Last month, I had a problem with a popular mobile banking app in Southeast Asia. Nothing exotic. A transaction didn't go through, and my support ticket had been sitting untouched for two weeks.
So I opened the app's chatbot. It greeted me warmly, asked how it could help, and then couldn't do a single useful thing. It couldn't look up my transaction. It couldn't check the status of my ticket. It couldn't tell me why my issue was unresolved. It could answer FAQ questions, and that was it.
I called the hotline instead. Spent an hour navigating prompts, got bounced between menus, and every path ended the same way: "Please contact our chatbot or check your existing ticket." The system was built for deflection, not resolution. The ticket that nobody had touched for fourteen days.
I gave up. And somewhere in that company's dashboard, my interaction counted as a successful AI chatbot deflection.
The uncomfortable part: if you shipped a deflection-optimized bot this quarter, a customer somewhere is living this exact loop right now. Your dashboard is calling it a win.
The Deflection Metric Everyone Loves (and Nobody Questions)
Deflection rate measures the percentage of customer contacts handled without a human agent. It's cheap to track, easy to celebrate, and it maps directly to cost savings. Industry benchmarks citing McKinsey's 2026 service operations data put AI resolutions at $0.62 per ticket versus $7.40 for human agents. That's a 12x cost difference. Of course executives love this number.
But deflection doesn't measure whether the customer's problem got solved. It measures whether the customer stopped asking. Those are very different things.
This is Goodhart's Law applied to customer experience: when a measure becomes a target, it ceases to be a good measure. Deflection is cheap and easy to optimize. Resolution is hard and expensive to track. So companies optimize the proxy and stop looking at the goal.
Gartner data, as reported by Forbes, confirms the gap: only 14% of customer issues are fully resolved through self-service channels. Even for the simplest cases, that number climbs to just 36%. Meanwhile, companies report deflection rates north of 60% and call it progress.
What It Actually Costs
The financial damage is already showing up. A Laivly study published in June 2026 found that 28% of leaders said AI had directly contributed to lost revenue because it couldn't handle complicated support issues. Another 20% knew there was lost revenue but couldn't even quantify it. Nearly half of leaders are aware their AI is costing them money, and many can't tell you how much.
The market is already correcting. A Sinch survey of 2,527 senior decision-makers found that 74% of enterprises have rolled back or shut down a customer-facing AI agent after deployment. The rollback rate was highest, 81%, among organizations with the most mature AI governance. The companies monitoring most closely found the most problems. The ones not monitoring? They're still celebrating.
PwC's Consumer Intelligence Survey, cited in the same Forbes analysis, puts the downstream cost plainly: 44% of customers stopped buying from a company entirely after a trust breakdown. That revenue leaves quietly, with no complaint and no exit survey that ever reaches leadership.
What That Bot Was Actually Missing
I build agentic systems. Not customer-facing support bots, but the same architectural components that make any AI agent useful instead of decorative: record lookup, scoped write access, escalation gates, anti-fabrication checks. These components work the same whether the agent serves you or your customer.
The chatbot I dealt with had none of them. What frustrates me as a builder: the fix isn't a bigger model or a more expensive API. It's basic architecture.
The bot knew who I was. I was logged in. It had my account ID, my phone number, my transaction history sitting in a database somewhere. But nobody gave it permission to read any of that, wired it to the ticketing system, or built an escalation rule that said "if a ticket has been open for 14 days with no response, flag it."
That's not an AI problem. That's a permissions and plumbing problem.
What a properly wired support agent actually does
| What the customer needs | FAQ-skin chatbot | Properly wired AI agent |
|---|---|---|
| Recognize me | Generic greeting, asks for details I already entered | Auto-lookup: I'm logged in, it already knows my name, account, and open tickets |
| Check my records | "Please visit our help center" | Pulls my transaction history, sees the failed payment, checks my ticket status |
| Take a basic action | Can't | Adds a follow-up note to my ticket, triggers a callback request, updates my case priority |
| Read my frustration | Oblivious, keeps looping through FAQ scripts | Sentiment detection: my second message is sharper than my first, route to a human now |
| Escalate without losing context | "Let me transfer you" (I start over from scratch) | Human agent inherits the full thread, my account state, and what the bot already tried |
| Enforce SLA | Ticket sits for two weeks in silence | Auto-escalate: no agent response in 48 hours triggers a supervisor notification |
| Feel like a real conversation | Corporate, stale, robotic | Warm, direct, culturally aware persona that doesn't read like a terms-of-service document |
None of this requires a frontier model. A well-structured agent with read access to CRM records, write access to ticket notes, and three escalation rules would have resolved my issue in two minutes. The technology exists. The architecture and permissions don't.
If you want the general-purpose test for whether an AI tool is actually intelligent or just conversational, I wrote a framework for that. What matters here is the customer-facing version: can your bot take an action on behalf of a specific customer, or can it only describe what a customer should do themselves?
Where Chatbots Actually Work
This isn't an anti-AI argument. Bots are genuinely excellent at killing volume that should never have been a ticket in the first place. Password resets, account balance checks, delivery tracking, store hours, return policies. Deterministic, high-volume, low-context requests where the answer is the same regardless of who's asking.
The failure boundary is precise: bots break when the answer depends on your specific data, your specific history, or your emotional state. A well-scoped bot that handles the first category and honestly hands off the second is a good system. A bot that pretends to handle both and quietly counts the frustrated customer as "deflected" is a liability dressed as a KPI.
The Philippine Irony
There's a specific irony here for the Philippines. This is the country that became the back office of the world by being better at customer empathy than anyone else. Filipino agents built a global BPO industry on patience, warmth, and actually solving the problem.
Now Philippine banks and telcos are deploying deflection bots that strip out the one thing the BPO industry proved Filipinos do best.
A Twilio/YouGov survey of 7,331 adults across seven Asia-Pacific markets captured the tension: 90% of Filipino respondents said their society values patience and politeness in daily interactions. But only 32% said they have any patience left for automated customer service. Filipinos become more frustrated than their Asia-Pacific peers when AI systems give scripted or robotic responses.
The Philippine AI Report 2025 found that 92% of Philippine organizations have experimented with AI, but 65% remain stuck at proof-of-concept. The gap between experimenting with AI and wiring it into the systems that matter is where the deflection illusion lives. The technology arrived. The integration didn't.
The Monday Morning Test
Pull your chatbot's deflection rate. Now try to pull its resolution rate. If your dashboard can't show you the second number, you already have your answer.
The fix isn't bigger models or more sophisticated NLP. It's plumbing. Give your bot read access to customer records. Give it write access to ticket notes. Build three escalation rules: sentiment spike, repeated question, SLA breach. That's the difference between a support agent and a search bar with a personality.
A sticker on a broken car is still a broken car.
Originally published at tokita.online.
Related reading:
Top comments (0)