Skip to content
ConvoAssist

How it works

Classify by risk. Automate what should be automated. Evidence it monthly.

Every contact driver in your support operation is classified into one of three risk tiers. What gets automated, what gets gated and what is never auto-resolved all follow from that classification.

£7,500 one-time Launch · £1,950/month Management · 12-month term, break clause each quarter

The risk tiers

What the AI never touches, first.

Start with what the AI will not touch. Tier 3 contacts are never auto-resolved — they are detected, routed and logged every time.

Tier 3

Restricted

Never auto-resolved

Complaints, vulnerability, financial difficulty and fraud. The AI never resolves these. It detects them, routes them to a human and logs the detection.

  • Complaints
  • Vulnerability signals
  • Financial difficulty
  • Suspected fraud

Tier 2

Transactional

Automated with a full audit trail

Account actions the AI can take once identity is verified and the action is gated. Every step is logged.

  • Card freeze and replacement
  • Payment status
  • Statement and document requests
  • Limit and status checks

Tier 1

Informational

Fully automated

Questions with a factual answer and no account action. Fully automated against corrected knowledge.

  • Product and fee questions
  • Balance and transaction queries
  • KYC document guidance
  • How-to and app questions

Launch

What happens before anything goes live.

Typical timeline 5–7 weeks.

Launch starts with 90 days of your contact data. Every driver is classified by risk and volume, and a baseline cost-per-contact model is built from your numbers and agreed with you before any build work begins.

Tier 1 and Tier 2 journeys are then built, knowledge is audited and corrected, identity verification and action gating are put in place for Tier 2, and detection and routing is built for complaints, vulnerability, financial difficulty and fraud.

Launch closes with a pre-deployment impact assessment, a test evidence pack across all three tiers, audit logging design and a signed scope statement of what the AI handles and what it never handles.

How it works

Assess. Build. Run and evidence.

01Assess
I review 90 days of your contact data, classify every driver into Tier 1, 2 or 3, and agree a baseline cost-per-contact model with you.
02Build
Tier 1 and Tier 2 journeys are built, knowledge is audited and corrected, identity verification and action gating are set for Tier 2, and Tier 3 detection and routing is put in place. You get a test evidence pack and a signed scope statement.
03Run and evidence
Every month I run the Optimisation Loop against real conversations and deliver the Monthly Assurance Pack.

The Optimisation Loop

The method that produces the Assurance Pack.

The Optimisation Loop is the method. The Monthly Assurance Pack is the deliverable. Every item in the Pack is an output of the Loop — nothing in it is generated separately or after the fact.
  1. 1Analyse

    Read real conversations. Find wrong answers, missed Tier 3 signals and new contact drivers.

  2. 2Prioritise

    Rank by risk first, then by volume and cost.

  3. 3Improve

    Correct knowledge, adjust journeys, tighten detection and escalation rules.

  4. 4Test

    Test every change across all three tiers before it goes live.

  5. 5Report

    Everything above is written up in the Monthly Assurance Pack.

  6. 6Repeat

    The following month starts from the previous month's findings.

See what the Assurance Pack contains

The deliverable

The Monthly Assurance Pack.

Six components, delivered every month. This is what your compliance function reads.
Monitoring records
What was monitored, how much of it, and over what period.
Intervention log
Every change made to the deployment, dated and attributed.
Outcome data
Resolution rate, escalation rate, recontact rate and CSAT split by AI and human.
Tier 3 performance
Detection performance, any misses, and the remediation for each.
Cost report
Net saving against the baseline cost-per-contact model agreed at Launch.
Change register
A running register of every change, request and decision.
Read the Assurance Pack in full

Who this is for

Three questions.

Built for UK consumer fintechs — neobanks, lending, savings and investing apps, credit builders, BNPL and consumer payments — running FCA-regulated support operations.
Do you handle 800 or more customer contacts a month?
Below that, the cost-per-contact case usually does not carry the fee.
Are your customers retail consumers?
The risk-tier framework is built for consumer-facing, FCA-regulated firms.
Are you on Fin or Zendesk AI, or actively choosing?
I implement and manage on those two platforms.

Who does the work

I'm Rob Darren. Six years running customer operations inside UK challenger banks, including building a bank's support automation stack from scratch. I run every ConvoAssist deployment directly.

Rob Darren

Founder, ConvoAssist

LinkedIn
More about ConvoAssist

FAQ

Process questions

AI customer support your compliance team can sign off, and keep signing off.

If this is the way you would want an AI deployment run inside a regulated firm, book an assessment call and we will start with your contact data.

Rob DarrenFounder, ConvoAssist