Learn how to overcome the four biggest challenges & roadblocks in applying AI to understanding users

Artificial Intelligence (AI) is no longer science fiction. With the very tangible and practical applications now available for everyone, like the tools from OpenAI, AI has reached a new level of acceptance. However, there are still many roadblocks for businesses in overcoming challenges in AI deployment. In this article, we name these challenges and provide solutions on how they can be overcome.

Please note: The header picture was created by the DALL·E system from OpenAI. It creates realistic images from your own entered description. Click on this link to find out what the description was.

What makes AI deployment difficult?

As the travel industry finally emerged from the pandemic and had a remarkable year of recovery in 2022, we at BD4 talked to numerous OTAs and airlines, discussing their particular challenges and growth plans.

Despite the industry´s comeback in 2022, we’ve experienced contrasting moods in these conversations – from “let’s make the most of this positive momentum before the next pandemic wave” to a more gloomy “let’s see how consumers will manage the fallout of the war with the economic uncertainty”.

A common perspective acknowledged by OTAs & airlines in many of our discussions was the importance of significantly improving understanding of individual users and personalizing the “next best action” for every single consumer. This was considered as particularly relevant due to the pandemic changing traditional user behaviors and interests to an extent that common user understanding and historical CRM data have become less relevant.

However, the vast majority of the industry players haven’t yet deployed real-time capabilities to profile users coming to their websites (particularly the anonymous ones, representing 90%+ of their traffic). Profiling these users and simultaneously profiling tens of thousands of product attributes, to understand in real-time every single user’s intent to book, already poses a technological challenge.

Taking into account predicted price sensitivity and other data-driven individual insights to take appropriate actions, like steering certain products and pushing promo codes or incentives to the relevant users adds a high level of complexity to eCommerce and development teams.

So a lot of interest was given to BD4´s platform capabilities, allowing travel e-commerces to dynamically profile and understand in real-time every single user to create AI-powered, self-learning processes, with light integration requirements on top of existing tech stacks. Common targets to achieve and KPIs to improve were conversion rates and basket values, marketing expenditure or even technology costs.

What are the four biggest challenges in deploying AI solutions?

Despite the clear interest, some typical challenges and roadblocks kept coming up in our discussions, hindering some players from really taking their user profiling and understanding capabilities to another level. We categorized these and summarized our experiences on what worked well for our clients to overcome these challenges.

Challenge 1: Development roadmaps & legacy systems

The pandemic forced a pause on many IT developments, growth and innovation projects. As the industry recovers, quite often different departments are competing for the same scarce IT development resources, with pre-Covid projects stacking up on top of the new ones. To make matters worse, frequently in-house legacy systems, or rigid third-party tools reduce the ability to flexibly and quickly adapt to additional business requirements.

Solution oriented approach:

  • Work with the internal stakeholders and, if involved, external IT partners to properly size the ROI and business benefits in parallel to the technical effort.
  • Handle the project in a phased and agile manner, looking for use cases that are quick wins with little implementation effort.

Once these have generated concrete revenue uplifts and cost savings, expanding the project scope and getting the required resources assigned is more likely.

Old development roadmaps & legacy systems are a big challenge to deploy ai

Challenge 2: Lack of data & digital maturity and top line management backing

Even though digital teams understand the value and importance of AI-driven user profiling and automated decisioning projects e.g. to personalize the shopping experience 1:1, they often find reluctance from other departments or the top management, as they are not seen as mission critical, with the business value not being clearly understood.

Solution oriented approach:

  • Work with internal departments and external partners to properly analyze and describe the instantly beneficial use cases – beyond just using the usual technology buzzwords.
  • Clearly qualify and quantify the COI (cost of inaction) or risk missing out on major industry advancements.
Lack of data & digital maturity often cause challenges in ai deployment

Challenge 3: Missing data strategy or data science capabilities

When it comes to Data Science Capabilities and long-term data strategies, you find it all in the industry: From the fully stacked, cross-divisional teams with clear targets and data-driven roadmaps, to the one data-modeling ‘kid wonder’ that wants to do it all alone.

Solution oriented approach:

  • Develop an internal data-strategy document by listing and prioritizing important, future-proofing capabilities that need to be developed, trying to attach target business metrics to each one.
  • Analyze and agree on which of these can be advanced internally with existing or to-be-hired resources – and which capabilities would make sense to be created with external partners, minimizing time to market, efforts and risks of a total in-house strategy.
Missing data strategy or data science capabilities is a big challenge when deploying ai

Having a mid to long term data strategy with a hybrid inhouse/partnership philosophy is still a rare business advantage that will accelerate the go-to-market of your data science outputs, minimizing risk and reducing costs.

Challenge 4: Smoke & mirrors of tech vendors

One of the biggest challenges in today’s diverse and multilayered tech space is that it is tough to be a buyer. Marketing messages are sounding somewhat similar and often innovative providers have limited track records in your specific industry. Particularly challenging is the sheer amount of overlapping players and value propositions that make proper due diligence feel like solving Rubik’s Cube.

Not only do you need to have crafted a strategy with the big picture mindset to understand where you need to bring in an external partner, you then also need to have the resources and knowledge to properly vet and discuss with these vendors. Once you have selected the right partner, buying their service and solution is an additional challenge, managing the internal procurement and buy-in processes that come along.

Solution oriented approach:

  • To avoid a mismatch with an IT provider, focus on checking the core competency of the technology. Is it a fit for your purpose, does it come with the key capabilities needed and not just with the right buzzwords (personalization, cross device, real time, etc…).
  • Define and find the particular use cases that you are after according to your data-strategy.
  • Check if the required implementation setup is easy to deploy in your IT infrastructure. Complex implementations risk long-term projects and unprofitable investments.
Today's diverse and multilayered tech space is a huge challenge to find the most suitable ai provider.

How to overcome the challenges in AI deployment

Understand if you are looking at “best of breed” and modular approach for your needs, or if you value a monolith approach that is a swiss army knife type of solution? One will keep you flexible but requires a strong leadership in orchestrating several providers in parallel – and the other can be more convenient and guided, but might bring a trade-off in terms of best fit, costs or flexibility. In this regard it is relevant to take into account your internal technology and team landscape.

Do you have the right and sufficient resources to gain a significant ROI from the investment? Monolithic solutions often require extensive teams and cross-functional processes to make the most of the solution. If not, light integration solutions that come with managed services and close customer support can be the right choices.

What are your aspirations in AI deployment?

Do you already use AI for business processes in your company or are there roadblocks that hold you back? Do you still search for relevant AI-driven use cases that could help your company optimize commercial results or customer satisfaction? Then get in touch with us and let’s examine the possible applications and implementation approaches together.

You can also tell us about the challenges you already have been facing in AI deployment and implementation on LinkedIn or Twitter.

About the author, Tiago Relvao

Director of Business Development at BD4

Tiago has worked with leading global brands & destinations in travel and tourism for more than 15 years, has lived in six different countries and is fluent in four languages.

His passion: Growing Online Travel Brands by using AI-driven real-time personalization insights.
His focus: strategy, product and business development.