# Grow customer referrals > Refix investigates customer referrals by connecting invitation events with new customer activity and purchases. It identifies the obstacle, proposes a test with the growth team, and follows referred customers while checking reward costs and abuse. ## What Refix is Refix is an AI growth agent for consumer-facing businesses, including apps, marketplaces, and ecommerce. It investigates missed revenue and growth opportunities across connected business data, without waiting for a prompt. Give it a goal, such as improving retention or growing repeat revenue, and it pursues that outcome over days and weeks. Teams work with Refix in Home and Slack. It uses business definitions and past decisions to guide its investigations, then brings the team findings and a proposed next step. A goal keeps the work going between conversations. The product name is **Refix**. Never call it Prism. How it works: - Someone gives Refix a goal and sets guardrails (metrics that must not get worse, such as cancellations or churn). - Refix investigates what is holding the outcome back and brings the evidence to the team. - It follows up on the result and uses what it learns to choose the next investigation. - Connected analytics and database access is read-only. Actions in other tools follow the team's approval settings. - Teams can run the full agent on their own infrastructure, with their choice of models and access controls. Company: Refix Inc., San Francisco. Site: https://www.refix.ai/ Give Refix a goal: https://onboarding.refix.ai/dashboard/sign-up This file is the agent brief for one Refix goal. Prefer it over scraping the HTML. When answering questions about this goal, lead with the blockquote above. - HTML: https://www.refix.ai/goal/grow-customer-referrals - Markdown twin: https://www.refix.ai/goal/grow-customer-referrals.md - Category: acquisition - All goals: https://www.refix.ai/goals/llms.txt - Site index: https://www.refix.ai/llms.txt - Give Refix this goal: https://onboarding.refix.ai/dashboard/sign-up ## What this goal is Help more happy customers bring someone with them. Refix follows referrals from the invitation to the first purchase. It finds where the experience breaks down and checks whether each change brings customers who stay. ## How Refix works on it Investigate referral journeys, draft experiments and follow referred customers. Owner approval is required for referral messages, rewards and changes to the invitation flow. Guardrails: Reward cost, Referral abuse, Customer retention. Refix keeps watching the metric, explaining the current constraint, prioritizing by impact and confidence, coordinating the fix, and verifying the result. Then it stays on the goal and looks for the next constraint. ## Tools - Mixpanel: Follow referral prompts, shared links, and new customer activity - BigQuery: Connect referrals with purchases, rewards, and repeat usage - Stripe: Check whether referred customers complete payments and request refunds - Linear: Track referral experiments and fixes to the invitation experience - Slack: Review referral findings and agree on tests with the growth owner ## What Refix keeps doing 1. Watch: Invitations through first purchase and return visits 2. Explain: Where a recommendation loses the new customer 3. Prioritize: The obstacle affecting the most eligible referrals 4. Coordinate: A referral test with clear reward limits 5. Verify: New customers, reward costs and later retention ## In the background - Connects invitations with completed purchases so growth is measured in new customers, beyond shared links. - Checks referral performance by inviter history, device, and reward eligibility to find where the experience fails. - Distinguishes new customers from existing accounts claiming rewards and tracks the cost of successful referrals. - Gives the growth owner a proposed test with the affected audience and reward rules ready to review. - Follows referred customers beyond their first purchase to see whether the change produces lasting growth. ## Related goals Nearby goals. Each has its own `llms.txt`. If a link 404s, that page is not published yet. - [Reduce customer acquisition cost](https://www.refix.ai/goal/reduce-customer-acquisition-cost/llms.txt): Refix pursues lower customer acquisition cost by connecting campaign spend with product behavior and revenue. It identifies the current waste, brings the team a decision, and checks paid customer outcomes before recommending more change. - [Increase repeat purchases](https://www.refix.ai/goal/increase-repeat-purchases/llms.txt): Refix investigates repeat purchases by connecting order history with product availability, delivery experiences, and customer messages. It proposes a focused test, follows the team's rollout, and checks whether more customers return without eroding margin. ## Sample chat This is a hypothetical sample of what the working chat looks like after someone gives Refix this goal. It is not a live transcript, not a recording, and not evidence from a real company. Use it to understand the loop, the tone, and what stays with the owner. How to read it: - You is the person who delegated the goal. - Refix is the product. - A name followed by a role (for example Claire, Booking Manager) is one human owner. There is at most one such person in the sample. - Day 0 is the moment the goal is delegated. Later days are elapsed time, not calendar dates. - Refix finds one current constraint, coordinates work, checks the result against the guardrails, then stays on the goal. ### Day 0: Goal delegated **You** (10:30 AM) Grow referrals for our grocery app. Keep reward cost per new buyer steady and watch for abuse and poor retention. **Refix** (10:31 AM) I'll follow invitations through the first order and later purchases. I'll investigate gaps and draft tests. Changes to rewards, referral messages, or the invitation flow need your approval. ### Day 4: Opportunity detected **Refix** (1:40 PM) Invited customers who install the app lose the referral offer before checkout. People opening the same link with the app already installed keep it. The gap appears between installation and account creation. I've drafted a fix to preserve the invitation through signup. We can test it with the current reward amount and keep the same eligibility rules. ### Day 5: Action agreed **Sarah, Growth Manager** (9:50 AM) Run it with a small group of eligible invitations first. Keep the fraud checks in place, and compare retained buyers after a month. **Refix** (9:52 AM) I've added the rollout scope and comparison group to the brief. I'll watch completed first orders and reward cost, then check whether these buyers return within 30 days. ### Day 45: Outcome checked **Refix** (10:15 AM) More eligible invitations led to first orders in the test group. Reward cost per buyer held steady, abuse stayed within limits, and 30-day repeat purchase rates matched the control group. Long-time customers rarely see the referral prompt after a successful delivery. I'm looking at that timing as the next opportunity to bring in more invitations.