AI for financial services todeliver better customer outcomes.reduce complaints.prevent customer harm.keep more customers.cut cost-to-serve.prove fair treatment.

Q: Why are customers leaving, and who is winning them?

SentiSum reads every customer conversation across calls, chats, emails, complaints, reviews and AI-assistant chats, and ranks the drivers of churn by cost. Kyo, the SentiSum AI analyst, answers in plain English with the recommended fix, the team that owns it and the source conversations behind it.

Q: Are we treating every customer fairly, and can we prove it?

SentiSum scores every human and AI conversation against your own procedures, including vulnerability procedures, rather than a sample. Every finding links to the conversation it came from, and each fix is tracked before and after, so banks and insurers can show regulators and auditors that customer outcomes improved.

Q: Why are customers contacting us at all?

SentiSum finds the biggest sources of avoidable contact across every channel, ranks them by cost and gives each fix to the team that owns it. It then tracks repeat contacts and cost-to-serve until the number moves.

Q: What are our AI assistants telling customers?

SentiSum checks AI-assistant conversations against your current policies and procedures. It flags where an assistant quotes out-of-date information, refuses a hand-off to a person or fails to log a complaint, so human and AI conversations are held to the same standard.

Q: Where are complaints coming from, and how do we stop them?

SentiSum links complaints to the conversations that came before them, traces each one to its root cause and ranks drivers by cost, redress and volume. For a top-five UK bank, SentiSum read a year of complaints and traced more than £10M of redress to its root causes.

SentiSum reads every customer conversation against your own policies and procedures. It finds why the same problems keep happening, gives each fix to the team that owns it, and tracks it until the number moves.

See week one on your data Watch how it works

ComplaintsUK · Retail banking · flagged week 1

Direct debit cancellations promised on calls are not being actioned.

214 calls miss a step in the cancellation procedure. Complaints likely within weeks.

Recommended fix

Add a cancellation check to call wrap-up

Owner

Payments Operations

Merchant churnPayments · Merchants · flagged week 1

Merchants are leaving after payout delays.

Payout-delay contacts tripled in three weeks. Merchants who raised one are twice as likely to leave.

Recommended fix

Alert affected merchants before the payout is late

Owner

Merchant Operations

Cost-to-serveGermany · Insurance · flagged week 1

Customers call back because renewal letters are unclear.

Repeat contacts doubled in two weeks. Same paragraph quoted in most calls.

Recommended fix

Rewrite paragraph 3 of the renewal letter

Owner

Renewals, Germany

AI assistantAll markets · flagged week 1

The assistant is quoting an out-of-date fee.

Answer conflicts with the current pricing policy in 1,140 conversations.

Recommended fix

Update the fee source the assistant draws on

Owner

Conversational AI

RetentionItaly · Insurance · flagged week 1

New single-policy customers are leaving after their first claim.

Recommendation falling against the main rival in this segment. Claims callbacks promised and missed.

Recommended fix

Set a 24-hour callback rule for first claims

Owner

Claims, Italy

Fair treatmentUK · Digital banking · flagged week 1

Customers in financial difficulty are missing out on support options.

Vulnerability signals in 312 in-app chats this week with no support options offered.

Recommended fix

Add a support prompt to collections chat replies

Owner

Collections

Illustrative findings from week one

What it delivers

Ask Kyo the questions your board asks.

Kyo is the SentiSum AI analyst. Ask in plain English, inside SentiSum or from Copilot, ChatGPT or Claude. Every answer comes back with the fix, the team that owns it and the conversations behind it.

Try a question

01Why are customers leaving, and who is winning them?

02Are we treating every customer fairly, and can we prove it?

03Why are customers contacting us at all?

04What are our AI assistants telling customers?

05Where are complaints coming from, and how do we stop them?

Kyo · SentiSum AI analyst

Illustrative

Why are customers leaving, and who is winning them?

Kyo is reading 62,480 conversations across calls, chats, complaints and reviews, last 90 days

✓ 62,480 conversations across calls, chats, complaints and reviews, last 90 days

Churn is concentrated in new single-policy customers in Italy, straight after their first claim.

01Claims callback promised and missed38% of churn mentions
02Excess higher than the customer expected21%
03Cheaper rival quote at renewal17%

Recommended fix

Set a 24-hour callback rule for first claims

Owner

Claims, Italy

Moves

Retention, recommendation

View 1,326 source conversationsAssign to Claims, Italy

Are we treating every customer fairly, and can we prove it?

Kyo is reading 41,900 calls and chats scored against your vulnerability procedure

✓ 41,900 calls and chats scored against your vulnerability procedure

312 in-app chats this week showed signs of financial difficulty with no support options offered.

01Support options not offered312 chats
02Payment plan offered too late96 chats
03Vulnerability not recorded on the account58 calls

Recommended fix

Add a support prompt to collections chat replies

Owner

Collections

Moves

Fair treatment, conduct

View 466 source conversationsAssign to Collections

Why are customers contacting us at all?

Kyo is reading 128,300 contacts across every channel, last 30 days

✓ 128,300 contacts across every channel, last 30 days

Unclear renewal letters are the biggest source of avoidable contact in Germany.

01Renewal letter, paragraph 3doubled in 2 weeks
02Card replacement status11% of contacts
03App login resets9% of contacts

Recommended fix

Rewrite paragraph 3 of the renewal letter

Owner

Renewals, Germany

Moves

Cost-to-serve, repeat contacts

View 4,812 source conversationsAssign to Renewals, Germany

What are our AI assistants telling customers?

Kyo is reading 22,700 AI-assistant conversations checked against current policy

✓ 22,700 AI-assistant conversations checked against current policy

The assistant is quoting an out-of-date fee in 1,140 conversations.

01Out-of-date overdraft fee quoted1,140 chats
02Hand-off to a person refused410 chats
03Complaint raised but not logged185 chats

Recommended fix

Update the fee source the assistant draws on

Owner

Conversational AI

Moves

Conduct across human and AI

View 1,735 source conversationsAssign to Conversational AI

Where are complaints coming from, and how do we stop them?

Kyo is reading 9,840 complaints linked to the conversations that came before them

✓ 9,840 complaints linked to the conversations that came before them

Direct debit cancellations promised on calls, then never actioned, are the fastest-rising complaint driver.

01Cancellation promised, not actioned214 calls
02Refund timeline missed+32% in 4 weeks
03Fee disputes after switching140 complaints

Recommended fix

Add a cancellation check to call wrap-up

Owner

Payments Operations

Moves

Complaints, redress

View 214 source conversationsAssign to Payments Operations

Delivered as early warnings to the owner, a weekly brief, answers inside Copilot, ChatGPT or Claude, and board-ready reports.

How it works

One layer between your customers and your fixes.

SentiSum sits across every channel, brand and market. It reads what customers say against what your firm already knows, and turns it into fixes with owners.

CallsChatsEmailsComplaintsSurveysReviewsAI-assistant chatsPoliciesProceduresProduct termsKnowledge baseYour taxonomy

SentiSum Intelligence Layer

Read.
Every conversation, checked against your own procedures.
Spot.
New issues flagged the week they start.
Explain.
The cause, ranked by cost, with an owner.
Fix.
Tracked until the number moves.

Live in weeks

First findings within 4 weeks of data access.

SentiSum plugs into Genesys, NICE, Salesforce, Zendesk, Medallia, Qualtrics, Intercom and Trustpilot. Your teams keep the tools they use today.

Step 1
Connect
Standard connectors go live in days.
Step 2
Calibrate
The Intelligence Layer learns your policies, procedures and taxonomy. Your team approves every change.
Step 3
Act
First findings, named owners and a board-ready report.

Results

Every result is a fix that moved a number.

Top-five UK bank

£10M+

of redress traced to its root causes

SentiSum read a full year of the bank’s complaints and ranked every driver by cost, redress and volume, so each fix could be prioritised by what it was worth.

~195,000

complaints read

12 months

of history

~1,800

drivers ranked by cost

Payments platform

~70%

fewer complaints on one driver

Customers kept trying to cancel a payment. SentiSum found the cause behind about 4,000 complaints, the owning team shipped one fix, and complaint volume fell.

Payments platform

100%

of conversations quality-scored

Every merchant conversation scored against the same procedure. Sampling is gone, and so are the blind spots it left.

See the Quality Agent →

Trusted by CX teams at

Governance

Built for accountability. Designed to improve customer outcomes.

Your risk team will ask how the AI reaches its answers. Every answer can be checked, down to the conversation it came from.

Why not build it with a general AI model?

A general model stops at a summary. SentiSum is calibrated to your procedures and taxonomy, calculates every count, traces each finding to its source and follows the fix through to the owner. Live in weeks.

Traceable

Every finding links to the conversation it came from.

Calculated

Counts come from queries on your data, never generated by the model. When the evidence is missing, SentiSum says so.

Your taxonomy

The AI proposes changes. Your team approves them.

Isolated

Each client runs in its own environment.

EU AI processing

Every AI model call goes to EU endpoints only.

Tested

Annual external penetration test. SOC 2 Type 2 audit in progress.

The same evidence trail serves the people who oversee you.

Regulatory oversight
Demonstrate how customer outcomes are monitored, risks identified and issues resolved, backed by auditable evidence.
Quality assurance
Assess every interaction against your standards. Identify gaps, emerging risks and opportunities to improve.
Evidence of customer outcomes
Go beyond tracking actions. Demonstrate measurable improvements in customer experience and outcomes.

The outcomes challenge

See week one on your own data.

Pick one market or one journey. In a short paid pilot, we show you what to fix first, who owns it and what it is worth. You get a board-ready report within weeks.

See week one on your data Talk to our team