AI for Business Leaders
Operational implications in the age of AI
This article starts off with the usual disclaimer - none of the content published on our newsletter has been or will ever be written by chatGPT. And that’s because AI, in its current form, can summarize the past but it can’t predict or create the future.
AI has taken the world by storm and it’s here to stay.
Most of my twitter and LinkedIn feed is full of technological mumbo jumbo, doomsday predictions or euphoria for the impending productivity gains. I do partake in a reasonable amount of click bait, because I like to keep myself up to date. Here’s my analysis of where things stand at the moment. I hope this will help you navigate the challenges in your business as you plan for the future.
Grab a coffee and let’s dive in.
The Current State of AI
Technology
Let’s start off with the technology. Anyone who has spent a few minutes with chatGPT knows that it’s a big deal. Here are 5 technical terms you should be aware of
LLM - large language model. Text question, text answer. This is the base innovation that has sparked the entire uproar. It’s an AI system that can respond like a human.
Vectorized databases - AI models work with numbers representing probability. If you were to look inside a model you’d see different weights like z1=0.1673, z2=0.9775. Vectorized databases store such numbers which don’t mean much to humans.
Function calls - convert a conversation into an API call. Unthinkable, 6 months ago. Magic? Definitely.
Code interpreter - Throw in a huge csv file and query the file instantly. chatGPT writes Python code for you and makes the data sing. Pivot tables and reports as we know it are dead.
Models - chatGPT (OpenAI) is the frontrunner for blazing to 100m users in 2 months but there are numerous other emerging models. Llama2, Cohere, Claude, etc. These are LLMs. Other AI models focus on other tasks, Dall E2, Midjourney, Stable Diffusion for images, InflectionAI as a personal psychologist, etc.
Coming soon - Gemini by Google. Allegedly allows users to combine text queries with images, graph interpretations, etc. All eyes on this launch.
Social + Economic Implications
AI will become a commodity like AWS. Sure, there will be differences but the competition will likely benefit you, the end user.
As the world’s largest tech companies pour billions into AI, 3 things are certain
It’ll get faster, better and cheaper.
What’s important to note though is that AI will be drastically cheaper than hiring a human. And that’s terrifying but realistic. Jobs will disappear and new jobs will get created. The past is littered with people who have predicted the downfall of civilization.
It’s unlikely that your business demand will be impacted because people will always need physical services like a comfortable bed for the night, a freshly brewed beverage, etc. However, costs are about to reduce drastically.
Legal
There will be legal troubles for sure. Politicians are motivated to seize the narrative to retain power. OpenAI and other market leaders are motivated to hype AI-fear for regulatory capture - so that they can remain the anointed ones protecting us from evil AI variants.
Ultimately, fighting against technological progress is a losing battle. Imagine fighting against electricity, telephones or the internet.
This is an arms race. Once an organization (or country) deploys AI and gains superpowers, competing organizations (or countries) are forced to do the same. You cannot afford to think that laws (if and when enforced) will protect you and your business. Do not concern yourself with how the LLM was trained. The genie is out of the bottle. The time for your business to act is now.
So What?
What this means is that you’ll be able to drive more revenue with less money and time.
The holy grail.
How?
Convert inbound customer inquiries into bookings - not to mention instant, 24/7, multi-lingual responses
Internal tooling to turbo-charge productivity for social media posts, product descriptions, photos, etc.
Automated / assisted support to free up team’s time and improve guest satisfaction
ChatUI powered software interfaces that requires zero training
Custom reports generated instantly on demand (is this end of manual account reconciliation? 😰)
Predictive analytics telling you where your business needs to improve
For example
Hotels - AI isn’t going to replace housekeeping but it definitely can help automate guest messaging + concierge, ultimately freeing up staff time and providing assistance to guests with the following
Instant
24/7
Multi-lingual
Personalized
Experiences - AI can’t take people skydiving. But it can help drive bookings while you’re 10,000 feet in the air. Customers don’t have to wait for you to respond but can chat with an agent that has been trained on your historical messages, rates, reservations, availability, mimics your brand tone and even your voice 🤯
Online marketplaces - sales / support agents are expensive and slow. Agents cost between $15-60 per hour and the average first response time is 11mins. In that time, customers have already moved to their 10th browser tab and have booked with a competitor. Human agents should be spending their time on resolving hairy issues.
Travel - have you ever called an airline to reschedule a flight? ‘nuff said..🤣🤣
Putting AI to Work For You
Step 1 - Define your objective
You know your business the best so you should be in a position to understand the place to get started. Broadly speaking, there would be 3 possible areas to evaluate
Sales
Search and discoverability - how can you better assist customers and match them to the right product?
Booking funnel - can you simplify the booking process by understanding implicit information? (obviously I’m traveling with my kids - I’ve booked this 17 times!!)
Delivery
Servicing customer requests - omni-channel customer communication delivered with the right context
Guest experience - delivering more than just a service. Leveraging your data to provide a memorable experience
Back office operations
Team productivity - stop doing busy work and work on things that matter
Automation - get rid of cumbersome manual processes
Step 2 - Data
Now that we have a chatty AI (chatGPT), it’s utility will only be as good as the underlying data. There are four types of data that every organization deals with
Online static information - business description, location, phone number
Gated static information - historical communications, SOPs, recommendations
Online dynamic information - rates, availability, reviews
Gated dynamic information - reservations, customer preferences, promotions
LLMs currently have access only to Online Static Information. You should be very careful about keeping your gated information private. This is your edge.
Step 3 - Generalist Chatbot for Sales & Support
Every single organization should be deploying an online AI chatbot for 2 reasons
Forcing function to gather your data in one place to train a model
Low cost method of collecting user input and feedback (customers will figure it out or call you if urgent)
The long term implications are huge though. You would be on your way toward understanding and developing an AI strategy.
Step 4 - Expand
Once you have a working model, you can then think about expanding the scope of your project with the following questions
Revisit your big picture goal(s)
Can you connect your base model to new channels (Facebook, Email, Whatsapp, Zendesk, Twilio, etc.)?
What data + APIs does the AI model need to level up it’s capabilities? Can you feed in historical messages to improve responses.
Is it time to plug in dynamic data (rates, availability, reservations)?
Is there any tooling out there that can help you achieve your objective? No prizes for reinventing the wheel.
How can you break up the work into smaller pieces and iterate?
Your AI Journey
Acknowledge the elephant in the room
Define an objective that AI can help you with
Take stock of your data
Train a base model for your business
Apply it to a simple use case (generalist chatbot for sales / support)
Monitor customer interactions
Process feedback and retrain
Ingest additional data (historical messages)
Connect to new channels
Expand functionality using APIs and dynamic data
Other resources
Skift AI travel newsletter - weekly newsletter capturing latest news
Everything AI in travel - new publication from Tony Carne
Michael Parekh - more technically oriented AI news