Improving customer experiences and services
Customer Interaction
Understanding and interacting effectively with customers is one of the critical elements of a company’s success. It ranges from understanding their needs and preferences to prioritizing which customer segments to focus on, and extends across all interactions — from the first sales contact to after-sales support. All touchpoints share one common risk: mistakes can result in significant opportunity costs in the form of profits never made.
AI tools can be great helpers in these efforts. If applied well, solid machine learning approaches can find the hidden gems in a large customer database, identify the potential needs of those customers, and guide them to buy more products with higher profitability. Then, using Natural Language Processing (NLP) tools can greatly improve sales conversion rates, reduce customer service efforts and increase customer satisfaction and retention.
Unfortunately, in many cases, AI-based customer analysis and interactions are implemented with the opposite effect, where customer analysis data generated by an AI-powered black box creates results far inferior to human analysis. The biggest problems, however, are created by AI-driven interactions: most of us have already experienced AI chatbots that were the opposite of helpful. This is also evident in our AI chatbot audits, which highlight substantial room for improvement in many implementations.
Below, you find more information on examples of customer-focused AI and its potential. Each reveals more details when you click them. And: please feel free to reach out to us to learn how we can help you use these technologies!
The 9senses Chatbot Audit evaluates the performance of your chatbot from a user perspective.
When we are using generative AI, to create new output we are exploiting knowledge previously generated by humans. But who is rebuilding knowledge for the next generation?
9senses Market Analysis: Only 16 out of 129 automotive providers in the DACH region use AI chatbots for customer service - with significant variations in quality
Large Language Models judge your work differently every time you ask. That is not just a quirk of the technology — it is a fundamental challenge to how we create,…
AI hallucinations aren't random — they cluster, systematically and predictably, in the topics you cannot independently verify.
Conversational AI is dominated by English, with serious consequences for other languages that are structural and can only be resolved with significant effort.
Renowned AI experts attribute a significant chance to AI that it could lead to the end of humanity. Is that a realistic prediction or doomsday talk? Short answer: it's not…Chatbots
A chatbot on your website that helps prospects and customers is available 24/7 and has access to all the relevant knowledge.
Call Centers
Supporting or fully automating voice-based customer interactions is an area where carefully designed AI can play a major role.
Customer Care and Services
Fully or partially auto-mating front-facing workflows improve customer experiences and reduces cost.
Customer Insights
Understanding and prioritizing your customers better, finding opportunities and detecting fraud helps to better focus business effort.
Chatbots
They are springing up like mushrooms on many websites, those little boxes on the lower right corner offering instant help by an "AI Assistant". And the idea is indeed fascinating: an electronic employee that is available 24/7, never gets tired and knows more than even the most experienced sales or customer care person. On top of that it can eliminate workload on employees, reduce wait times and provide answers much faster than a human could.
In most cases, as we all experience when trying out those AI bots, they fail at very mundane tasks and are plagued by many of the problems inherent to Natural Language Processing. The Chatbot Audit developed by 9senses shows that many bots do not even get close to exploring their potential.
Benefits of a good chatbot
A well-designed chatbot can create big benefits, driving sales and customer satisfaction up, while decreasing labor cost for sales and customer service employees. Below are a few key numbers from various studies evaluating key benefits.
%
reduction in resolution time
%
reduction in customer service minutes
%
increase in direct sales conversions
%
increase in customer satisfaction
Call Centers
Adding AI-based interactions to phone-based services can also significantly improve both revenue and customer satisfaction, particularly when it comes to intent recognition and distribution of calls. Compared to chatbots, the implementation paths are similar, while the challenges are vastly different.
First and foremost, automated call flows almost always need the potential for a human escalation path and thus must be integrated with call center solutions that can transfer to a human. Often, AI-driven call management is most successful once it excels at intention detection and directing calls into the correct pipeline - be this an automated response solving a problem, a redirection to a human agent, or a dedicated customer care and service workflow that may be fully or partially AI-supported.
Then, there are the challenges that exist far less in text-based interactions. Understanding dialects, non-native speakers, filtering background noises or dealing with bad voice line quality can negatively impact conversations, and needs clear exit paths, making the design of call center interactions more challenging.
And last, but not least, we have rarely ever seen pure AI solutions in successful call center applications, most are hybrid approaches, where AI-driven Natural Language Processing is heavily used in intent recognition and output generation, while many of the workflows are supported by logic and rules. Finding this balance is the most important element of successful call center implementations.
Customer Care and Services
AI-supported customer care solutions help streamline service and support processes, deliver consistent, high-quality responses across channels — while reducing operational costs and response times. For simple requests, they are already capable of handling them completely independently.
Often, these services are integrated into chatbots and call center solutions, sometimes they are only implemented in the backend processing of inputs received on traditional pathways, e.g. through letters or online forms.
The key differentiator is that these implementations are almost never plain AI solutions based on Machine Learning, but rather hybrid systems that combine the ease of using normal language in interactions (e.g. compared to filling a complicated long form) with strict and well-tested business logic that then turns the inputs into actionable business workflows with a tangible outcome, a service process completed with all feedback and audit checks, a satisfied customer, and - if possible - limited to no involvement of humans in simple and boring tasks that consume time.
Customer Insights
Understanding your customers at a deeper level is the foundation of sustainable growth. Data-driven customer insights reveal behavioral patterns, unmet needs, and value drivers that often remain hidden in complex data environments. AI-powered analytics enable companies to act on this data in ways that were previously out of reach. Typical applications include:
- Customer segmentation — moving beyond basic demographics to behavioral and value-based clusters that allow truly targeted engagement.
- Churn prediction — identifying customers at risk of leaving before they do, enabling timely and personalized retention measures.
- Next-best-action and offer optimisation — using purchase history, browsing behaviour, and contextual signals to guide customers toward products and services most relevant to them.
- Fraud detection — identifying anomalous patterns in real time to protect both the business and its customers.
- Lifetime value modelling — prioritizing effort and investment toward the customers and segments with the highest long-term potential.
What these applications share is a dependence on clean, well-structured data and thoughtful model design. A segmentation model built on poor data will produce poor segments — and acting on them can be worse than acting on intuition alone.
For a deeper look at the data foundations that make reliable customer insights possible — including data pipelines, knowledge management, and analytics architecture — visit our Data and Knowledge Management page.
Setting up customer interactions successfully
Successfully setting up customer-facing AI - irrespective of it being a chatbot, call center automations or service workflows - requires structured planning, iteration, and ongoing refinement. Like training and coaching a new employee is essential, even advanced out-of-the-box systems need significant tuning before they reliably serve customers and support business goals. The most effective implementations define clear objectives, prototype rigorously, implement with governance in mind, and continuously test and improve responses.
building customer-facing AI
From idea to a reliable AI assistant
Good customer-facing AI is not bought, it is built and then supervised. The first four steps happen once, before launch. From the moment it goes live, the last three repeat - forever.
Start with the business, not the bot
Every project begins with a clear answer to “what?” and “why?”. What will the assistant do, which use cases does it cover, and what measurable benefit does it create — for the company, for staff, and for the people using it? Use cases are prioritised by value and effort before a single line is built.
- Goals first, technology second
- Prioritise by value and effort
- One agreed, realistic scope
Prototype before you commit
There are many ways to build customer-facing AI, so we prototype. We test several models and toolsets, find the right balance of retrieval, structured logic and machine learning, and prove a first few use cases. Architecture and hosting decisions — cloud or on-premises — and operating cost are settled here. The result is a scoped, ready-to-build plan.
- Compare models and tools
- Decide hosting and cost
- Prove the first use cases
Ground answers in your own knowledge
Retrieval-augmented generation looks up relevant passages from your own content first, then asks the model to answer using them. The model no longer answers from memory alone — it answers from your documents, and the reply can point back to the source it used. This is the single biggest lever against confident, wrong answers.
- Retrieve, then generate
- Your content, chunked and searchable
- Answers carry a source
Wire it into the real systems
The assistant only helps if it can reach the systems that hold the answers and perform the actions. We set up the technical environment, connect the CRM, order and knowledge systems, and make sure intent is recognised and output is well structured. These are rarely pure-AI builds: language models handle understanding, business logic handles the action.
- Backend integrations
- Intent in, structured output out
- Hybrid: model + logic
Catch the confident wrong answer
A fluent answer is not a correct one. Before a reply reaches a customer it passes through checks: is it grounded in a retrieved source, is it within the agreed scope, and is the system confident enough? Answers that fail are held back or handed to a human, rather than shown. Hallucinations cluster in exactly the topics nobody can quickly verify — so the gate matters most there.
- Grounding and source checks
- Scope and refusal thresholds
- Held back, not guessed
Test like a user, refine forever
The hardest part — and the one that never ends — is testing each use case the way a real user would, including other languages, and refining the weak cases. Like any new employee, the assistant needs supervision and retraining. Failing questions feed an evaluation set that drives the next round of improvement.
- Real, user-style questions
- Scored across dimensions
- Weak cases loop back
Supervision that never ends
Once live, quality, speed, escalation rate and language coverage are monitored continuously. Logging, human review and update cycles decide whether the system stays a controlled assistant or drifts into a fluent black box. When content or data changes, or quality slips, the loop runs again — back to scoping and refinement.
- Live quality dashboard
- Logging and human review
- Drift triggers a new loop
Please contact us to discuss your needs and ideas. We are looking forward to working with you on a successful solution that truly helps your business.