Artificial Intelligence (AI) is transforming many industries, and many CX teams are navigating mandates to incorporate it into their strategies. But let’s face it — today, most of the ways we use AI in CX are pointed at our teams and customers, not with them. Consider the following scene: Somewhere in New York, an exec is half-listening to a SaaS sales pitch. She got the call because an algorithm marked her profile as a “hot lead” and matched her with a sales agent who has sold to her industry. When she declines a free trial afterward, a manager skims an automatic transcript of the call to see what went wrong.
She is half-listening because her phone is lighting up with daily notifications, all timed to maximize the likelihood that she’ll be online. As she dismisses weekly app summaries and activity reminders, she pauses on a survey regarding her airline travel last week. “Your feedback is important to us.” She recalls already leaving direct (and scathing) feedback with the rebooking agent after the airline canceled her first flight without explanation. Skip.
The AI and personalization features underneath these interactions have permeated our professional and consumer lives in the last five years. And yet, it’s hardly the stuff of science fiction. Has our exec’s life been enriched? Do the SaaS vendors’ staff feel smarter and more efficient? Are CX organizations executing better decisions with more coordination than were possible without these features?
From Artificial Intelligence to Intelligent Infrastructure
Like many new technologies, AI has started its career by taking over simple tasks that drive high costs. In CX that includes lead scoring, scheduling, and bots to deflect common questions. At their best, these automation tools help businesses scale faster, operate more efficiently, and focus on building better products instead of giant contact centers.
But when it comes to understanding or building relationships with customers, the current standard is usually “no worse,” not “better than” traditional approaches. Along the way, automation often contributes to information overload and distraction (as with our busy exec). If we treat AI as one more tool in an arms race to claim customer attention without spending our own, it will not have improved lives.
At Frame AI, we believe the next chapters of AI for CX are about empowering CX leaders and entire organizations to respond effectively to their market. We draw inspiration from Stanford University professor Michael Jordan (no relation to the basketball player, sorry), who proposed that we consider AI adoption as a three-stage journey:
- Imitative Artificial Intelligence means getting computers to do what humans already do well (and this is where most of our CX tools are right now).
- Intelligence augmentation (IA) means building systems where people and machines do more together than they could alone.
- Intelligent infrastructure (II) means building systems that coordinate and complement the work of many.
Frame AI is on a mission to build Intelligent Infrastructure for CX, so companies can listen and adapt to their customers with a speed and precision that raises the bar for customer relationships and internal operations.
What Does the Intelligent CX Organization Look Like?
Quantitative views on qualitative concerns
Keep the viewpoints, lose the silos
Proactive vs. reactive default
Customer interactions are opportunities vs. costs
Building an intelligent CX organization isn’t a question of waiting for the future — this technology exists today. But it involves making some non-intuitive choices. Here are three from our playbook:
Listen to Your Unstructured Data
Recording and storing data about customer interactions has become so central to most teams, we are running out of metaphors to describe it. Chances are that your organization has already invested in a data warehouse, data lake, or even a data lakehouse to make sure data is safe and available. But, after storage, structured data — customer profile, marketing, and product lifecycle events, multiple-choice surveys, etc., commands nearly all organizational attention. Analyzing structured data is easier because it loads easily into graphs and spreadsheets, so we are constantly tweaking instrumentation and surveys to answer new questions with structured data. Unfortunately, this means many organizations are perpetually stuck fighting yesterday’s battle — structured data only answers questions we already know to ask, and only after enough new data has been collected, while the multitude of factors driving today’s CX remain unknown.
Organizations are missing an opportunity to continuously learn about their customer experience in their customers’ own words. Organic feedback is unstructured data and represents raw data from support tickets, community forums, call centers, and webchat. Examples of organic feedback include frustration with billing policies, praise for your Support team’s ingenuity, and feature suggestions. At least 80% of customer-generated data is unstructured, and by 2022, IDG predicts that will grow to 93%.
This unstructured data already lives in your CRM, Helpdesk, or Data Lake(house) — but it often gets ignored because most teams lack the tools or expertise to extract value. But recent advances in Natural Language Understanding (NLU) mean that AI systems like Frame AI can pull accurate, detailed summaries out of unstructured data, allowing you to drive quantitative analytics and processes. Building capacity with unstructured data will allow your team to humanize their analysis, efficiently digest feedback and activate what moves the needle for your customers and your organization. And most importantly, it may unlock major new opportunities in the data you’ve already collected. That’s AI letting individual teams accomplish things they never could before.
Let Technology Translate Between Teams
When two teams — say, Support and Product, try to compare notes on customers, they almost always find some misalignment in their viewpoints. Support tickets may be tagged based on the customer’s initial intent, but Product wants to align them with a customer journey diagram they’re using to plan the next roadmap. Sales defines segments one way in order to set territory, but Marketing defines them in another based on motivation.
This often leads in one of two equally bad directions:
- The teams disengage, deciding that they prefer to conduct brand new user research rather than using each others’ experiences. The outcome of this is a lot of repeated effort, and continued misalignment in the future.
- The teams digress into building “one taxonomy to rule them all”, convinced that _this time_ they will find a way of describing customer interactions that works for every use case. Most often, this project never completes but manages to block any more tactical efforts for months or more.
In reality, there is no perfect view of the customer — just the perfect view for a particular task. But by automatically tagging customer interactions very granularly based on unstructured data, AI can make sure that interaction data captured by each team can be aligned against all taxonomies, acting as a kind of Rosetta Stone to help restate insights in the terms that matter most to your peers. For example, Support recognizing that 30% of their inbound volume relates to “Login Issues” (their term) is much more actionable when surrounding text allows Product to split those between “Account Creation” and “Teammate Invite” journeys (their terms). That’s AI helping your teams collaborate more effectively.
Invest in Feedback Loops, Not Features
Applying any new technology to CX often gets billed as a project. “This quarter we are implementing a Customer Success management system.” “We have a 4-week sprint to include conversation sentiment in our Account Health scores.” But the nature of CX means that these projects are never truly finished. As your product and market evolve, the tools you use to assess and direct your efforts need to co-evolve. And out-of-date tools are actually a negative asset — if that account sentiment score seems inaccurate and gets ignored, it still blocks the way to creating something genuinely useful!
At Frame AI, we’re focused on building NLU applications where continual reassessment and improvement are baked in. Our CX Opslifecycle model includes making sure that models we train are aligned with outcomes we expect and can measure so that models can be periodically reassessed and tuned with new inputs or objectives — not just machine learning.
When embarking on any project involving machine learning, make sure to clearly separate the data feedback loop that most vendors promise (“we’ll improve with experience!”) from the business feedback loop that makes sure the system overall continues to generate results. Ask, “how can we evaluate the business outcomes of deploying this, and how will we tune it if things aren’t working?” Attention and iteration are the keys to ensuring that your teams and their tools keep working together, instead of ignoring each other.
Conclusion— The Past as Prologue
If today’s Artificial Intelligence landscape seems riddled with false starts and dead ends, it’s actually in good company. When mainframe computers began replacing rooms full of Human Computers in the 1970s, they made economic sense, but were difficult to operate, had embarrassing failures, and acted as awkward appendages to the human team. But the decades that followed saw business computers move from imitating labor to augmenting individual workers (a PC on every desk!), to helping coordinate whole teams (CRM, chat, etc.)
Many AI applications today feel like that mainframe era. But CX teams actually have the technology today to leap ahead in that cycle by focusing on AI that genuinely helps their teams and their customers to better understand each other.