Field notes
Thoughts from people working in customer service, and what we are learning about AI as we use it every day.
42 notes
A high AI deflection rate doesn't mean your customers got help
Your dashboard says the AI handled the ticket. It cannot see the same customer come back through another channel an hour later.
Priya Raman · Support operations and measurement
Why Salesforce's contact center AI won't work until you fix your data
The demo works because it runs on one clean customer record. Your company has that customer's details spread across four systems that disagree.
Daniel Okafor · AI systems and data architecture
When Cisco automates your customer support, what is customer success for?
The routine work of customer success is becoming a feature your software vendor sells. What is left is the work the software cannot do on its own.
Claire Bessette · Customer success strategy
How much freedom should you give an AI customer service agent?
Some tasks are safe to hand to a bot. Others need a person to check first. Here is a simple way to decide which is which.
Marcus Bell · AI governance and safety
Why customers reject your chatbot even when it gives the right answer
Customers do not recoil from your chatbot because it is weak. They recoil because it showed up in a moment that needed a person.
Sofia Lindqvist · Customer experience and customer behaviour
Your AI agents will get worse over time. Make sure someone is watching.
An AI agent that worked well last month can quietly start failing. One named person has to be responsible for noticing.
Tomas Reuter · AI operations and reliability
What should you check before buying a conversational AI chatbot?
The vendor demo shows you the part that matters least. The parts that decide whether it works are the ones the demo skips.
Naomi Tan · Conversational and voice AI
Agree how you will measure AI success before you spend the budget
Most AI customer service projects decide what counts as success after the rollout. By then the only numbers left are the flattering ones.
Gregory Fenn · AI economics and ROI
Customers forgive a human mistake. They are harder on the same mistake from AI.
When a person gets it wrong, customers see an accident. When AI gets the same thing wrong, they see a choice the company made.
Hannah Vogel · Customer trust and AI ethics
Why so many customers are searching for a way to reach a human
Every month, tens of thousands of people search for how to get past a company's bot to a real person. That number is a warning sign you can measure.
Andre Costa · Contact centre workforce and operations
Why your support ticket volume didn't fall after you added AI
The AI is closing thousands of tickets, and your total volume is flat. That is not a failure. It is two predictable effects nobody planned for.
Priya Raman · Support operations and measurement
Why your AI agent passes the demo but fails real customers
A demo tests the questions the AI is ready for. Real customers ask the ones it is not. The gap between the two is where deployments go wrong.
Tomas Reuter · AI operations and reliability
Will AI replace customer service jobs?
Some companies tried full replacement, then quietly hired people back. What that reversal costs is the number worth knowing before you cut.
Andre Costa · Contact centre workforce and operations
Why AI customer service projects fail before they go live
The failure is usually set before launch. A company deploys AI without deciding what it wants, and the AI optimises into the empty space.
Claire Bessette · Customer success strategy
What HIPAA actually stops an AI agent from doing in healthcare support
The retail AI playbook gives the agent everything it might need. In healthcare, one rule makes that approach unlawful, and most teams design around it too late.
Marcus Bell · AI governance and safety
The AI security gap in your contact centre is probably already open
While leadership debates whether to adopt AI, agents are already pasting customer data into AI tools nobody approved. That is the incident.
Daniel Okafor · AI systems and data architecture
What AI automation means for your outsourced contact centre
AI removes the simple, high-volume contacts. That is exactly the volume your per-seat outsourcing contract is priced on.
Andre Costa · Contact centre workforce and operations
Four metrics that tell you more than your deflection rate
Deflection rate counts how much volume the AI absorbed. These four count whether the customer was actually helped.
Priya Raman · Support operations and measurement
Who is accountable when an AI follows every rule and still harms a customer?
Most AI governance covers rule-breaking. The harder case is the AI that broke no rule, and the customer was harmed anyway.
Marcus Bell · AI governance and safety
How to build the business case for bringing human agents back
Finance cut the agents on a cost-per-head number. To bring people back, you have to win on finance's own terms.
Gregory Fenn · AI economics and ROI
What it costs to take an AI customer service pilot to production
The pilot worked. The distance from there to production is mostly data work, and it has a real, plannable cost that the pilot hid.
Daniel Okafor · AI systems and data architecture
What to do when a better AI platform launches mid-contract
The AI customer service market moves faster than your contract term. You have more levers than the renewal date, but only if you look for them now.
Gregory Fenn · AI economics and ROI
Does your voice AI work as well for every accent?
Speech recognition is a well-studied technology with a well-studied weakness. In a contact centre, that weakness lands on a specific group of customers.
Naomi Tan · Conversational and voice AI
Someone has to own the handoffs between your AI agents
A contact centre running several AI agents fails in the gaps between them. Those gaps belong to no team unless you assign them to one.
Tomas Reuter · AI operations and reliability
How can you tell if customers have stopped trusting your chatbot?
Lost trust does not show up as a complaint. It shows up as customers quietly working around the bot. Those signals are in data you already hold.
Sofia Lindqvist · Customer experience and customer behaviour
The operating change that makes AI customer service actually pay off
A return on AI in customer service comes from changing how the work is routed, not from adding a model to the workflow you already had.
Gregory Fenn · AI economics and ROI
How a failed AI support ticket becomes a renewal you lose
On a B2B account, a support ticket the AI closed but did not solve is not a CSAT dent. It is an early renewal risk, sitting in a system nobody connects to revenue.
Claire Bessette · Customer success strategy
How to buy AI customer service: test it for resolution
Most AI buying is decided by the demo and the deflection rate. Both reward the appearance of working. Make a measured resolution rate the contract requirement.
Tomas Reuter · AI operations and reliability
Your frontline staff are quietly fixing your AI's mistakes
Agents correct what the AI gets wrong all day long. With no way to feed those corrections back, the AI never improves and the cost hides in agent time.
Hannah Vogel · Customer trust and AI ethics
The customer success work AI cannot see
The most valuable customer success work leaves no record in any system. When AI automates the recorded tasks, the unrecorded work is what a budget review puts at risk.
Andre Costa · Contact centre workforce and operations
Intercom renamed itself Fin: what it means if you buy the product
A 15-year-old customer service company has taken the name of its AI agent. That is a signal about where the roadmap and the money will go.
Gregory Fenn · AI economics and ROI
AI is being used to change how human agents sound on calls
Telus Digital is altering its agents' accents in real time. The technology works. The questions it raises for CX leaders are harder.
Naomi Tan · Conversational and voice AI
How the contact centre agent job is being redesigned around AI
AI is taking the routine contacts. For most service leaders that does not mean cutting agents, it means changing what the job is.
Andre Costa · Contact centre workforce and operations
Does outsourcing customer service still make sense once AI handles the routine work?
AI absorbs the high-volume contacts that made outsourcing cheap. That reopens the question of whether to keep a BPO or bring support back in-house.
Priya Raman · Support operations and measurement
Running a fleet of AI agents, not just one
Most companies do not have a single AI agent. They have several, spread across channels and teams. Managing the group is a different job from managing one.
Tomas Reuter · AI operations and reliability
Why your customer journey map breaks once AI handles contacts
A journey map draws a fixed path through fixed steps. Once an AI agent is answering customers, the path is no longer fixed, and the map describes a route they stopped taking.
Sofia Lindqvist · Customer experience and customer behaviour
What conversational AI is in customer service, and what it cannot do
A plain explainer for anyone new to the term: what it means, how it is different from the old phone menus, and where it still falls short.
Naomi Tan · Conversational and voice AI
How voice AI works in a contact centre
A plain walk through what happens between a customer speaking and the system answering, and what makes the difference between a call that works and one that does not.
Naomi Tan · Conversational and voice AI
Microsoft is shipping AI agents inside Dynamics 365 Contact Center
A platform that big building agents into the product changes the buying question. Here is what it signals.
Daniel Okafor · AI systems and data architecture
A recurring support problem is usually a business problem
When the same complaint keeps arriving, the fix is rarely in support. It is upstream.
Sofia Lindqvist · Customer experience and customer behaviour
When not to automate a customer follow-up
Some follow-ups should be automated. The ones where a real concern just landed are not them — and the dashboard cannot tell the difference.
Priya Raman · Support operations and measurement
Your AI agent can't see the customer's shadow workflow
Adoption metrics measure participation. They do not measure migration. The customer's parallel workflow is the trust your platform has not yet earned.
Daniel Okafor · AI systems and data architecture