The evidence that an account is leaving usually sits in your own systems for months. Nothing is watching it. That is the part we build.
Keeping a customer costs less than winning one. Most owners accept that and keep spending on the winning half anyway, because winning is visible and keeping is not. The customer retention software sold to Australian businesses is usually a loyalty program or an email tool. Neither of those tells you that an account is quietly on its way out.
By the time a customer has actually gone, the evidence has generally been sitting in your own systems for months. Orders got smaller. The gap between them got longer. Support tickets went up. The person who championed you internally changed roles and their replacement has never spoken to you. None of that is hidden. It is just not being watched by anything.
We build the systems that watch it: churn indicators drawn from your own transaction and service records, renewal and re-order prompts that fire early enough to act on, account health visible to the person who owns the relationship, and post-sale follow-through that does not depend on somebody remembering.
This is deliberately different work from sales growth systems. That side of the business is about winning. This side is about keeping, and it runs on behaviour you already have a record of rather than intent you have to infer.
Order frequency slipping against a customer's own baseline, average order value falling, a plan downgrade, a support ticket rate that has doubled, usage or logins dropping away, a contact who has stopped responding entirely. Each signal is measured against that customer's own normal rather than a company-wide average, which is the difference between a useful alert and noise. The data analytics work behind it makes the thresholds defensible.
Contracts, subscriptions, service agreements, warranties, maintenance intervals and consumables all have a date attached. The system tracks them from the source records in your accounting, ERP or booking system and prompts the right person at 90, 60 and 30 days rather than in the week the thing expires.
One screen per customer: what they buy, how often, what has changed, what is open, what is overdue, when someone last spoke to them and what about. Built as part of your dashboards and reporting so the account manager and the operations manager are working from the same numbers rather than two exports.
Response and resolution times reported against the customer they happened to. Aggregate service reporting hides the one account having a terrible quarter, and that account is usually the one about to leave. Breaches surface attached to a name, an owner and a next action.
Onboarding steps, delivery confirmation, installation sign-off, first-90-day check-ins, review requests, seasonal service reminders. These get dropped when everyone is busy, which is exactly when they matter. Each one gets an owner, a due date, and visibility when it is missed.
Marketing consent, communication preferences and unsubscribe state kept against the customer record and honoured by every system that touches it. A customer who opts out of marketing should still receive their renewal notice — those are different categories of message, and the system needs to know the difference.
First we define what "leaving" actually looks like in your business, because it differs. For a retail group it might be a rewards account going quiet for two ordering cycles. For a financial services firm it might be a renewal that suddenly goes to market. Vague definitions produce vague alerts.
We find where the evidence already lives — Shopify, Cin7, Xero, MYOB, the booking system, the support inbox, the CRM — and check whether it is complete enough to base a signal on. Sometimes the honest finding is that a gap needs closing before any of this is worth building.
Thresholds, who gets alerted, and what happens next. An alert nobody owns is just another notification. We also agree here how consent and preference data will be handled across every connected system.
We build the signal logic, the prompts and the account view, and connect them to the systems that hold the underlying records. Client data we host sits in Australian regions.
After a few months we check whether the signals predicted anything real. The ones that did get tightened. The ones that did not get retired, because a retention system that cries wolf gets ignored within a quarter.
A loyalty program rewards customers who are already engaged. It does not tell you that a customer's ordering pattern has changed, that their support experience has deteriorated, or that their renewal is 60 days out and nobody has called. Retention systems watch behaviour and prompt action. The two can sit alongside each other, but they solve different problems.
Using records you already hold to serve your own customers is ordinary practice, but it is governed. The Privacy Act and the Australian Privacy Principles cover how personal information is collected, used, stored and disclosed, and the Spam Act covers consent, sender identification and unsubscribe for marketing messages. We build consent and preference handling in from the start. We build systems, not legal advice — confirm your own obligations with your adviser.
Drawn AI works on a subscription: monthly billing, a three-month minimum term, and 30 days' notice to cancel. Not fixed-price project work, not hourly. What you pay each month depends on how many signals you want, how many systems hold the data, and how much of the account view you need. We quote after discovery rather than from a brief.
Enough to establish what normal looks like for each customer. For a business with weekly or monthly ordering, roughly twelve months of transaction history is a reasonable starting point. With less than that, renewal and expiry prompts still work immediately because they run off dates, while the behavioural signals get more reliable as data accumulates.
Less than a sales build, because most of it reads existing records rather than changing how people work. The effort lands in two places: agreeing definitions and thresholds during design, and deciding who owns each alert. If nobody is assigned to act on a signal, the system produces reports instead of retained customers, and that is a decision about people rather than software.
A CRM records what your team entered. It rarely watches what your customers did. Most of the signals that matter here live in transaction, support and usage data outside the CRM. We usually connect those sources and surface the result back into the CRM, so the account manager sees it where they already work rather than in a second system.
A first conversation costs nothing and usually ends with a clearer idea of what is worth building — sometimes that answer is “not yet”, and we will say so.
No lock-in templates. Built around you.