A view from the Valley
How tech companies are riding the next wave of AI
This week I’ve been accompanying a financial services client on a study tour, meeting with various tech companies who are developing the next wave of AI capabilities in customer service. What we found were groups of people devoted to addressing the opportunities and challenges of making AI work at scale - and less hype than I had been prepared for.
Here are my biggest takeaways.
1. Agentic AI is more about orchestration than autonomy, for now
A lot of the headline buzz about agentic AI talks of fully autonomous AI agents that can seek high level goals without supervision. In practice, few companies are seriously focusing on full autonomy in the short term. Instead, much of the thinking is about how to set up networks of AI and human employees who can work effectively together to achieve more complex tasks, including cross-functionally.
An example would be:
Personalised Offer Agent owned by product team, which uses a combination of Generative AI and Machine Learning to build a highly personalised set of offers for a customer
Knowledge Agent owned by the customer service team, which tests the offers against a ground truth knowledgebase to ensure accuracy
Service to Sales Agent owned by the customer service team, which guides a human Customer Service Representative to choose and present a personalised offer to the customer in the course of a customer service interaction
Sales Compliance Agent owned by the compliance team, which monitors the whole process in real time for compliance to company policy and regulatory rules
To scale these complex ecosystems, companies must decide how to set up enterprise AI platforms which allow them to build, deploy, manage and improve potentially thousands of AI agents.
Dust off some of that thinking about Robotic Process Automation Control Towers from 8-10 years ago! A lot of that is relevant to this next wave of technology.
2. No code / low code AI agents will require a whole new set of human skills
No code / low code AI agents mean that you don’t need a PhD. in data science to build one, but there are specific technical skills needed to be successful. Roles like the four below are likely to be in demand over the next few years:
Strategic Service Designer setting the overall strategy for AI in customer service. Where, how and when to deploy it? Where to differentiate vs. using out-of-the-box templates?
Solution Architect concerned with how to stitch together networks of AI agents, core tech platforms (CRM, ERP, Digital etc.) and data sources. When to use specialist AI tools vs. a cross-department/cross enterprise AI Agent platform for a particular job?
Workflow Designer decides how to break down a full process into a set of AI agents and human tasks, reengineering the process at the same time. How to define AI agents that can be re-used across multiple different processes?
Prompt Engineer writes the detailed instructions for each AI agent: objectives, activities, guardrails. How to instruct AI agents to optimise for efficiency and accuracy?
3. Context is critical. Data and knowledge management are critical for creating context
Customer Service AI works best when it has access to context: who is the customer? What are their previous interactions? How are they feeling right now? This customer context is not static. It develops over the course of an interaction as new information comes to light, and it develops in the longer term over the course of the customer’s life.
Knowledge content is also a key part of creating context. It grounds any conversation within the bounds of what is actually possible to solve the customer’s need. Some of the most valuable knowledge content is often not held in formal knowledgebases, but instead resides in documents on hard drives, emails, and in the heads of experienced employees.
A critical activity to fully scale AI in customer service is to build the means to collect and use contextual data within a single platform.
4. Model performance and UX are two critical success factors
Tech companies are looking to differentiate their products on two main fronts:
AI model performance: accuracy, speed, tone-of-voice and industry relevance. Often this is achieved by offering a selection of public and custom models, to be used for different purposes
User experience: in a world where companies could be managing thousands or even tens of thousands of AI agents, it’s vital to make it easy to manage these agents in a single interface. Simplicity of prompt builders and visualisations of workflow and performance are all areas where there is a lot of innovation happening
Thanks to teams from AWS, Clinc, Cresta, Kore.ai, Observe.AI, PolyAI, Salesforce and Sardine for joining the discussion and providing your insights.
Amsterdam event - 11th June
I’m delighted to be chairing the GenAI & Hyperautomation in Finance Summit in Amsterdam on 11th June.
We’ll be joined by top flight speakers from leading financial institutions including JP Morgan Chase, ING, UBS, ABN Amro, Citi and BNP Paribas - covering AI topics across the enterprise, including customer service.
You can register to attend here.
Recommended news articles
HR News: How can organisations address the contact centre retention crisis?
CMS Wire: The loyalty program boomerang: when rewards drive customers away
HR Dive: A 2-point rating scale can reduce racial bias in customer feedback, research finds
World Economic Forum: Meet the AI agents defining the new customer experience
Latest perspectives from BCG
New GenAI Tools Offer an Edge. Why Aren’t More Professional Services Firms Using Them?
Specialized GenAI delivers greater time savings and quality than general-purpose tools yet is used by less than 40% of legal, tax and accounting, and financial professionals.
Significant barriers—including data privacy and confidentiality concerns, limited customizability, and the need for training—have stood in the way of increased uptake.
To surmount these challenges, information services providers need to work closely with their professional customers, while professional services firms need to prepare their data and workforce to leverage the new tools’ expanding functionality.
Thanks for sharing Nick. Completely agree that orchestration is key here, making any human interventation add value rather than merely a gatekeeper of processes and tribal knowledge.
The thinking from 8-10 years ago is interesting. Many of the problems being solved by AI today aren't new, though the companies doing the solving may well be.