As we move beyond the initial buzz around AI/ML usage, we enter the next consolidation phase over the upcoming months. Here are my thoughts on what will be important in the next six months (and likely beyond) for many E-commerce and Travel & Hospitality companies:
1. Data Pipeline Optimization (DPO) to Feed AI Foundation Models
Every AI model is only as good as the data it can access.
However, in reality, we have a mix of legacy and cloud systems that are often not connected to each other. Consider financial data hosted in an on-premise ERP solution, payment data via a SaaS provider, internal knowledge systems (Confluence, Intranets, etc.), mixed offline/cloud-based data warehousing (e.g., contracts for travel suppliers), CRM/ticketing systems via cloud-based solutions (Salesforce, Lime, etc.), and messaging platforms. The list goes on…
An average company likely uses over 30 systems, and usually, they are not interconnected (except for some transactional data stored in databases – and even that is not always the case). Accessibility, especially via API hubs, was not often business-critical when systems were setup.
However, AI foundation models need to absorb all that information like a sponge, necessitating the design of robust, resource-optimized architectures to facilitate information flow from all systems. Consequently, many systems will require a fresh evaluation (we can assist with that).
2. Augmentation of Live and Historic Data
Current AI models almost exclusively work on historic data. However, in a business context and with real-time developments, this approach might only allow limited application and insight into AI outcomes. Therefore, efforts must be undertaken to merge data streams for better results (enabled by the proper pipeline optimization mentioned above).
For example, combining key performance indicator (KPI) data from monitoring systems with historical transaction or error data (e.g., Kibana, Grafana) can provide quicker and better estimates of system health.
An upgrade here would be integrating user-behavior data from live website interactions with the most-visited areas/clicks for real-time content optimization.
3. Multiple Specialized AIs Working Together for Governance and Outcome Optimization
To achieve a multiple-AI setup, a business analysis must first be conducted to determine where valuable data resides and how and when it should flow into the foundation models. These foundation models can already be specialized and trained on specific data structures or sources.
For example, in a customer care case for an Online Travel Agency (OTA) dealing with a disrupted flight where the passenger is stranded and has lost luggage amidst a labor strike, the following AIs could be used:
- An AI trained on customer-care responses integrated deeply into a CMS system (scanning previous tickets).
- An AI trained on (the data warehouse) of supplier contracts, aware of specific conditions negotiated with the airline (including refunds and compensation).
- An AI with a foundation model on legal systems, familiar with laws across different jurisdictions (considering the traveler’s origin, the airline, and the country where the traveler is stranded).
- An AI scanning various news sources for travel disruptions (strikes, governmental updates, and security information).
- A Voice AI capable of contacting the airline’s Lost & Found desk to inquire about lost luggage (this capability might be possible within the next six months or sooner).
- Finally, a Governance AI (trained on company principles, vision, training materials, and staff memos) which collates information, aligns them with internal policies and integrates it back into the ticket for updates and speedy resolution.
In conclusion, focusing on these areas will likely be a key driver for better efficiency, improved use-cases of AI and more productive outcomes.
