Conversational AI Training Data & Orchestration Are Lagging
According to Deloitte it’s no longer about AI adoption or automation per se, but rather realising value. However, there are two impediments raised by respondents…training data & organisational orchestration.
Training Data
In Deloitte’s State of AI in the Enterprise Report, respondents were asked about challenges they are facing in terms of scaling AI initiatives, 44% of respondents raised “obtaining training data for models” as an impediment to scaling.
When considering organisations which are achieving success in their AI implementations, a key differentiator is having a well documented process for managing data and ensuring the quality of training data.
Data privacy and consent management together with safety concerns about AI systems loom large and still serves as an impediment. This is one of the drivers of advancement in the field of intelligent PII redaction tools, from Microsoft, Google and others.
Organisations are often overwhelmed by the sheer amount of both structured and unstructured data streaming into disparate data platforms. Leaving enterprises with new challenges in establishing and delivering human-accessible insight from these vast amounts of accumulated disparate data.
One of the case-studies raised the need to be able to accurately select training data, and fine-tune the most effective model, for a particular task, focussed on detecting customer patterns.
The best tool to achieve this is via a latent space by which an approach of weak supervision can be followed to detect true customer signals from noisy, limited, or imprecise data sources.
➡️ Please follow me on LinkedIn for updates on Conversational AI
Orchestration
The discipline of orchestration is raised on two levels…
1️⃣ Orchestrating Technology and Talent
With focus on AI delivering value, organisations are moving away from attempting to build their own AI solution in-house. Organisations are rather opting for AI as a product or a service. This was the feedback from 65% of respondents.
With many open-source products and available projects, there is the temptation to think implementation will be easy. Organisations are finding that taking solutions to production are complicated in terms of existing talent , systems and integration.
53% of companies are hiring experienced professionals with AI skills whilst 34% of companies are re-training internal resources.
In the case of the latter, advances in no-code studio approaches to data ingestion, analysis and designing training data bodes well.
2️⃣ Existing Systems & Technologies
Deloitte notes that the market is accelerating rapidly, but outcomes are lagging. Focus has shifted from AI adoption for the sake of it…to realising value, achieving outcomes and realising promised potential.
One of the challenges raised in scaling AI initiatives is the integration of AI into the organisation’s existing day-to-day operations and workflows. This, together with integrating with supporting organisational and business systems are some of the biggest challenges.
➡️ Please follow me on LinkedIn for updates on Conversational AI
Conclusion
Worth mentioning is the problem of identifying the use-cases with the greatest business value, which was reported by 42% of respondents.
Customer Experience remains as one of the most convenient avenues to a successful AI implementation, to list only a few key use-cases:
- Customer Service Operations
- Voice Assistants, Chatbots and Conversational AI
- Personalisation
- Customer Feedback Analysis
I’m currently the Chief Evangelist @ HumanFirst. I explore and write about all things at the intersection of AI and language; ranging from LLMs, Chatbots, Voicebots, Development Frameworks, Data-Centric latent spaces and more.
