Business-driven AI: limitations, frameworks, opportunities

Hey there,

It’s Saïd from Stimpack,

Welcome to the 3rd edition of 99founder-tactics, I’m thrilled to share this new issue with you.

If you received this newsletter from a friend and would like to get it directly in your inbox, here’s the link. I would love to have you with us.

This third issue is about how businesses can make the most of AI. AI isn’t just hype, yet businesses have a hard time selling AI-powered products. We will introduce a new framework intended to empower AI businesses and help them make informed decisions.

In this issue:

• 🚀AI 2024 quick tour.

• 😬AI adoption limits.

• 🎯The QPCS Framework for AI-powered Businesses.

• 💼New business-driven AI-cell architecture.

•  💵Business Opportunities & Impact.

Reading this issue will take ~ 5 minutes.

Let’s dive in!

AI 2024 Quick Tour

Let’s first explore the current AI landscape in 2024.

Generative AI vs. Discriminative AI:

• 🎨Generative: creates art, generates text, used in Q&A, and for inspiration.

• ⚡️Discriminative: separates classes of items, categorizes, detects.

Models & Techniques:

• 💬LLMs for text data.

• 🎞️Stable Diffusion for images/video.

• 🤖All of them are “Transformer”-based.

Providers:

• Chips/IP: think Nvidia that provides H200 GPUs for AI.

• Cloud providers: think AWS, Google, or Microsoft Azure.

• AI providers: think OpenAI, Google, IBM (who remembers Watson?😅).

General AI vs. Specific AI:

• Specific (narrow) AI: ChatGPT is considered a specific AI.

• General AI: an AI capable of human-level reasoning, referred to as AGI.

  • AGI: still lacking reasoning capabilities😅

Adoption

Did you know that more than 80% of businesses have already integrated artificial intelligence technology in some capacity, as per a recent report? It's quite astonishing, isn't it?

Adoption has been crazy so far, will it last? Furthermore, the question that arises is, what is the reason behind this widespread adoption? Won’t this inevitably end in mass churn?😅

If AI goes beyond the hype, there’s absolutely no reason for mass churn.

AI Adoption Limits

Many limits to its mass adoption, behind the hype:

Can discriminative help with the business-tech limitations?

• 😬Overall the same issues, but..

• Better performance frameworks.💯

• Conceived to detect/classify/score & cluster.💯

• More repeatable (white box models).💯

tl;dr: Yes, discriminative AI is complementary to generative AI.

A fun article about why generative AI is overrated🙃

[1] Tactics:

• 🎯 Make sure you know when to choose a generative AI model vs. a discriminative one.

• 🎯 Thinks of ways to combine models.

The QPCS Framework for AI-powered Businesses

We want to introduce a framework that’ll fulfill what AI-powered businesses need:

• 1. Quality (accuracy, precision, consistency)
  2. Model Predictability (deterministic output, focused)
  3. Creativity
  4. Stability (availability, performance)

  📢The QPCS framework is still experimental.

🚩Problem: Generative AI is very good at Innovation, but quite bad at Precision & Consistency. We need to consolidate the tech part first.

🎖Predictability: Using lower model temperatures and fine-tuning can help with getting predictable products. Prompt engineering can help.

🎖Creativity: Behind the hype, your product should provide a unique way to tackle a problem.

🎖Stability: the service you provide should be available and robust. Make sure your providers are established in the field. Ideally, ask for SLAs.

🎖Quality: We need to improve the quality of generative AI, by:

• Making it more accurate.

• Removing out-of-context outliers.

• While not degrading its creative abilities.

The QPCS framework will help us confidently:

• 🌟Select AI-powered products.

• 🚀Build our own AI products.

• 📈Unlock outstanding business value and not rely on the sole genAI.

[2] Tactics:

• 🎯 Use the QPCS framework when integrating AI models.

• 🎯 Evaluate external AI services before buying them using the QPCS framework.

• 🎯 Measure the QPCS and use the metrics for marketing purposes.

New Business-driven AI-cell architecture

Business-driven, why?🤔 Because it reinforces many pillars of the QPCS.

What does a new AI cell look like:

1. Traditional generative AI model; eg: an LLM.

2. A new discriminative step, either:

   • classifier, detector, filter, anomaly-detector, optimizer

   • that takes genAI as input

3. A feedback loop.

🏗️Hybrid systems: will combine cells in sequences & parallel flows.

[3] Tactics:

• 🎯 Implement the QPCS into your AI-powered applications.

• 🎯 Evaluate external AI services before buying them using the QPCS framework.

• 🎯 Goal: Introducing such an AI cell should lower your churn rate.

Business Opportunities & Impact

List of business use cases

Let's consider 2 crucial business goals:

1. Increase Revenue
2. Reduce Churn

Increase your revenue with AI-powered A/B test

• Generate AI variants of your Landing Page.💯

• 📊A/B test, measure conversions.

• 📊Keep the highest one.

• 📺Run Ads.

 Hint: Use Stimpack to create multiple LPs and select the best variant.

Increase your customers’ revenue by building AI-powered products

• Make sure to fulfill the QPCS framework💯

• 📊Measure the QPCS metrics.

• 📊Ideally, measure an ROI score.

• 📢Market these metrics: "80 accuracy", "95% repeatable", "99.99% stable", +400% ROI.

Lower churn with AI detecting in-app small signals...📊

• 🚨"User has a bad user experience" => trigger alert.

• 🚨"User probability of downgrading soon" => trigger alert.

• 🚨"User connects & uses less" => trigger alert.

Note: Dynamic pricing is also a field AI can totally help with

Hope you enjoyed this week’s issue folks, let me know.❤️

We are still small, but looking for partners.

🙌I would love your feedback on this issue. If you like it, feel free to tell your friends and partners about it. I also aspire to engage with all of you on a personal level: feel free to send me DMs on social media.

I’m also involved in various projects, such as an early-stage product growth platform called Stimpack; as well as a MicroLaunch (micro tools launchpad).

That’s it for this time! Thanks again, see you soon and take care.🙏

Saïd