Working at the frontline of conversational AI for over two decades and with the copious amounts that is written about AI, I am often asked what advice I would give teams and departments considering an implementation.
The conversational AI solution you may be considering could be for customer service, HR, internal support, sales, or marketing. Additionally you may be looking for a virtual assistant, chatbot or voicebot for a website, app, social channel, or voice-enabled agents for a contact center.
Irrespective of the how your conversational AI solution is being used, organizations and teams will benefit by being aware of the following, to ensure successful and transformative deployments:
Observability is essential
Observability should be a foundational element when designing and building AI solutions. Its importance in conversational AI is that it ensures:
- Optimisation and performance assurance
- Quality control
- Continuous improvement
- Compliance, trust, audit, and governance
AI is not a one size fits all
Every organization is on a unique journey with AI and bespoke, domain-aware AI, delivers better outcomes. When it comes to conversational AI, specialization matters and so partnering with a trusted advisor to help you co-design a tailored AI strategy, aligned with your business goals and future roadmap, will deliver the best results.
The diversity of use cases that enterprises need and use conversational AI for are vastly different. For example, a retail chatbot needs industry and brand specific understanding whilst healthcare compliance demands great precision and regulatory handling. These needs cannot be met with general-purpose AI solutions built to work more broadly.
Getting to 80% is easy. The final 20% matters most
Initial AI development often progresses rapidly, with most of the core functionality built during this time. It is the final stages however, where delivering enterprise-grade reliability, trust, user experience, edge cases, and refinement, matter. These elements require a deeper effort and laser attention to overcome complex, interconnected challenges.
Getting to live deployment efficiently is best achieved by leveraging rapid prototyping to test and learn early. This is followed by investment in refining the critical last 20%. A trusted partner can be invaluable in navigating this phase effectively.
AI is a capability not just a technology
Successful AI delivery is grounded in knowledge, collaboration, and change leadership. It is much more than a technology deployment and must be the jurisdiction of more stakeholders than just the tech team.
There are two key aspects to conversational AI. The technology side of conversational AI, in terms of the chosen tools such as the models, platform and cloud services, is one aspect of conversational AI. The second relates to building capability to get insights, shape decisions and transform how an enterprise uses the information to adapt and evolve. This is much more complex.
Organisations must be committed to educating stakeholders, running collaborative workshops, and fostering cross-functional ownership. It’s important to select an AI partner who will engage with your team, share expertise, and help build your internal AI capability. AI must be seen as a strategic capability and not a product. It requires enterprise-wide integration and so a partner who understands this is critical.
Establish AI guardrails early
Clearly define your ethical, operational, and compliance guardrails at the start of your AI initiative. It is so much more difficult to retrofit safety, ethical or performance controls once deployed. Thinking about guardrails ‘later’ should never be an option. By addressing guardrails at the get go, safety, trustworthiness and compliancy will be built into your systems from bottom to top.
Having early clarity on what AI should and should not do, ensures ethical, safe, responsible, and compliant AI, as well as aligned outcomes and simplified decision-making throughout the project lifecycle.
