Operational diagnosis for regulated sales floors

You pay a fortune for every lead.
Nobody can tell you why they didn’t buy.

We diagnose the sales floor, not the salesperson.

Your media budget buys the conversation. What happens inside it decides whether you ever get paid. Cognifai reads every conversation a customer has with you, however many people they speak to on the way, and tells you which line of your own process is costing you money. Not a score. A cause, located in a script, a system, a handover or a rota, quoted from the recording and priced against your own funnel.

Show me what leaked →

No migration. No IT project. Nothing for your agents to change.

100
conversations is all we need to start with. Every missed close, every dropped objection, every recoverable deal: named, timestamped and priced. In 7 days.
Get your leakage report ↓
<5%
of conversations reviewed manually by the average firm.
You paid for 100% of those leads.
McKinsey & Company, 2024
£0
of extra media spend required. The leads are already bought. The conversations already happened. The only question is what was said in them.
Scroll
Built for regulated sales floors that pay for their leads Insurance Claims Management Mortgages Consumer Finance Debt Solutions Wealth Management
Where It Leaks

The conversations you never hear
are where the money is.

Acquisition is the biggest line in your budget. Every pound of it is spent to create one thing: a conversation. And the entire return on that spend then rests on how that conversation goes, which nobody in management ever hears.

Fewer than 5% of those conversations get reviewed, and the few that do are scored for compliance rather than for money. The rest is recorded, stored, paid for twice, and never listened to by anyone. That is not a QA problem. It is a P&L problem.

A 100-conversation review surfaces every leak: the buying signal nobody picked up, the objection nobody handled, the renewal nobody defended, the close nobody attempted. Not modelled or estimated from an industry average. Specific calls, specific customers, each one timestamped. So your team can go and listen, and in a lot of cases pick up the phone and ring them back.

Your sales floor has more than 100 conversations before lunch.

We will not put a number on your firm until we have heard your conversations. You want your own figure, not an industry average, and you want to be able to check it call by call.

And the same pass tells you where the risk is.You are regulated whether you like it or not, and the identical analysis that finds a missed close also finds the exclusion that was not explained, the vulnerability signal that was not logged, and the card number read aloud onto a recording. You do not have to buy a second product for it, or run a second project. One pass over the data. Three answers.

For context: a Section 166 skilled person review typically costs a firm £50,000 to £200,000+, paid before the regulator reaches any conclusion. Insurers and brokers were billed £2.71m for eight such reviews in the year to March 2024 (FCA Freedom of Information disclosure).
What We Find

Four things your own reporting
cannot show you.

Findings from paid Cognifai engagements on live regulated sales floors, one of them a market leader. The ratios are real. Client, sector and revenue figures are withheld, because on a floor this specific they would identify the firm.

01
Your two scripts contradict each other
One asked whether the customer lived with a partner, because the credit file needed it, and never whether that person had to agree to anything. The decision maker was properly established on 8% of qualification calls. Fourteen advice calls then reached the fact-find before the customer said they could not decide alone, on average 48 minutes in. The second script opens by apologising for repeating the first.
02
The busiest outcome code was one nobody could define
The telephony export flagged 21 transfers. The recordings showed 30. We asked the provider what the most-used code in the operation meant and were told it belongs to the firm; the firm said it belonged to whoever configured it. Your reporting is not lying to you. It simply was not there for the conversation.
03
The word that costs you most is the one that is missing
The word documents did not appear in that firm’s qualification script. It was mentioned before handover on 8% of those calls, then chased on 76% of the advice calls that followed. Fees were mentioned on 4%, call length on 7%. A keyword tool cannot find an absence.
04
The thing you pay for twice is not the lead
17 of 100 qualification calls ended with a customer who was ready and no adviser free. Every one was given another time, so nothing recorded a failure. Six of the seventeen had already been rebooked once, and one says on the recording that it is the third. The lead is bought once. The qualification is what gets spent again.

The leak shows up in the sales call.
The cause is almost always upstream.

We read both halves: the qualification call and the call that follows it. Almost every expensive problem we find was decided before the adviser picked up, in a script line, a handover or a rota. None of them is a training problem, and none of them shows up in a report.

The Blind Spot

You optimise every step of the funnel
except the one where the money is actually decided.

BEFOREListening to 4 in 100
NOW REVIEWING
REC-2841.mp3
04:12 / 08:34
4calls reviewed today
Remaining95 unreviewed
Fewer than 5 in every 100 heard  ·  95+ paid for and never opened  ·  revenue lost: unknown
AFTERListening to 100 in 100
100%
CONVERSATIONS ANALYSED
REC-2842
REC-2843
REC-2844
REC-2845
REC-2846
REC-2847
PROCESSING
Recoverable revenue
Conversion coaching
Compliance flag
100% of calls and chats analysed  ·  every leak named and priced
The Second Reason

Worth doing for the revenue alone.
The regulator just made it unavoidable.

FCA Voluntary Requirements (VREQs) Secured Per Year
15
2019/20
21
2020/21
53
2021/22
82
2022/23
102
2023/24
15 → 102 in five years. Roughly 7× growth. The FCA's supervisory net is getting wider, not smaller.
Source: FCA interventions team data, obtained via Freedom of Information disclosure (as reported by Compliance Corylated): 15 voluntary outcomes (2019/20) rising to 102 (2023/24).
01
Evidence, not assumption
Under Consumer Duty, firms must demonstrate they are delivering good outcomes, not assume it. The FCA has been explicit that collecting data and reporting MI does not, by itself, show anything. Firms must show what the information told them, what they did, and whether it worked. That audit trail cannot be built from a 5% sample of your conversations.
02
Vulnerability, by driver
In July 2026 the FCA specifically criticised firms for aggregating vulnerability data rather than segmenting it by driver: health, financial resilience, life events, capability. Those drivers surface in conversation and almost nowhere else. A customer mentions a bereavement, a diagnosis, a redundancy. Structured data will never capture it. FCA, 27 July 2026
03
Sales practice scrutiny
The FCA is examining whether customers genuinely understand terms, exclusions, fees and risks before they buy. A keyword tool can hear a term mentioned. It cannot hear whether the customer understood it.
04
Continuous, not periodic
The FCA’s Mills Review (July 2026) describes firms moving beyond “periodic review, manual sampling or fixed workflow rules” towards continuous monitoring of customer interactions across every channel. The direction of travel is settled.
See what leaked out last month.
Get your leakage report.
→
What Cognifai Delivers

One pass over your conversations.
Three answers. Every call. Every chat.

Recovery
The revenue is already in your conversations. You just can’t hear it.
Every buying signal your closer walked past. Every objection that was never handled. Every quote that went cold because nobody followed up, and every cancellation where the customer could have been saved and nobody tried. Surfaced, categorised, priced and routed to the people who can act on it, while the customer is still warm enough to ring back.
Conversion
Your best closer and your worst one are working the same leads.
The gap between them is worth more than any media optimisation you will run this year, and it is made of things neither of them can see. Cognifai scores every conversation, so coaching is grounded in evidence rather than the four calls a QA manager got to this week. Fair, specific, auditable. And it drags your average closer towards your best one, on leads you have already bought.
Risk
And find the problems in week one, not in a regulator’s letter.
Compliance issues flagged, timestamped, attributed and evidenced against the calls they came from. A firm that spots a systemic sales practice failure in week one fixes it in week two. The alternative is finding out from someone else, much later, with a lot less control.
Ask Better Questions

Stop sampling calls at random.
Start asking where the money went.

A keyword tool only finds what you told it to look for. The deal you lost is almost never the thing anybody thought to search for. Cognifai understands meaning and context across every conversation, voice and chat, so you can interrogate all of them in plain English, and get an answer with the calls behind it:

→“Show me every lead we lost where the customer was ready to buy.”
Revenue
→“Find every call that ended without anyone asking for the sale.”
Revenue
→“Which closers are losing deals at the price objection, and what are they saying?”
Conversion
→“Show me every conversation where a customer showed signs of vulnerability.”
Risk

Legacy speech analytics can't answer a single one of these. It matches words against keyword libraries you spend months building, then rebuilding every time a product, script or campaign changes. A phrase-spotter can hear a product mentioned; it cannot hear whether anyone actually offered it, or why the customer said no. Cognifai reads meaning, not word lists. Nothing to build. Nothing to retune.

Scoring tools have the opposite problem. Including the new ones with a language model behind them, they grade the last call and hand you a number per agent. That number goes to whoever owns quality, gets argued about, and changes nothing, because whatever caused it was sitting in a script or a rota that nobody scored. One answers a question about vocabulary. The other answers a question about a person. Neither answers a question about a process, and a process cannot be coached.

The FCA’s Mills Review (July 2026) describes firms moving beyond “periodic review, manual sampling or fixed workflow rules” towards continuous monitoring of customer interactions, on every channel.

The Leakage Report

This is what lands
in your inbox.

Value demonstration. A full report drawn from 100 of your own conversations: calls, chat transcripts or a mix. Every pound of leaked revenue named and sourced back to a specific call. Every closer scored. Every compliance flag evidenced.

Figures below are from a real review, anonymised. The pound values are withheld: the sector is small enough that they would identify the firm. Every ratio is real. A live report carries your own numbers, traced call by call.

19%
Agree to go ahead, then never convertOf everyone who says yes on the call. The selling is done, the fact-find is done, the adviser time is spent. Priced in the live report against the firm’s own funnel data.
76%
Of calls spent chasing paperworkPortal logins, payslips, bank statements. Hundreds of agent hours a month, and the single most common activity on the calls, more common than any objection.
3.7/5
Closing: the weakest agent scoreBehind rapport, fact-find depth and product explanation. The conversations are good. The last ninety seconds are not.
39%
Of calls ended unresolvedAn agreement with no date attached, a question left hanging, an objection nobody answered. Ten worth starting with, each with the timestamp to jump to and what to listen for.
Pages from a Cognifai review: the customer's own words, and the diagnosis with the fix beside it
The Chat Leakage Report

Your customers told the bot everything.
The bot told nobody.

The same review, run on the channel most firms never look at. Webchat leaves a perfect written record and almost nobody reads it. Every buying signal quoted and priced. Every vulnerable customer identified. Every unlogged complaint surfaced. The figures below are illustrative. They show the format, depth and evidence standard of a chat review, not findings from a live one.

£16.9k
Revenue signals, in 100 conversations43 conversations carried a priced signal; sixteen asked outright to buy. Each priced at the firm’s own premium for the product discussed. That is £169 per conversation reviewed.
14
Vulnerable customers unflaggedBereavement, critical illness and financial difficulty, all disclosed in writing.
6
Complaints never loggedComplaint language in seven conversations. Six appear nowhere in DISP records or root-cause MI.
35%
Bot containment rateAll simple lookups. Escalation succeeded 3 times in 13 attempts.
ILLUSTRATIVE REVIEW · 100-CONVERSATION FORMAT
Illustrative review  ·  UK personal-lines scenario  ·  webchat & WhatsApp  ·  AI assistant, human handover 9-to-5
Chat & AI Bot
Analytics
£16.9k
Revenue
Signals
14
Vulnerable,
0 Flagged
6
Complaints
Unlogged
35%
Bot
Containment
⚑ Revenue: Asked To Be Called. Never Was.
One customer, three chats, five weeks. He buys a second car and asks to be put through to someone. The bot logs a callback “within 2 working days” for a vehicle he cannot legally drive. He comes back; the bot confirms twice that it cannot see the status of callback requests, then logs a second one. He returns a third time: “Ive insured the second car elsewhere and Ill move this one across at renewal.” The bot offers him a third callback. Two unactioned callbacks: £520 of new business lost, plus £470 of existing premium now flagged to lapse, with eight months’ written warning nobody will read until the lapse report.
Vulnerability & Conduct: Top Findings
01
The Assistant Does Not Recognise The Word "Died"
A recently widowed customer asked to remove her late husband from the policy. First disclosure returned a fallback message, the second a phone number. No condolence, no bereavement process, no flag.
Critical
02
Confident, Verifiably False Answers
Where retrieval came up short the model answered anyway: an excess quoted at £250 against a schedule that says £500, a cooling-off period that does not apply, an add-on the firm does not sell.
Critical
03
Complaints Invisible To The Complaints Process
"I want to make a formal complaint" hit the fallback message verbatim. Six of seven were never logged anywhere, absent from DISP records and root-cause MI.
Critical
04
23 Material Disclosures Carried Nowhere
Undeclared courier work, a lodger in residence, unspecified convictions, a restricted dog breed. The firm holds written notice of facts it does not know it has been told.
High Risk
Transcript Evidence · Verbatim
Customer · CHT-0011I need to take my husbands name off the policy. He passed away in April.
AssistantI'm sorry, I didn't quite get that. You can try rephrasing, or type AGENT to speak to the team.
Customer · CHT-0023callback please, ideally today, the car is sat on the drive uninsured
Bot Performance Scorecard · Measured, Not Sampled
Containment
35%
All simple lookups & document resends
Escalation
3/13
8 queue abandonments
Callbacks
0/8
No status visibility, no SLA tracking
Complaints
0/7
No complaints intent exists
Missed Revenue · 100 Conversations
Signal Type
Found
Est. Premium
Explicit Product Requests, No Route To Sale
16 chats
£6,740
Renewals Lost With No Save Attempt
11 chats
£5,280
Life-Event & Cross-Sell Signals Ignored
16 chats
£3,040
The Callback Black Hole
8 chats
£1,180
Referral & Leakage Lines
£610
Total, 100 conversations
£16,850
Basis for the illustration: each signal priced at an average annual premium for the product discussed. In a live review the figures are your own. One hundred conversations, one month, one channel. We have not multiplied it across the book: mix changes by product and season, and a sample is a sample. Where a signal could not be priced honestly it was left unpriced rather than guessed at. What it does tell you is that one hundred chats held £16,850, and a sales floor handles rather more than one hundred.
Harrison, Founder of Cognifai Intelligence
Built by an operator

Built by someone who ran these sales floors, and paid for the leads.

Cognifai didn't start in a lab. Before founding it, Harrison spent years building and running regulated contact centres, living the exact problem the product now solves: signing off enormous lead invoices every month, then having no idea what was actually said on the calls those leads turned into. The quiet certainty that the next deal, and the next breach, were both sitting in a recording nobody had time to open.

He then spent six years founding, scaling and exiting an international executive search business, placing leaders into some of the fastest-growing technology companies in the world. This venture taught him what most software vendors never learn: that a contact centre is a revenue engine first and a compliance obligation second, and that any tool which forgets that gets switched off within a year.

"I've sat in your chair. I'd paid the huge web traffic invoices and I know exactly what is hiding in the calls nobody listens to."

Cognifai is the ‘always-on’ sales management system he wished he'd had. Every conversation heard. Every pound of leakage surfaced. Every risk visible, long before anyone else finds it.

Harrison Scott
Founder, Cognifai Intelligence
What Happens After Your Report

The report is the proof.
100% of your calls & chats is the product.

The review reads 100 conversations. It exists so you can see what is in them, on your own data and in your own customers' words, before you commit to anything. What follows is the same analysis running across every call and every chat you handle. The connection is already made: the read-only API access that produced your review simply keeps running, across every call instead of a hundred, with chat transcripts on the same cycle. Not a 5% QA sample. Not a keyword library for someone to maintain. No telephony migration, no new software for your agents, no engineering resource pulled from your side. Your team keeps working exactly as they do today.

01
Connect
Nothing new to set up. The read-only API access from your review stays in place and widens to your full call volume, with chat transcripts exported on an agreed cycle.
02
Analyse
Every conversation assessed against the same framework as your review: revenue, vulnerability, complaints, conduct and disclosure. 100% coverage, not the 5% a QA rota reaches, and nothing for your team to do.
03
Act
Recoverable revenue, conversion coaching and compliance flags land with the people who can act on them, with the evidence trail Consumer Duty expects you to be able to produce.

Platform access, reporting cycle and pricing are agreed at your findings call, against your own conversation volumes and the systems you run. Nothing is committed before you have read the review and validated every number in it. We would rather prove it on your data than ask you to take it on trust.

The Revenue Leakage Report

Start with 100 conversations.
We’ll show you what last month cost you.

No keyword libraries to build. No migration. No new software for your agents. Nothing for your team to do. Calls are retrieved by read-only API, chat transcripts by simple export. We take 100 conversations and handle the rest. Full findings within 7 days.

You’ll see the revenue your team walked past, which closers are losing the leads you paid for and where, and anything you’d want to know about before someone else finds it. For the first time you will know what your acquisition spend is actually buying.

The report is where it starts. From there, Cognifai becomes your always-on layer, with every call and chat analysed continuously, so the next recoverable deal surfaces while the customer is still warm.

1
30-min scoping call
2
NDA & DPA signed
3
Calls via read-only API · chats by export
4
Full findings in 7 days
5
Data feedback session

Request your Revenue Leakage Report

Tell us about your firm and we'll be in touch within 24 hours to scope it. NDA signed before any recordings are shared. Scope and fee agreed on that call, with no obligation to go further.

All conversation data deleted within 30 days of report delivery.
Strict confidentiality, governed by a signed Data Processing Agreement (DPA).
Transcribed in the EU. Analysed in AWS eu-west-2 (London). Never used to train AI models.
Customer names are never reproduced in the report. Calls are referenced by timestamp so your team can locate them.
Scope and fee are agreed at your scoping call, before any work begins. The review is a standalone engagement, with no obligation to move forward.
Something went wrong. Please email info@cognifai.uk directly.

Request received.

Thanks, your request is in. A member of the team will be in touch within 24 hours to book your 30-minute scoping call and get the report underway. Need us sooner? Email info@cognifai.uk directly.