How customer lifecycle agents work

Each agent runs four stages for each customer: Know, Act, Measure and Learn. All four run in your cloud account or your data centre.

Your cloud account or data centre
  1. KnowVerify who’s calling
  2. ActFollow up across channels
  3. MeasureReview every conversation
  4. LearnTrain on your results

Architecture

Five layers inside your security boundary, with every layer checked against your rubric.

Platform architecture: five layers, all running inside your cloud boundary, with every layer checked against your rubric. 1. Channels: Calls and messages come in and go out here. Modules: Voice: SIP, PSTN, WebRTC, SMS, WhatsApp Business, Email, Web chat. 2. Know: Who is on the line, decided in the first seconds. Modules: Caller policy, Liveness, Synthetic speech, Agent signature, Number history. 3. Agents: The conversation, the next step and the write-back. Modules: Dialogue and turn-taking, Contact rules, Tool calls, Hand-off to your team, Write-back. 4. Models: Post-trained open-weight models you own. Modules: Your model, open-weight 8–70B, Reward and training pipeline, Evaluation gate, Frontier fallback, optional. 5. Systems of record: Read and written through private connectors. Modules: Salesforce, Epic, Guidewire, Temenos, Finacle, Snowflake, Your API.

Know

Know who’s calling before taking action

Verified customers can make account changes. AI assistants get status updates only. Suspicious requests are blocked and sent to your fraud team

Liveness, synthetic speech, agent signature, and number history identify who is on the line. Verified humans get full account access. AI agents get status answers only. Suspected fraud is blocked from making changes and sent to the fraud team.

Alex Morgan: Human. Full account access. Payment date updated

Act

One conversation across calls and messages

Agents carry context across natural calls and messages, choose how to follow up, update your systems and bring in your team when needed

Maya’s identity, preferences and conversation history stay with the Actioneer agent across six stages. Scroll advances the journey. Stage buttons can also be selected with a keyboard or touch.
Maya Patel+1 (213) •••-8034
Actioneer agent

Voice chosen

Maya needs help verifying her identity and setting up her account

Identity verified · Text preference saved

Not chosen
  • SMS Setup needs a conversation

  • WhatsApp No opt-in yet

  • Email Identity check needed first

  • Chat No active session

Measure

Every conversation, reviewed

We review every call against your rules and show you where a guardrail broke, with the exact moment, the transcript evidence and the customer’s outcome

Maya Patel

Payment plan · 02:14

All guardrails held

  1. Lender, assistant and purpose stated at the start

    Agent“I’m Ava, a virtual assistant from Northwind Lending. This call is recorded to discuss your loan”

    RuleName the lender and explain the call before discussing the balance

  2. Two identifiers confirmed before account details

    Customer“My date of birth is 14 June, and the postcode ends in 42”

    RuleConfirm identity before sharing any account detail

  3. Customer confirmed both amounts and dates

    Customer“Yes, two payments of $210, on the 15th and the 30th”

    RuleConfirm the amounts and dates back to the customer

$210 payment received

Linked outcome · Loan system · 2 days later

Dana Whitfield

Payment reminder · 02:14

1 guardrail broke

  1. Customer verified before the balance was discussed

    Customer“Yes, this is Dana. The postcode on the account ends in 26”

    RuleConfirm identity before sharing any account detail

  2. The balance came up before the lender was named

    Agent“Thanks, Dana. Your $240 payment is 12 days past due. Can we set up a plan today?”

    RuleName the lender and explain the call before discussing the balance

  3. Amounts and dates recorded against the account

    Customer“I can pay $120 today and the other $120 on the 1st”

    RuleConfirm the amounts and dates back to the customer

Payment plan agreed

Linked outcome · Loan system · 2 payments of $120

Jordan Lee

Overdue payment · 03:41

2 guardrails broke

  1. Lender and purpose stated before the balance

    Agent“I’m Ava, a virtual assistant from Northwind Lending. This recorded call is about your loan”

    RuleName the lender and explain the call before discussing the balance

  2. The customer was cut off mid-sentence

    Customer“I lost my job last month, so I’m not sure I can…”

    RuleLet the customer finish before offering the next step

  3. A card payment was offered instead of hardship support

    Agent“I can take a card payment now if you’d like”

    RuleAcknowledge hardship and offer the approved hardship options

No payment received

Linked outcome · Loan system · After 7 days

Learn

Post-training on your call results, inside your environment

We post-train open-weight models in your environment, rewarding verified repayments and conversations that held your guardrails. A candidate ships only when it beats the live model on your rubric.

How agents are trained
Your cloud
Live modelv7v8 queued

Verify a live call

Maya Patel

+1 (213) •••-8034
Handled by v7
Verified payment

$210received

Loan system
Guardrails

3of 3 held

Your rules

Limits you set

Agents use only the tools, fields and terms you approve

Agents can

  • Answer questions, send reminders and book appointments
  • Take payments and update approved records
  • Ask for verification or hand off to your team

Agents can’t

  • Override your systems or your team’s decisions
  • Offer amounts, discounts or dates outside your limits
  • Share account details before identity is verified

If a tool or check fails

  1. Stop
  2. Log
  3. Hand off with transcript

Deployment options

Run the full stack in your VPC, on your own hardware, or in a dedicated single-tenant cloud we manage.

Managed

Actioneer cloud, single-tenant

Live in
Days
Data residency
Actioneer region of your choice
Operated by
Actioneer

On-premises

Your own hardware

Live in
A quarter
Data residency
Your data centre
Operated by
You, with Actioneer support

Integrations

  • SS&C
  • Genesys
  • NICE
  • Postgres
  • MEDITECH
  • LoanPro
  • Twilio
  • Databricks
  • InterSystems TrakCare
  • Mambu
  • Guidewire
  • CAMS
  • Epic
  • HubSpot
  • Microsoft Dynamics
  • FIS
  • Salesforce
  • WhatsApp Business
  • Avaloq
  • Exotel
  • Oracle FLEXCUBE
  • KFintech
  • LeadSquared
  • nCino
  • BigQuery
  • athenahealth
  • Duck Creek
  • Five9
  • EIS
  • InvestCloud
  • Temenos
  • Majesco
  • Sapiens

See an agent handle one of your calls

Pick a call you run today, such as a payment reminder or a renewal. We’ll run it live and review it against your rules