDTiHI.png" 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/blocks/aIcKJKXZYTuqpVxlf0lZlMjtsAOpdCwCRJa8VQkvCKeOiDcXAe1bH3fAQ9eDTiHI.png" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure><figure id="figure-rh1zlere9hmhdnc4ra" class=""><img id="rh1zlere9hmhdnc4ra" src="https://cdn.durable.co/blocks/c0ozxOaJC3bv3tzyxJTnzwLGTgMWykmvJjpHUMGDcrFSCPSJz84clvpW9OxD7DDi.png" 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/blocks/c0ozxOaJC3bv3tzyxJTnzwLGTgMWykmvJjpHUMGDcrFSCPSJz84clvpW9OxD7DDi.png" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure><figure id="figure-p8w8bfpcilmhdncefo" class=""><img id="p8w8bfpcilmhdncefo" src="https://cdn.durable.co/blocks/dxZIguzkjesNquf0snPsRFYzts6LX6DYpG5hwRJsNsUXiFkAc6RfUS8cSViUFUjM.png" alt="" title="" class="object-cover bg-gray-100 rounded-sm md:rounded-md lg:rounded-lg selected" 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/blocks/dxZIguzkjesNquf0snPsRFYzts6LX6DYpG5hwRJsNsUXiFkAc6RfUS8cSViUFUjM.png" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure><figure id="figure-d3iq6oioimhdncocx"><img id="d3iq6oioimhdncocx" src="https://cdn.durable.co/blocks/fwSsIGt1q1BcsrP9J3J1tAYa1VRGhL9bBpa7EixfwMt16n6G3MUOql6Ulawpv4sm.png" 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"/></figure>

Maia Overview Presentation

<a data-mce-href="https://mypeasesoria.org/blog/maia-s--intellectual--architecture--situated-entrepreneurial-coaching" href="https://mypeasesoria.org/blog/maia-s--intellectual--architecture--situated-entrepreneurial-coaching" target="_blank" rel="noopener">The previous post</a> described Maia's appraoch to situated entrepreneurial coaching through a continuous coaching cycle, as well as its basis in the literature. How do we actually implement that in practice?About 90%+ of the time, the cue is Maia's tip of the week*. This used to be human-curated and fixed by sector, but now our coaching engine derives a customized tip for each user, the next idea/best practice with the highest marginal impact on their business. This is derived based on the user's own priorities and history as reflected in their entrepreneurial profile**, Maia's 8-dimension diagnostic of their business***, the country-sector specific knowledge bases we've built with local experts, and also considers emerging and seasonal themes via search. These tips are expressed via a set of WhatsApp template shells with a few different approaches.Interaction with the tool following the cue quickly takes the user towards the action plan, which is calibrated to the businesses' level of development and priorities, both as expressed in the conversation and in the profile. Along with the plan, Maia provides every asset it can to make taking the next step as easy as possible, such as templates, images, checklists, if-then scenarios, etc.The guided reflection and reward stage is implemented through a second weekly message. Unlike the cue, which brings in new information and ideas, this follow-up message looks backwards, encouraging and rewarding action and reflecting on results to overcome obstacles and find new opportunities. Again, implemented though a series of WhatsApp template shells.Interaction with the tool on this follow-up message leads directly to adjusting plans and taking corrective action when things worked poorly, and building habbits when things worked well. It naturally leads to a new cue, to start the cycle again.* The other 10% are inbound user questions. Current cadence is fixed weekly, our roadmap is to move to adapting the cadence to user preferences** A structured profile of each user, including key business characteristics and goals. Will be discussed in a future post.*** This is synthesized based on diagnostic tools and training programs from a variety of organizations like the ILO, the IFC, the FAO, and consulting companies. Will also discuss in a future post.

Blog

Building an AI Coaching Service: The Key Ingredients

Impact Numbers

Context is King

MAIA @AMPYME Panama

Maia Overview Presentation

Part 2: Implementing this Architecture - MAIA's Coaching ...