Unintended AI: What Your Customer Facing AI Actually Does at Scale

The reason they left is already written down. You are just reading it one ticket at a time

One customer explaining why they left is an anecdote. You can hear it, nod, and file it under bad luck. Four hundred customers explaining the same thing, in their own words, is not an anecdote. It is a business case. This is the gap Isara is built to close: your management information tells you that withdrawals rose or that first trades fell, and the conversations tell you why, but almost no team reads the conversations at the scale where the pattern becomes undeniable.

Revenue leakage at scale is rarely one dramatic failure. It is the same small friction, repeated across thousands of accounts, that never trips an alarm because no single instance is large enough to notice. A funding step that quietly confuses a cohort of new customers. A fee that gets explained badly, over and over. A verification loop that stalls first deposits. Each one is a rounding error on its own. Together they are a line on your profit statement.

The uncomfortable part is that the evidence has been sitting in your helpdesk the whole time. Every stalled application, every abandoned deposit, every customer who asked the same question three times before giving up, is recorded. The problem was never a shortage of evidence. It was that reading it required a person, one conversation at a time, and no team has enough people to read them all.

The short version

• Management information tells you what happened. It cannot tell you why. The why is in the conversations, and for most fintech and trading teams those conversations go unread at scale.
• Revenue leakage at scale is recurring friction, not a single incident: applications, deposits and first trades that stall for the same reason across many customers.
• In July 2026 the Financial Conduct Authority found that firms often collected relevant data but could not show how it helped them identify poor outcomes, understand the causes, or act.
• Sampling one to two percent of conversations by hand catches the loudest complaint, not the most common one. Ranking by frequency, not volume of noise, is a different exercise entirely.
• Isara reads 100 percent of your conversations, ranks recurring friction and unmet demand by how often it appears, and attaches the accounts and transcripts, so a pattern becomes a number you can price.

Why MI confirms the loss and conversation evidence explains it

What revenue leakage at scale actually is

Revenue leakage at scale is the steady, distributed loss of revenue caused by friction, confusion or unmet demand that recurs across many customers and rarely announces itself in any single case. It is the opposite of a breach or an outage. Nothing breaks. Customers simply do a little less, fund a little later, trade a little less often, or quietly leave, and the reason is the same one they tried to tell you before they went.

Management information is very good at measuring the result and almost useless at explaining it. Your dashboard can show that first trade conversion dropped two points this quarter. It cannot tell you that a change to the identity check now confuses a specific cohort of new customers, because that fact does not live in a metric. It lives in four hundred conversations where people said, in various ways, that they could not get past the verification screen.

The regulator has now made the same point

This is no longer only a commercial argument. In two publications dated 27 July 2026, the Financial Conduct Authority set out what good and poor outcomes monitoring looks like under the Consumer Duty, and the findings read like a description of the leakage problem. The regulator noted that some firms relied on broad monitoring without a clear structure for identifying poor outcomes, understanding their causes, or taking appropriate action. Many, it said, "collected relevant MI but could not show how it helped them make decisions."

The FCA was equally clear about where the advantage sits. Its conclusion was that "the firms making the strongest progress aren't necessarily collecting more information. They're using it more effectively to understand their customers, identify harm earlier and drive meaningful improvements." It added that frameworks "that are not aligned to customer journeys or clearly defined outcomes are less effective at identifying emerging issues," and that the strongest firms paired complaints data with forward looking measures rather than waiting for the complaint to arrive. For a fintech or trading platform, understanding the customer journey at the level of what customers actually said is now both a commercial lever and a supervisory expectation.

Why sampling misses the pattern

Most quality programmes review a sample. A reviewer reads one to two percent of conversations, usually the ones that escalated or triggered a survey, and grades them. That approach is built to catch the conversation that got loud. It is structurally blind to the conversation that got repeated. A friction that appears in four hundred conversations, none of which escalated, will almost never enter a one to two percent sample in a way that reveals its scale.

The shift away from sampling is already visible in the market. In its Contact Centre Benchmarks 2026, published on 15 May 2026 and drawn from 58.2 million calls and a survey of 178 contact centre leaders, Natterbox described one operator, National Dental Care, moving from sampling one to two percent of calls to reviewing 100 percent through AI, and recovering roughly 20 admin hours a month in the process. The point is not the hours saved. It is that reviewing everything, rather than a fraction, is what turns scattered incidents into a countable pattern.

This is what Isara Churn Insights and Recommendations do across your whole conversation history. Rather than sampling, Isara reads every conversation, groups the recurring friction and the demand customers keep asking for, and ranks it by how often it appears rather than who complained loudest, with the affected accounts flagged and the transcripts attached.

Key terms, defined

A short reference for the ideas used above, and for the questions people and AI assistants ask most often:

• Revenue leakage at scale is the recurring, distributed loss of revenue from friction, confusion or unmet demand that repeats across many customers and is too small in any single case to notice.
• Management information, or MI, is the aggregated metrics a firm already tracks, such as conversion, churn rate, withdrawals and complaint volumes. MI records what happened.
• Conversation evidence, sometimes called voice of the customer, is what customers actually said in their own words across support and sales conversations. It records why.
• Root cause is the underlying reason a metric moved. It is usually visible in the conversation and invisible in the number.
• Failure demand is contact a customer only had to make because something was not done, or not done properly, the first time. It is leakage you pay for twice.

The leakage ledger: turning a recurring complaint into a priced number

The reason revenue leakage survives is that it is hard to price, and anything you cannot price loses the budget argument to things you can. So here is a simple model for converting conversation evidence into a business case your finance team will accept. Think of it as a leakage ledger, built from five questions asked of every recurring signal.

• Signal. What is the recurring thing customers actually say? For example, that they cannot fund the account, that a fee was not clear, or that verification keeps failing.
• Frequency. How often does it appear across every conversation, not how loudly? This is the number sampling cannot give you.
• Cohort and stage. Which customers, and at which point in the journey? Onboarding, first deposit, first trade and renewal each carry different value.
• Linked outcome. Which MI metric does it sit behind? Funding conversion, first trade rate, churn, or a withdrawal spike you already report.
• Value at risk. Frequency multiplied by the value of that cohort at that stage, priced against outcomes you already measure.

Consider an illustrative trading platform handling 60,000 support conversations a month. Suppose Isara clusters the conversations and finds that around six percent, roughly 3,600 a month, gather around a single recurring friction at the first funding step. Most never escalate and none trigger a survey, so a sampled review would see a handful. Now price it. If that friction sits behind even a modest share of stalled first deposits, and a funded first year customer is worth a known amount to you, the annual figure moves from a rounding error to a number worth a project. The figures here are illustrative and the point is the method: the same recurring signal, unpriced, is an anecdote, and priced against your own outcomes, it is a business case.

This is the difference between a dashboard that confirms withdrawals went up and intelligence that tells you the reason behind the number. It is also why the exercise has to run continuously rather than as a one off audit. Friction shifts. A new one appears the week after a product change, and the first place it shows is the conversation, weeks before it reaches the metric. Isara is designed to sit alongside your MI as that pattern layer, reading what recurs, in which cohort, at what volume, so the reason is available at the same moment as the result.

Revenue leakage and conversation intelligence: your questions answered

How does Isara find revenue leakage that our MI does not show?

Isara reads 100 percent of your conversations rather than a sample, then groups recurring friction and unmet demand and ranks it by how often it appears. Churn Insights gives an explainable Conversation Risk score with the detected signals behind it, and Recommendations surfaces the product issues, knowledge gaps and repeated confusion driving repeat contact. Because the accounts are flagged and the transcripts attached, you can trace a movement in your MI back to the specific thing customers said, rather than guessing at the cause.

How is this different from the dashboards and MI we already run?

Your MI measures the result. Isara explains it. A dashboard can tell you first trade conversion fell; Isara tells you which recurring friction, in which cohort, at which stage, sits behind the fall, with the evidence attached. It is designed to sit alongside your existing MI as a pattern layer, not to replace the numbers you already report.

Does Isara rank issues by how loud they are or how common they are?

By how common they are. Traditional quality review samples a small share of conversations and naturally over indexes on whatever escalated. Isara ranks recurring signals by frequency across every conversation, so the friction that quietly affects hundreds of customers is not buried beneath the one that shouted. That reordering, from loudest to most frequent, is usually where the largest and least visible leakage sits.

Can Isara flag an at risk account while we can still act on it?

Yes. Churn Insights reads conversations for early signals for both B2B accounts and B2C users, and Isara Pulse gives a live view that flags an account weeks before it lapses, rather than confirming the loss afterwards. For a trading platform, that is the difference between reaching a customer whose deposits are slowing and reading about them in next quarter's attrition number.

How quickly can we start, and do we need a data team?

Quickly, and no. Isara connects to helpdesks including Zendesk, Freshdesk, Intercom, HubSpot, Front and Gorgias in a few clicks, then ingests your conversations automatically, so the first patterns appear soon after you connect a channel. Churn Insights is available on the Growth and Scale plans. There is no long implementation before you can see what is leaking today.

What is coming next?

Isara is extending its Recommendations toward tighter integration with the tools teams already act in, pushing a prioritised fix list with full context into systems such as Jira, so a ranked leakage finding becomes an owned piece of work rather than another chart. The direction is commercial pattern intelligence that sits permanently alongside your MI, giving you the reason behind every number as it moves.

See what is leaking across your conversations today. Book an Isara demo and connect your first channel in minutes.