New product development

Q: What have you learnt about using AI in NPD?

A: Every time we have sat down with a publisher to brainstorm new AI services they can deliver, we have been impressed by the original concepts they have come up with.

Often the starting point has been to focus on the support, sales, marketing and customer service aspects first for creating AI services and then move on to the editorial side.

This is because, quite often, those things are well documented already and content can simply be ingested and used to train up a ‘chat AI’ for say, customer services or sales support, or an ‘event AI’ which can advise users on how to buy tickets, how to enter an award, how they can get to the venue and who’s speaking at 4pm on day two.

For more value-add services, publishers need to identify where they have already got strengths and depths in content. This is key to an early start with AI products and services, especially ones provided at a premium.

We have found that, frequently, there are topics that can quickly be used to create great AI services. Directory content also lends itself well to this.

There is also a great opportunity for creating bespoke services for significant business and government clients by using a combination of content from both sides. In this way, you could roll out multiple custom AI services for your major clients.

Q: In which use-case has AI proved most effective?

A: One publisher has created an incredibly powerful ‘expert AI’ service for their community. We used their 2,000 in-depth articles on all things guitar-related to create an exceptionally powerful AI that has deep answers on almost any topic and theme.

Crucially, the way the AI is trained is that there is a near real time sync between the publishing side and the AI service so that if any question is either not answered or is answered incorrectly, then the publisher simply can create an additional article or update the existing one(s) to improve on the answers provided.

The debug mode, notifications and reporting are also critical here. The debug mode allows the AI manager, who is in this case also the editor, to see which parts of which articles are being used to provide the answers, they can then quickly and easily use this information to update those if appropriate.

The alerts let them know if users are unhappy with answers, or if indeed the AI is unable to provide an answer to the questions.

By being able to see exactly what their industry is interested in, in real time, they are also better able to know what content to focus on. This is particularly important for Generative Engine Optimisation (GEO, which is rapidly becoming more important than SEO), since the articles they create to train their own AI are then picked up by Google and others and cited.

This was a journey though, because initially there was a disconnect between what the publisher thought the AI could do, and what it excelled at. For example, it could not give an exact number for all the products it had reviewed, but it could give a list of the top 10 or top 100 products for certain categories.

Once it was understood how the AI service worked, the editor has been able to quickly fill the gaps and refine the content to ensure an outstanding quality of answers to the questions being asked.

Critical to the success of the service has been ‘suggested questions’ — these inform the community of the different types of scenarios in which the AI can assist them.

Three best practice top tips

1. Identify which content you can best leverage for your initial AI products and services. Quite often, this might be on the sales / marketing / customer service side. In the event that it is published content, make sure that you select the content which works best with an AI; it is better to focus on a single or small number of topics where you have in-depth content than a whole range of topics where you have more lightweight content.
2. Understanding what your content is good for in an AI context is really important. The sooner you start training AIs and asking questions the better. In this case, there was some confusion and disappointment from the editorial team when initially presented with the AI service. That has changed dramatically though as the underlying LLMs (Large Language Models) have improved their answers, and the editor has understood what kinds of questions their content is great at answering. They have also added critical additional content, including what we refer to as ‘bridging content’ to make sure that the AI consistently gives great answers to the questions being asked.
3. Creating and showcasing example questions is also critical as it guides the user not only on what types of questions they can have answered, but also the style in which they can get answers. This is so important because AIs can be extensively configured in how they interact and deliver their answers. The related aspect here is the ‘system role’ which instructs the AI how to reply to any questions.

Founded in 2009, ours is the leading AI-infused Unified Business Platform, suitable for most businesses and organisations. The Affino SaaS Platform provides a complete single customer view solution for engaging with and monetising audiences. It combines audience CRM with automation, ecommerce, subscriptions and memberships, events, ad serving, and expert AI solutions.

Email: markus.karlsson@affino.com

Website: www.affino.com