The Idea in 60 Seconds
- It seems likely that smartphones from Samsung, google and apple will have ai assistants in 2024.
- Those agents will then visit websites to do the bidding of their owners.
- That means the AI assistants need to understand how to interact with the websites they visit.
- The AI assistants will prioritize the websites they learn how to use.
- High authority websites with millions of visitors like Amazon will be learned quickly.
- Long tail websites (including many Australian brands) might be harder to train.
- Since the source of the traffic that websites receive is going to change dramatically in the next few yeas, away from search and towards AI Assistants, this is an important topic.
How An AI Assistant Would Like To Engage With Your Website
It seems likely that smartphones from Samsung, Google and Apple will have AI assistants by the end of the year. Samsung may announce theirs today. (January 17th 2024.) Those agents will then visit websites on behalf of their users. So, what happens then?
There are usability themes which are consistent across the web but every website is different. AI Assistants are probably going to have to be trained on the idiosyncrasies of how each website works. The only rational way to do that is to prioritize the ones they learn.
Having an AI Assistant operate a website like a human is not a very efficient solution. A more medium term solution would be for each website (especially those long tail websites which won’t be learned by the assistants for a long time) would be to enable their Evaluate, Buy & Self Service facilities to be available as eservices / via an API.
Computers can talk to each other through APIs and it will be important for brands to put this facility in place because the source of their traffic is going to change when AI Assistants come online.
Traffic from search (paid and organic), direct and affiliates will be vastly reduced and an increasing proportion of traffic will come from AI assistants. Getting your API strategy right is an opportunity to grow the traffic you receive.
What is an API?
An API, or Application Programming Interface, acts as a mediator between different software applications, allowing them to communicate and exchange data.
Imagine the lobby of a company you’re visiting. You go in, tell them who you are, provide some ID and there is a map showing all the features which are available in the office you’re visiting. Canteen. Toilets. Car Park. Maintenance Department. Security. That’s very much like an API. APIs direct software ‘visitors’ to the right place to get the things the website can do (eServices) they want.
One of the key changes Generative AI provides is to allow conversational interfaces. To add to the analogy we already have, this is a bit like adding a concierge (representing the API’s interface, particularly those with NLP -Natural Language Processing – capabilities) assists visitors (AI assistants) by guiding them to various services (website functionalities/eservices) based on their requests.
A concierge in an office is a bit like an API enabled with NLP.
Building an API for AI Assistants ? What Should It Be Able To Do?
The primary thing a new API should do is to help the visiting agent achieve the user’s goal. As a result, the first feature I’d work on is including Natural Language Processing (NLP) Support. Natural language is how most AI assistants will prefer to interact with them. It’s faster than pressing buttons on the site and waiting for screens to load. Natural Language is also a more direct route to the answer to the user’s question than fishing around on the website looking for it.
As such, the API should include :
- Intents Cataloging:
The concierge, equipped with NLP, is able to discern the intent behind each query. He then directs them (provides a link, for example) to the resources to satisfy the need. It’s basically a menu of services the API can provide. The API needs to allow visiting agents to facilitate common tasks like placing orders, booking appointments, or requesting services. - Personalization options :
Given the data AI assistants have about their users, APIs that allow for personalized recommendations or services will be more effective. - Security and Privacy Features:
As AI assistants handle sensitive user data, APIs must include security protocols to protect user information.
After that, you need to have a framework for improving it
- Overflow Management:
When the concierge encounters a request or question it cannot fulfill or answer, it should be logged and answered for next time. Companies should create a list of things their website doesn’t do so the visiting bot is clear on that as well as what it can. - Human Intervention and Learning:
The overflow tool creates an opportunity for human staff to review unmet needs or queries, allowing them to update the API’s capabilities accordingly. This continuous learning loop can lead to a more robust and responsive API over time.
How much will it cost to build this sort of API?
Creating such APIs can be expensive (probably at least $10,000, possibly in the $100ks for larger companies.) Exactly how much it will cost comes down to the complexity of your existing environment and depends on the scope, the quality of your in house technical skills, the number of people it needs to service and your specific security requirements,
It’s probably too early to start building them but it’s possible to start planning. If I was working for a corporate, I would be assessing my current digital capabilities, getting some high level technical advice, agreeing principals for the API requirements and staying flexibility.
If and when API standards are released by Google, Apple and Samsung for how a website should engage with an AI Agent, companies which have prepared like this will be in a better position to implement If every website in the world decides to do this at the same time, the resources involved will become expensive so it will pay to prepare.
Which Industries Will Be Most Affected ?
Websites that engage in e-commerce or those selling products and services are likely to benefit the most from adapting to AI assistant compatibility.
- E-Commerce Platforms:
These sites stand to gain immensely as AI assistants can facilitate smooth shopping experiences, from product discovery to transaction completion. Simplified navigation, and easy check-out processes are particularly beneficial for visiting agents. You want to make it as easy as possible for AI Agents to buy. - Local Businesses and Services:
Local businesses like restaurants, salons, or service providers can benefit from being easily discoverable and bookable via AI assistants. For instance, users could ask their AI assistant to find the nearest Italian restaurant or book a haircut appointment. - Information and Content Providers:
News websites, blogs, and educational platforms can attract more visitors by being ready to provide entertainment, news and information faster to the user.
In each of these cases, producing APIs will simplifying the user’s interaction with the website when it’s an AI assistant, rather than the user themselves, visiting the site.
There Are Good Reasons To Think About APIs – Even Before Standards Come Along
Suggesting that smartphones will have AI assistants very soon is not a controversial idea.
Developing sophisticated, AI-compatible APIs, businesses can ensure they not only keep pace with this change but also leverage it to enhance user engagement and satisfaction and, critically, drive more traffic when their old paths to visitors dry up.
Of course, Multimodal AI which can ‘see’ a page and identify buttons and such will make it easier for AI Assistants to navigate sites. There are tools like Rabbit which are specifically designed to overcome the need for these APIs. My view is that ultimately, API access is still going to be better. It will still be more efficient (for both sides) to have the agent engage with an API rather than serve millions of pages to users who aren’t there and use limited compute resources with multimodal AI to analyze billions of web-pages when API access would render that unnecessary.
APIs will make websites easier to access and more efficient for AI assistant visitors. The simpler and more efficient it is for an AI assistant to access and retrieve information from a site, the more likely it is to visit that site. Loading screens and waiting for buttons to respond is not the most efficient way for an AI Assistant to interact with a website.
For those that implement them first it could provide a short term competitive advantage: Finally, the overflows (see above) are a good source of new information about your visitors wants and needs.
A convergence towards a standard set of API features across these Assistant Ais will come at some point. Audis, BMWs and Toyotas all have similar components, and an overlap in features (horns, indicators, windscreens) but they also have unique elements. Standardization could streamline development processes and ensure broader compatibility, but it also requires websites to stay abreast of evolving trends and be flexible in their digital strategies.
Neil Aitken - Blog 