AI Product Asks - Part 1
How to break down a business request into an AI product?
Hi there, it’s Matt! This is part of a larger series on AI product. Also check out our AI strategy thinking series.
Time to Read: 5 mins
So just how do you break down business requests for an AI Product? I’ve seen several ways that are successful, and couple that weren’t as a lead data scientist.
These are often difficult to classify. No two AI products or use cases are the same. Even ones in the same industry can take different approaches to business requests.
Breaking down business requirements is critical if you want to tease out good AI product features.
In today’s article I’ll:
Highlight how to break down a business request for an AI agent
Run a demand analysis
Break down operational level requests
This is the first in three-part series for breaking down AI product requirements. I’ll be starting this week with the business requests.
1. What do with a business request.
Have you ever gotten a business request you’re not sure about, with multiple business units asking? Then not sure what algo or features to build?
We’ve all been there. Too often it gets confusing. Lots of requests can on the surface seem deceptively similar. Even worse, they can use the same wording.
I’ve been watching different methods, and I’ve found this works:
Break down business requirements
Follow up on algo performance
Design Product features
This article series will focus on breaking down business requirements first.
I’m going to use a case study using an AI agent that analyzes user feedback. Let’s visualize these requests from four departments: data science team, administration, marketing, and customer service.
In this case, you have several requests for your AI Product:
Data Science Team: Wants an AI tool to find useful info in hundreds of megabytes of user complaints.
Administration Department: Needs an AI tool to automatically summarize meeting minutes.
Marketing Team: Wants an AI-driven promotion strategy for precise targeting of a new product.
Customer Service Team: Needs an AI system to handle the surge in user inquiries.
These demands can seem overwhelming. These can be not just one but multiple projects - your team has limited time. You need to break it down an prioritize.
Demand analysis is the best way to break it down, and we go into this in the next step:
2. How to do demand analysis
Demand analysis is a way of figuring out how much of a product or service people want. It's like asking, "How many users want to buy this and why?" or “How would this impact current processes?” Here’s four demand analysis foundations:
Understanding Customer/User Needs
Planning
Pricing
Marketing
Now let’s do demand analysis on the requests. For this article, we’ll skip pricing and marketing, and only focus on planning and needs since its internal facing.
As you can see there are two types of requests here:
operational requests (AI Text Analysis Tool and Meeting Minutes Tool)
strategic requests (AI generated promo plan & AI Customer Service)
Breaking down strategic requests are multifaceted and complex. They can affect alignment and business workflows. So, I’ll break that down in the next article.
Operational requests are focused on specific, immediate needs rather than long-term planning or overall strategy. They augment current user workflows, without too much change to the AI strategy.
They’re a bit easier to break down. You’ll be seeing these more often.
So, we’ll do demand analysis on operational requests on the first two rows:
Request #1: AI Text Analysis Tool
Original Request: The data science team hopes to mine effective information from historical feedback data.
Requirements Analysis: To refine this need, we should ask:
If a person manually analyzed the data, what conclusions would they draw?
What kind of information are they hoping to uncover?
If they can't clearly define the rules and expectations, the algorithm will struggle to make accurate predictions.
Conclusion: This is an unclear requirement. The team needs to specify what they want the AI to analyze or pattern match. Run opportunity discovery, and work with them to define more specific use cases. This should get you a better problem space to define the text data they need.
Request #2: Meeting Recording Tool
Original Request: The administration department wants an intelligent tool to record everything said in meetings.
Requirements Analysis:
The core technologies involved are Automatic Speech Recognition (ASR) and speech-to-text conversion. Let's break down the business needs:
Is it enough to record each word spoken?
Should the tool differentiate between multiple speakers or languages?
How will it format it for each person?
…to name just three!
For example, if 10 people are in a meeting, should it record each person separately? If so, we need to evaluate whether current ASR technology can meet these expectations.
Conclusion: This is a technology-dependent AI need. The scope should include technical requirements and consider whether to build or buy the solution. Audit the current infrastructure and tools to see if a solution can be integrated or implemented.
Takeaways
Breaking down an AI product request can be tricky. I’ll be honest, I’m so used to building first, that it took me awhile to get used to a product focus. Asking these questions help when you’re starting:
Who’s really requesting this AI feature/product? Not all requests are from initial person saying it. Sometimes it’s a telephone chain from farther up the chain. Clarifying the request from the original requestor is pretty critical. Assumptions can make or break how you classify a request.
Who can I work with on the demand analysis? Even if you’re a leading team, getting a second opinion is critical. Full stack tech is fine. Full stack product is near impossible. Find users, stakeholders, and product managers early. Front end your demand analysis work to get them all on the same page.
What requests get me traction to solve strategic ones? Start with these first. You need momentum and credibility with it. Break down operational level requests first. Strategic ones need advocates and sponsors. But you need a track record. Start by solving operational requests first.
These are questions that keep coming up in different forms. Reduce ambiguity around opportunity discovery as early as possible.
Before then, if you have any questions or comments, feel free to comment or reach out to me on:
In the next article, I’ll talk about how to break down an AI product strategic requests. It’s going to be fun!
Thanks for reading!
Enjoyed this article? Subscribe to get this newsletter every week 👇