0.00034%

Moative Labs is now building its first product.

There is so much that is different about building an AI product if you come from SaaS/Tech.

01.

Zoom out of user behavior. Zoom into process design

Examples:

Outright bad idea: Users are filling a form. Instead we can make it conversational. No. A form is a familiar construct. It shows the user what is ahead and how much time it will take. A chat is excruciating and anxiety inducing. It takes more time.

Bad idea that looks like a good idea: Let's get news about a prospect and automatically write an email. Great but its still humans who know which news will be a better hook. If you absolutely must auto-write an email, design a flow where each news item gets its own email and the junior sales person evaluates which looks the best.

But even this, is a bad idea. The data that is valuable as intelligence is the one where you have an edge over the others. The friction to acquiring intel and sending emails is so low that news alone doesn't cut it. Accelerating a low throughput process only accelerates bad outcomes.

Good Idea: A good idea is one which looks at current user behavior, asks the "Toyota 5 Why's" and compresses/eliminates parts of the workflow – while keeping in mind, the users' vested interests in mind.

Example: Talk about how form 1508 typically takes 11 minutes, and how fetching the data for the form, digitization of documents to upload, checking if the form is consistent with docs and submitting – all can be automated away, thereby eliminating a part of the process. Talk about how the same digitized documents can be stored and re-uploaded in the subsequent step, which the user was otherwise doing all over again.

User's pains, aspirations, and behaviors are good to know. But with AI, users' jobs are up for re-engineering and they may not see it or like it.

02.

Use Generative AI like a star ingredient in an energy bar

CPG Marketers talk about some new, fancy nutrient in their energy bar. But if you zoom in, its 0.09% of the ingredients and its not even 5% of the daily nutrition intake recommendation. But its the magical ingredient that sells the product.

LLMs are the magic ingredient. They cannot do the entire thing today. Dozen other boring ingredients go into re-engineering the process. But the magic ingredient is a great touch. Its getting the user something that wasn't available before but no one has put a gun to your head and ask that the LLM be 99% of it.

Example:

Bad approach: We train a SERP AI to browse pages and summarize the information of interest for us

Good approach: Let's get a SERP API to pull the top 10 results. Let's use a crawler to get the data from those pages and then use NLP (NER, Topic Modelling, Keyword Extraction, and maybe, even Sentiment Analysis) to deterministically decide which results have the information that we need, through a relevance scoring system. Extract those data points. Ask an LLM to summarize it.

Our JTBD is not building an AI product. It is making a process as efficient as possible using a judicious use of tech that is available at our disposal.

PMs have to become process consultants. While that isn't easy, good PMs always ask "Why?" In addition, we need to define the problems better, have measurement systems for what qualifies as good, use the right tools, analyze the outcomes, iterate the approach and toolsets without showing affinity to any one approach or toolset, and finally improve the experience/efficiency as the tools evolve, but at no point degrading the current level of consistency, quality, and experience.

If you have not found out already, the above paragraph describes DMAIC from the world of Six Sigma. DFSS (Design For Six Sigma) is the design process for building new products.

If you are becoming a process consultant, you need tools from the world where they have perfected things to a point where error rates can reliably be predicted to be under 0.00034%. AI is a process enabler like no other tool, but the process should be the focus and AI should be the side show.

Sponsor Message: If you are wondering where to apply AI, but think that is a lot of hype, we are with you. But we know what 99.97% ingredients with a sprinkling of AI will be good for your process. Message me.