Every customer has a hidden score that decides how well they get treated, and most businesses have never interrogated how it’s calculated. Hiba Hassan argues that AI in customer service is being aimed at deflection when the same capability, pointed one step upstream, would expose how often companies are charging customers for their own defects.
You have a score you have never seen. Not a credit score, something less regulated and far more consequential to how you get treated: what you cost the companies you buy from. It decides whether you reach a competent human in ninety seconds or a queue designed to outlast your patience. Nobody will show it to you, and in most businesses nobody senior has interrogated how it is calculated either.
That score is how firms fire customers, which they do constantly and almost never in writing. They route them to the slowest queue, quietly raise their fees, stop offering them anything, and wait for them to leave. The polite term is portfolio optimisation. The honest term is cost escalation.
The logic is not stupid. Some customers genuinely cost more to serve than they will ever return, and a business that refuses to admit that is being sentimental with someone else’s money. I have no argument with the principle.
My argument is with the number, and with what we have just handed it.
The Number That Hides Two Different Customers
Cost to serve bundles support contacts, returns, discounts, and handling into one line per customer. Cross it with revenue and the customer sorts into keep, grow, or quietly discourage. It is clean, defensible in a board meeting, and it hides two completely different things. Some customers are expensive because of who they are. Others are expensive because your systems are failing them. The person calling every week may be exploiting your returns policy, or may be reacting to a checkout that silently drops their basket and an app that logs them out on every update. Same ticket volume, same cost line, opposite customer. The model treats them identically, so you escalate the cost, they leave angry, and you record it as a successful optimisation.
Now watch what artificial intelligence is actually being deployed to do about this, because it is not what the sales deck says.
Success, Defined As The Customer Not Getting Through
Look at how support AI is sold and measured. The headline metrics are deflection and containment: what share of contacts were prevented from reaching a person. We have built an entire category whose definition of success is the customer not getting through. That is cost escalation with a friendlier interface and better margins, and it scales in a way a call centre never could.
Then follow the loop it creates. Your system has a defect. The customer contacts you. The bot, trained on documentation rather than on the defect, fails to resolve it. The customer contacts you again, and again, because the problem has not gone anywhere. Every one of those contacts lands in their cost-to-serve record. The score rises. The higher score routes them to a cheaper tier with less human access, which lowers the chance of resolution, which produces more contacts. Eventually the model declares them unprofitable and the business quietly shows them the door, having charged them for its own bug and then punished them for complaining about it. What used to take a year of institutional neglect now compounds in a quarter.
There is a quieter problem underneath. The AI increasingly writes the case summary that gets logged, categorised, and counted. The system that failed is also the witness. If the summary says “customer contacted repeatedly about login” rather than “authentication service dropped sessions after the March release,” then the defect never appears in the data, and the customer looks like the problem in every report anyone will ever read.
“The system that failed is also the witness.”
The Gain Is Real, But It Is Not Intelligence. It Is Reach.
None of this is an argument against the technology, and I want to be precise here, because the category is drowning in vagueness. The gain is real, but it is not intelligence. It is reach. A system can read what humans structurally cannot get to: complaint text, call transcripts, chat logs, session recordings, the unstructured evidence that explains why a customer is expensive rather than just recording that they are. Root cause was always sitting in that material. Nobody had the hours to find it. Everyone can buy the same models, so the tooling is not the advantage. What you point it at is. Most firms are pointing it at deflection when the same capability, aimed one step upstream, would do diagnosis.
The Gulf version Of The Problem
The regional version is the one I find most instructive. In the Gulf, cash on delivery is still a substantial share of e-commerce, and it generates real cost: failed deliveries, redelivery attempts, returns that never leave the doorstep. Easy to code those buyers as expensive and worth discouraging. But most are not choosing cash on delivery to inconvenience your logistics partner. They are choosing it because your checkout did not earn enough trust for them to hand over a card. That is not a bad customer. That is a trust problem wearing a cost-to-serve costume, and no amount of deflection will fix it.
“That is a trust problem wearing a cost-to-serve costume, and no amount of deflection will fix it.”
The Fix Is Cheaper Than Any Platform
So the practical change is cheaper than any platform. Before a customer is classified as unprofitable, split their cost. Code the tickets by root cause, product failure on one side, customer behaviour on the other, and check that coding against your own error and uptime logs, because what a customer says broke and what your system recorded breaking are not always the same thing. Then look at the ratio. My expectation, and I would like to be tested on it, is that much of the unprofitable segment is expensive because of defects the business pays for twice, once in support cost and once in churn.
The uncomfortable part is not the analysis, it is the accounting. Splitting cost to serve tells you exactly how much of your customer service budget is really an engineering invoice. Most organisations already suspect the answer. Very few want it itemised, and none of them want a machine itemising it in real time.
Hiba Hassan is a Marketing Expert and AI Adoption Strategist. A neurodivergent leader, she translates emerging tech and inclusive values into high-impact strategies.
Disclaimer: This article was originally written by the author. Views expressed are the author’s own. All rights belong to the original author.
- EMIR and Omnicom Media MENA Partner to Amplify GCC Thought Leadership - September 9, 2026
- We Are Now Using AI To Fire Customers For Our Own Bugs - September 8, 2026
- Initiative Appoints Darine Elkaissi as Managing Director in Saudi Arabia - September 2, 2026
