What problem does it solve?
Banks pay to acquire customers and products that then sit unused. A new card that never leaves the drawer earns nothing, a savings account opened for a promotion goes dormant, and a customer who keeps paying overdraft fees has a reason to look at another bank. The bank usually knows the moment to act (a card not used 30 days after delivery, a balance heading below zero, a fixed rate ending), but the tools it has are weak: push notifications and emails that are easy to ignore, or human outbound calls that are costly to scale beyond high value products.
What is missing is a way to have a short, useful two way conversation at scale: answer the customer's question, complete the action (activate, set up a transfer, accept an offer they already qualify for) and leave them alone when they say no. Because some of these conversations touch credit and fees, they are regulated conversations, not marketing copy.
- Forrester's survey of banking customers, as reported by Mi3 in 2025, found that the thing customers most want (60 percent) is to be alerted when there is not enough money in their account to cover an upcoming expense.Westpac blows app rivals away as Forrester rates Australia among world's best – but Big Four still missing key customer aspirations (2025)
How does it work?
- The bank decides who and why. Event triggers and campaign lists come from the bank's own systems (core banking, the decision engine, CRM), including eligibility for any offer. The agent does not choose targets.
- Check consent and contact rules. Before any message or call the agent checks marketing consent where needed, frequency caps, quiet hours and the customer's preferred channel.
- Open with the reason. The agent says it is an AI assistant, names the bank and states the specific reason for contact ("your new card has not been activated").
- Converse and act. It answers questions from the customer's own account data and approved product content, and completes the action through allow listed APIs after the right authentication: activate the card, set up a transfer to avoid a fee, book a call, accept a pre approved offer.
- Respect no. An opt out or a "not now" is recorded at once and suppresses further contact on that topic.
- Hand over. Complaints, hardship, vulnerability and complex product questions go to a human with the conversation attached.
- Write back the outcome. Results flow to CRM so the decision engine learns and the next campaign does not repeat the contact.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Emerging
- Channels
- WhatsApp, SMS, Phone and voice, Mobile app, Email
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Interactions handled | Not pooled | at least 3 billion | 1 | 1 organization |
Value drivers: Revenue growth, Customer experience, Lower cost to serve.
Indicative value
A card issuer that issues 100,000 new cards a year
USD 25,000 to USD 360,000
Annual revenue from cards activated by outreach per year
How this is calculated
Formula: newCards * inactiveShare * activationLift * revenuePerActiveCard. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| New cards issued per year newCards, cards per year | 100,000 | 100,000 | The reference issuer. |
| Share of new cards not used within 90 days inactiveShare, fraction of new cards | 0.1 | 0.2 | Editorial assumption, replace with your own activation data. |
| Share of inactive cards activated because of the outreach activationLift, fraction of inactive cards | 0.05 | 0.15 | Editorial assumption; measure against a control group that receives only the usual push and email. |
| Annual revenue from an active card revenuePerActiveCard, USD per card per year | 50 | 120 | Editorial assumption, replace with your own figure. |
What it leaves out: Covers card activation only. It leaves out fee avoidance, reactivation of dormant accounts, accepted offers and retention, the cost of messages, calls and the AI, and any complaints from unwanted contact.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Capital One
United States · Banking · 2026
Eno is Capital One's virtual assistant. Capital One says it helps protect card accounts by looking out for charges that might surprise the customer, and sends insights when it spots free trials and recurring charges, through text, email and app alerts. Capital One does not publish outcome figures for Eno on this page.
No outcome disclosed.
Bank of America
United States · Banking · 2025
Erica, launched in 2018, is Bank of America's virtual financial assistant in its Mobile Banking app. Beyond answering questions it delivers proactive, personalized insights: BankAmeriDeals cash back deals based on the client's spending, where balances are trending over the next seven days and eligibility for the Preferred Rewards program. It also gives guidance on investment topics for Merrill clients and hands off to people by scheduling appointments. The bank reports that clients have received and interacted with more than 1.7 billion of these insights, and that most users find the information they need, which it links to lower call centre volume. Bank of America says Erica selects answers from a predefined set and does not use generative AI or large language models.
- Users served: about 50 million, since launch in 2018, as of August 2025
"assisting nearly 50 million users since launch, surpassing 3 billion client interactions, and now averaging more than 58 million interactions per month"
Claimed by: organization - Interactions handled: at least 3 billion, client interactions since launch in 2018, as of August 2025
"surpassing 3 billion client interactions"
Claimed by: organization
Commonwealth Bank of Australia
Australia · Banking · 2022
Commonwealth Bank's Customer Engagement Engine (CEE), built on Pega Customer Decision Hub, suggests in real time the next best conversation to have with each customer, whether in the branch, on the phone, online or on a mobile device. Beyond suggesting conversations, the bank uses it to match customers to government benefits and rebates they may be missing (Benefits finder) and to reach customers hit by natural disasters with same day support such as a loan deferral. The same decisions feed digital channels and prompts for branch and contact centre staff.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Event triggers and campaign lists with the reason for contact
- Consent, channel preference and suppression data per customer
- Eligibility decisions for any offer, made upstream by the bank's decision engine
- Approved product and fee content for questions
Systems to integrate
- CRM or customer engagement platform (lists, outcomes, suppression)
- Messaging and telephony channels with opt out handling
- Core banking and card platform (activation, transfers, limits)
- Identity and step up authentication
- Contact centre for handover and callbacks
Complexity: Medium
The conversation is straightforward. The hard parts are the triggers and eligibility data, the consent and contact rules per channel and market, and authenticating a customer on a conversation the bank started.
- 1
Start with service triggers, not sales
Card activation, low balance and payment reminders help the customer and build trust in the channel. Add offers only after service outreach performs well.
- 2
Keep the campaign brain outside the agent
Targeting, eligibility and offer terms come from the bank's systems and are passed to the agent as facts. The agent explains and executes; it does not decide who qualifies.
- 3
Build consent and frequency into the trigger
Check consent, caps and quiet hours before the first message, per channel and market, and log the check with the contact.
- 4
Authenticate on the way in
When the bank starts the contact, the customer must still prove who they are before any action, ideally in the app. Never ask for secrets in an outbound message.
- 5
Test against a control group
Hold out a random share of each trigger and compare activation, fees and churn, so the results are yours and not the vendor's.
Guardrails
- The agent only offers products the bank's decision engine has already approved for that customer
- AI disclosure and the reason for contact in the first message or sentence
- Opt outs honoured immediately across channels
- No requests for passcodes or card details in outbound contact
- Frequency caps and quiet hours enforced before sending
KPIs to instrument
- Activation, reactivation and acceptance rates against a control group
- Opt out and complaint rate per trigger
- Fees avoided for customers who acted on a low balance alert
- Handover rate and reasons
- Churn of contacted customers versus control
Human in the loop
Marketing and product owners approve every trigger, script and offer before launch. Humans take over for complaints, hardship, vulnerable customers and complex product questions, and a sample of conversations is reviewed each week for tone, accuracy and fair treatment.
Common failure modes
- Outreach that feels like spam
- Too many contacts or vague reasons drive opt outs and complaints. Cap frequency, lead with the reason and stop when the customer says no.
- The agent starts deciding eligibility
- Free form offers creep into the conversation. Pass eligibility in as data and block offers that were not approved upstream.
- Scam lookalike messages
- Outbound bank messages are what scammers imitate. Never ask for secrets, and point customers to the app to act.
- No measurable effect
- Without a control group every activation looks like a win. Hold out a share of each trigger from day one.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
A customer facing agent must disclose that it is AI (Article 50(1)). It stays out of Annex III as long as eligibility for credit offers is decided upstream by the bank's own, separately governed credit processes; if the agent itself assessed creditworthiness it would be high risk under point 5(b).
Guidance
- Guide to Privacy and Electronic Communications Regulations (Information Commissioner's Office, Europe). UK rules on marketing by phone, text and email, including consent and opt out. The ICO says the guide is under review after the Data (Use and Access) Act 2025.
- Declaratory ruling on AI generated voices under the TCPA (FCC 24-17) (Federal Communications Commission, North America). AI generated voices count as "artificial or prerecorded voice" under the TCPA, so US outbound AI calls need the consent the TCPA requires.
Controls to put in place
- Consent, frequency and quiet hour checks logged per contact
- Approved scripts and offer terms per trigger with an accountable owner
- Audit trail of every contact, answer and action taken
- Suppression lists shared across channels
- Monitoring of complaints and outcomes for vulnerable customers
Frequently asked questions
- Is this the same as a marketing campaign tool?
- No. The bank's campaign and decision systems still choose who to contact and what they qualify for. The agent is the caller: it explains, answers questions, completes the action and records the outcome.
- Do banks already reach out proactively with AI?
- Yes, but mostly as proactive alerts and insights in the app and by message, from an assistant the customer can then talk to. Bank of America says clients have received and interacted with more than 1.7 billion proactive, personalized insights from Erica, and Capital One says its Eno assistant looks out for charges that might surprise the customer and alerts them by text, email and app. We found few published results for AI agents that hold two way outbound conversations or make outbound calls in banking, which is why this page treats the pattern as emerging.
- What consent do outbound AI calls need?
- It depends on the market and on whether the call is service or marketing. In the US the FCC has ruled that AI generated voices fall under the TCPA's rules for artificial voices; in the UK PECR governs marketing calls, texts and emails. Build the check into every trigger.
How to cite this page
Blits.ai AI Use Case Library, "AI agent for proactive customer outreach, activation and retention", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/proactive-outbound-engagement-agent. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published