Artificial Intelligence / Use Cases / Customer Service
The agent received assistance during the call, and the simple request was handled immediately
Artificial Intelligence in Customer Service and Contact Centers: Filter, Assist, Summarize, Track.The customer waits while the agent searches. Simple requests—a schedule, an order status, an address—clog up the queues where more complicated requests should be handled. Once the call ends, no one has summarized it. Artificial intelligence handles these three things without replacing the agent: a virtual agent that truly filters requests, assistance during the conversation, an automatically generated summary, and the conversation logged in the file. We’ve been building the telephony infrastructure for a long time; that’s what makes the rest possible.
Three Signs That Your Customer Service Team Is Operating in the Dark
They don't care about the agents. They care about what the agents see right in front of them while the customer is speaking —and what remains after the customer hangs up.
The customer speaks; the agent searches
The order is inERP, the history is at CRM, the procedure is on the intranet, and the latest email is in a shared inbox. The agent switches between four screens while the customer repeats what he already told his colleague yesterday. The wait time isn’t in the queue—it’s in the conversation.
The agent knows the business; that particular client doesn't yet.
Simple requests clog up the queues
“Are you open on Saturday?” “What’s the status of my order?” “Can you resend my invoice?” The answer is in the system, and the call lasts four minutes. Meanwhile, the complaint that deserved ten minutes of attention is put on hold with music playing.
What resolves itself should be left to resolve itself.
Once the call is over, nothing remains of it
A hastily typed three-line summary—or nothing at all. The customer calls back, and another agent has to start from scratch. The manager doesn't know why people are calling, what was promised, or how many times the same question has come up.
A conversation that leaves no trace never took place.
What the solution does, in plain language
Six functions, from the initial contact to case closure. None of them makes decisions for the agent; they all save the agent time.
A virtual agent that handles the simple tasks and delegates the rest
Whether by phone, chat, or on the website: they identify the customer, answer questions for which the answer is in the system—such as hours, status, or invoices—and forward the rest to the appropriate queue, along with the information they’ve already gathered.
The file on the screen when the ringtone plays
The customer is identified by their number or account, their profile, their open cases, or their most recent interaction. The agent begins the conversation with this information in mind, rather than starting by asking for it.
The suggested response during the conversation
The procedure that corresponds to the question, the relevant section of the knowledge base, the contract clause. The agent reads, adapts, and responds. He no longer has to look anything up.
The summary writes itself
What was asked, what was answered, what was promised, what remains to be done. Written during the call, reviewed by the officer in ten seconds, added to the file.
The right person, in the right language, the first time
By skill, by language—three languages in French-speaking Switzerland is the norm in this field, not the exception—by priority. The routing system takes into account what the customer has already told the virtual agent.
What a Thousand Conversations Reveal
Reasons for appeals, recurring issues, promises kept or broken, moments of tension. A weekly overview, not just a snapshot—and what the knowledge base should include.
A concrete, step-by-step example
A customer calls the customer service department of a retail company: their order has arrived, but with a different product code than the one they ordered. The contact center runs on Dynamics 365 Contact Center, and the telephony system runs on Teams.
What Remains Human
Artificial intelligence provides support; it does not make decisions on behalf of the company or manage the relationship. Five decisions remain the responsibility of designated individuals, and they are documented before the system goes live.
Decisions That Cannot Be Delegated
A virtual agent making a promise, a summary submitted without proofreading, or an angry customer being handled by a bot: these are the three ways to lose a customer—even with a good tool. We address these issues to prevent them from happening.
- Make a commitment. Whether it’s a deadline, a refund, or a goodwill gesture: never through a chatbot, always through a person who has the authority to make that promise.
- Decide what the virtual agent can handle on its own. The list of requests it processes from start to finish is short, created by the department, and grows as results come in—never before.
- Handling difficult calls. An angry customer, a complex complaint, or an important client: once identified, these calls are routed directly to a human agent.
- Reread the summary. Ten seconds—but by the agent who spoke. A false summary in a file is worse than no summary at all.
- Correct the knowledge base. An incorrect suggestion made during a call is flagged; someone corrects the procedure, not the template.
The data used, and what they require
A virtual agent who doesn't know the customer is worth less than an automated response system. The quality of the customer file and the knowledge base is what matters most.
Sources
- Dynamics 365 : the customer record, their files, their history, and their contracts.
- ERP: orders, delivery, invoices, inventory—what simple queries reveal.
- The knowledge base: procedures, requirements, hours, and sample responses. What the assistant suggests to the agent.
- Telephony and communication channels: calls, chat, and email, along with their transcripts when permitted.
- Past conversations: for context, trends, and tone—subject to the retention policies you set.
What they must have
- A single customer: one number, one account, one record. Having the same customer listed in three different records makes identification unreliable—that’s what Customer Insights is all about.
- A well-maintained knowledge base, with an owner. Without it, nothing is transferred to the virtual agent, and nothing is suggested to the agent.
- Defined queues and competencies: languages, topics, priorities. Routing does not dictate how your organization is structured.
- Recording and transcription rules: who is notified, what is retained, for how long, and where. This is a legal issue before it is a technical one.
- A real-world sample for the prototype: the last 500 calls and chats—not a demo.
Where artificial intelligence works, and which one we should study first
The contact center operates entirely online. There’s only one thing that sometimes brings a call to your office: the human voice.
At Microsoft
RecommendedDynamics 365 Contact Center or Luware Nimbus at Teams, phone services at Teams, and Copilot Studio for the virtual agent: the entire system is designed to work together online.
At your facility
RarelyA contact center located entirely on your premises is another option. What can remain there are the recorded voice messages and their transcriptions—see “hybrid.”
Partly at Microsoft, partly at your office
PossibleConversations and case files at Microsoft; recordings, transcriptions, and summaries at your location when a written record is required—such as in banking, healthcare, and the public sector.
Next to the machine
Things to AvoidA contact center does not have a workshop. The only exception is a remote location that must continue to respond without a connection—a special case.
Why Online, and How the Voice Channel Is Changing Things
Three criteria are decisive for this use case. The third applies only to part of the solution, and that is what makes the hybrid approach possible without having to change everything.
| Criterion | What tips the scales | Toward what |
|---|---|---|
| Where are the phone service and the file located? | Teams For voice calls, Dynamics 365 or Luware ; for the contact center, Customer Service; for your account: everything is online and designed to work together. Moving a single piece breaks the whole system. | Online |
| Features included with the products | Summaries, agent assistance, virtual agents, and skill-based routing are coming to Dynamics 365 Contact Center and are available on a pay-as-you-go basis. Building them elsewhere would cost more and be less effective. | Online |
| Where Recordings Belong | When a contract requires that the recorded audio and its transcript remain on your premises, only these specific steps are performed on-site: recording, transcription, and summarization, while the file itself remains online. | Hybrid, for voice only |
Dynamics 365 's Contact Center AI features are billed—per conversation, per summary, or per virtual agent message. The actual cost becomes clear in the third month: we calculate it based on your previous call volume, and someone reviews the bill afterward. The regions where the recordings are stored are documented by Microsoft and verified again before each proposal.
What the solution connects to
A contact center is only as good as its telephony infrastructure and the back-end system. That is precisely what we have been saying for years —long before artificial intelligence came along.
Infrastructure, When It Is Needed
As for the contact center itself, none: everything is handled online. When regulations require that voice calls remain on-premises, we also scale the necessary infrastructure —for that portion only.
Record, transcribe, and summarize without a single minute of audio being produced
Recordings stored on Dell storage in your datacenter, along with their legal retention requirements; transcription and summarization on a Dell server PowerEdge equipped with graphics processing units, on Azure locally or on an open-architecture system; only the summary is uploaded to the online file. This is the “voice and transcription locally” option, which will be described as a separate use case.
How We Carry Out the Project
Five steps, and the order matters: the knowledge base comes before the virtual agent, and the summary comes before filtering. The functions that assist the agent are deployed first; those that interact with the customer are deployed last.
The Context: The Contact Center as It Is
Actual reasons for calls over the course of a month, wait times, languages, handling times, what agents look for during a call, what is promised but never followed up on. And the status of the customer file and the knowledge base.
The knowledge base and the rules
The procedures and standard responses the assistant will suggest when interacting with a property owner. The virtual agent’s guidelines: what it resolves, what it escalates, how it introduces itself, and in which languages.
The prototype, in a single file and with one team
First, the agent support and summaries—for a team over the course of a few weeks: we measure processing time, the quality of the summaries as judged by the agents, and useful versus useless suggestions. The virtual agent comes next, for two simple reasons, running in parallel.
Connection and Commissioning
Telephony, eERP, Customer Service, compliant call recording. File by file, language by language. Agents trained on call reasons, not on the system.
Maintenance and Measurement
Changing patterns, incorrect suggestions corrected in the knowledge base, requests that the virtual agent can now resolve, and tracked usage.
Estimated timeline: ten to sixteen weeks from the initial assessment to go-live for one queue, one team, and two issues resolved by the virtual agent, within an existing contact center. This is a rough estimate, not a commitment: measurement is part of the process. If the contact center itself needs to be set up, that is a separate project that must be completed first.
Scope and Limits
What artificial intelligence does not do in a contact center—and what we do not promise.
The virtual agent has trouble understanding some of the requests
An accent, background noise, a request phrased awkwardly, two topics in the same call: he sometimes makes mistakes, but he does so politely. That’s why he transfers the call at the slightest doubt, and why the list of issues he resolves on his own remains short. The actual rate is measured against the prototype.
The summary may miss a nuance
An implicit commitment, a touch of irony, a “we’ll see” that really meant no. The agent’s ten-second review is the solution. A summary submitted without a review is a risk, not a benefit.
The quality of the knowledge base is what matters most
A wrong suggestion during a call is worse than no suggestion at all: the agent reads it to the customer. The database is maintained by an administrator and is corrected with each report. Without it, nothing is tracked and nothing is supported.
Recording and transcription are subject to a legal framework
Who is notified, what information is retained, for how long, and where. You should review this framework with your data protection officer before implementing anything—and sometimes the data protection officer decides that the voice data should remain on your premises.
The cost is charged per call
Summary, agent support, virtual agent: billed on a pay-as-you-go basis, in addition to license fees. With several hundred calls per day, the budget adds up. We calculate the cost based on your actual volume, and it’s included in the proposal.
What We Won't Say
Let a virtual agent take the place of human agents. It eliminates the time wasted on simple requests and frees up that time for the conversations that matter. A contact center that loses its agents will lose its customers soon after.
Related cases and their corresponding pages
The products behind this use case, projects that have already been carried out, and a similar case.
Choosing the right model for Teams
Four questions are key; the product name comes later: the three ways to connect a contact center to Teams, and the telephony features below.
View the page TechnologyDynamics 365 Contact Center
Voice, chat, and files all in one place; a virtual agent that really does filter; and a summary that writes itself.
View the page TechnologyLuware Nimbus for Teams
Teams 's in-house contact center: the media remains in your tenant.
View the page Published ProjectTelephony Connected to Dynamics 365 Contact Center
The voice connected to the platform that hosts the virtual agent, the agent assistant, and the summary.
Read the project Published ProjectTelephony on Teams and the Nimbus Contact Center Luware
The underlying layer, once in place, is what makes the artificial intelligence on top possible.
Read the project A similar caseAutomated Email Processing
A written request arrives in a shared inbox, is processed, and converted into a case file before an agent reads it.
View the caseLet's talk about this use case
Tell us why your customers call
Recurring issues, what your agents look for during calls, and your current contact center and phone system. We’ll come back to the features we’d enable first, the issues a virtual agent could resolve on its own, and what a prototype in a queue would allow us to measure.
What we offer is the opportunity to meet the engineers who will do the work. Assessment, prototyping, and installation are all part of the project. The initial consultation, however, is free of charge.
Renens, Sion, Châtel-Saint-Denis
Microsoft Solutions Partner, a Gold Partner of Luware and AudioCodes. Telephony, contact center, case management, and the artificial intelligence mentioned above—all with the same point of contact.
Renens VD +41 21 806 37 15
Sion VS +41 27 552 00 22
Châtel-Saint-Denis FR +41 26 322 59 05