By: Zohar Yardeni, Chief Product Officer
July 27, 2023
Generative AI (gAI) is exploding faster than almost anyone predicted, including experts and long-time AI practitioners. Large Language Models (LLMs) such as chatGPT in particular are quickly changing the way people interact with technology. These rapid seismic changes make it difficult to figure out what this all means for our companies. There just isn’t much historical pattern recognition we can apply here. And so we must reason from first principles. Here’s how we approached this at The Knot Worldwide.
The Knot Worldwide connects more than 4 million engaged couples with nearly 850,000 local wedding professionals through its global wedding brands and leading online Vendor Marketplace. The Knot Worldwide’s family of wedding brands offer a comprehensive suite of personalized wedding websites, planning tools, invitations and registry services that make wedding planning easier.
We started off by breaking up our gAI strategy into the following buckets:
- Product Features
- Impact on Existing Strategy
- Internal Efficiency
Product Features
Different companies in different industries will leverage gAI in different ways—but almost every company will be impacted. A good place to start is by asking where in your problem space or product surface can generative capabilities play a role? This might be text generation, image generation, or conversational interfaces. Planning a wedding is highly personal and involves a lot of pressure. We want to alleviate anxiety for couples. Hence conversational interfaces are an obvious place for us to start, whether to help get started, find the right vendors, or build a wedding website. The first project we started building is a conversational interface into our wedding marketplace. This will allow engaged couples to ask questions such as: “Find me hotel venues on the water in Long Island which can handle 200 plus guests and have been featured in wedding magazines.” Such an exchange quickly produces a relevant selection of wedding venues without having to spend time learning our interface, and without being constrained to structured data choices and filters.
One strategy for exploring blue-sky opportunities outside of your current product experience is to ask yourself: how would a new gAI-driven player disrupt my company or industry? At TKWW, we’re exploring this question with a small cross-functional “moonshot” team.
Public LLMs such as chatGPT will also play a role in user acquisition and product discovery, as SEO evolves to chatEO. It’s early days, but it can’t hurt to play around with plugins such as OpenAI’s. Essentially, you expose an API to chatGPT, and it figures out when and how to query it. Plugins are currently only available for chatGPT Plus users, and each plugin must be manually turned on. This, too, will quickly evolve.
Impact on Existing Strategy
Next up we asked ourselves: how does gAI impact our current strategic priorities? What is less important and what is more in a post-gAI world? Some of this will depend on your particular product and business. However, general patterns exist. Unique content and unique transactional capabilities probably become more important, while discovery innovation such as semantic search or step-by-step flows might be less important. On the unique content side, any product that has unique inventory such as articles, videos, reviews, or pricing data for example, will still be relevant in a gAI world. Similarly for marketplaces and ecommerce players, if unique transactions or connections happen on your platform, LLMs can’t easily replicate that—unless you choose to let them via an API. However, innovations in search or browsing capabilities are at far greater risk of being replaced or leap-frogged by LLMs.
At TKWW, we are working to provide engaged couples with even more of the unique content they need to find the best wedding pros who both represent the couple’s vision for their special day and who work within the confines of the couple’s personal budget and timing needs.
Internal Efficiency
This heading sounds boring, but there’s nothing boring about doubling engineering velocity. This is initially an offensive opportunity to lean in and be faster than the herd, but later will become a defensive must-have as more companies adopt gAI coding efficiency tools such as GitHub Copilot, AWS CodeWhisperer, Google Duet and the like. If you run a tech company, you probably already know this. If you don’t, then spend some time on YouTube researching. Upon first blush, it can appear that gAI tools are nothing more than a better way to grab off-the-shelf code versus cutting and pasting from Stack Overflow. But use it a bit yourself, watch others use it, or listen to others who have used it more extensively. The coding tools are less of a cut and paste, and more akin to a junior developer helping you with the more mundane heavy lifting of writing tests, more generic functionality, or quickly sifting through data. It’s a game changer. The faster your tech org gets good at it, and gets comfortable fitting it into your dev process, legal construct, and tech culture, the faster you will move as a company.
At TKWW, we ran a pilot with a few scrum teams who used a gAI coding companion in a compliant way, and we saw a velocity lift and good qualitative feedback. We’re working now to scale this across our entire team. An important consideration for us, and a reason for piloting, was to ensure we could navigate the legal complexity. Everyone wants to move fast, but you probably want to also work with your legal team to strike the right balance of speed versus risk management in an unclear and rapidly evolving regulatory and legal landscape. Some good practices include: creating a whitelist of approved tools that your teams can use, clear company policies (e.g. please don’t paste our customer data / PII into ChatGPT and ask it to summarize in a table), and establishing a communication channel with your legal team for speedy resolutions on gray areas.
There are many other efficiency opportunities for: generating text/imagery/creative, handling more human interactions at scale (e.g. customer service, sales), querying your data, distilling insights from or summarizing anything, and much more. Some will be niche opportunities that are unique to your particular operations, while most will be common needs shared with other companies. Startups are popping up every day to address the more common use cases. We are thus looking at off-the-shelf tools or simple prompt engineering solutions for those common uses rather than pursuing bigger builds like training models or building complex UIs.
TL;DR: Building a Generative AI Strategy
- Product Features: Better ways to solve your users’ problems + chatEO
- Impact on Existing Strategy: What is more important now (unique content, transactions, connection), and what is less (search, browsing, flows)
- Internal Efficiency: Coding tools! + other stuff