Building a referral/recommendation engine:
Strategy, Automation, and Measurement Blueprint

Part 1: Strategic architectural design of the recommendation engine

The success of a recommendation engine is built on two strategic pillars: ‘incentives’ and ‘triggers’. While incentives provide a strong motivation for customers as to why they should recommend, triggers capture the optimal timing of when to ask for them to be most effective.

1.1. Core incentive model design: mutual reward, one-way, grade differential

1.1.1. The Strategic Importance of Incentives: Why Recommendations Are More Powerful than Advertising

Referral programs are more than just a cost-effective customer acquisition channel. Purchase decisions based on friends' recommendations have a 90% higher purchase rate than regular advertisementsIt shows.

1.1.2. Model 1: Mutual Compensation (Double-Sided) - ‘Win-Win’ structural design

• definition: The most common and powerful incentive model that provides rewards to both the advocate and friend.
• Mechanism of action: Based on the principle of ‘Reciprocity’. The recommended person's barrier to entry is lowered with an immediate discount, and the recommender becomes the subject of an altruistic act of giving a 'gift' to a friend, gaining both psychological satisfaction and material rewards.

Success story analysis:

• Dropbox: When you invite a friend, both parties receive 500MB of additional storage space. Non-cash ‘product value’ based compensation 3,900% explosive growth attainment
• PayPal: If your friend signs up and spends 7,250 won, both parties will receive 14,500 won. Spread the initial user base like a virus with the help of cash rewards
• Robinhood: A unique reward called ‘Futures Stock’. When signing up, both parties are provided with random stocks, securing a waiting list of 1 million people before the service launches.

apply: It is most effective for most B2C, e-commerce, and subscription-based services that can offer clear and immediate value, such as “Give my friend 29,000 won as a gift, and I also get 29,000 won.”

1.1.3. Model 2: One-Sided - Strategic Use Case

• Referrer Reward Type (Advocate-Only): In fact, it is similar to ‘affiliate marketing’. Mainly used in B2B SaaS or expensive professional services. Since referrals are directly linked to ‘reputation’, clear financial rewards (e.g., KRW 11,600 to KRW 33,350 per user in Google Workspace) are a strong motivator.
• Referred person compensation type (Friend-Only): Recommendation activity without compensation because the recommender already has strong brand fandom (intrinsic motivation). Focus solely on lowering the barrier to attracting new customers and providing discounts only to referred people

1.1.4. Model 3: Tiered and Cumulative - ‘Super Fan’ Cultivation Strategy

• definition: Dynamic model in which the size or type of reward varies depending on recommendation performance (number of invitations, number of successes, sales)
• Mechanism of action: Introducing gamification elements to encourage continuous, rather than one-time, recommendation activities and foster ‘super ambassadors’

Case analysis:

• Tiered: Harry's legendary launch campaign. ‘Free shaving cream’ for 5 referrals, ‘Free razor’ for 10 referrals, ‘Free razor for 1 year’ for 50 referrals. Collect 100,000 emails in just one week
• CycleBoard provides 10% of the order purchased by a friend in cash to the referrer. Motivate your referrers to make larger purchases
• Multi-step: 14,500 won is paid for the first 5 successful referrals, and 36,250 won thereafter is paid to lower the initial barrier and provide long-term motivation.

1.1.5. The hidden power of non-cash rewards: lower costs and higher LTV

1.2. The Science of ‘Invitations’: Identifying Key Trigger Points in Referral Messages

Once you've designed the optimal incentive model, you now need to decide when to ask for referrals. The psychological peak (Peak Moment) when customers are most satisfied with a product and have the strongest recommendation intention.Capturing and sending automated requests at that moment is key to a recommendation engine's success.

1.2.1. Onboarding: Capture the ‘Aha Moment’ immediately

• What is ‘Aha Moment’? It's the first time customers emotionally realize the core value of your product and have an "Aha!" The moment you do it
• example: Facebook found that users who were ‘connected with 7 friends within 10 days of signing up’ did not churn. Users who experience this moment clearly recognize the value of the product, and the probability of sharing the positive experience with others increases rapidly.
• execution: Clearly define the ‘Aha Moment’ of your product with data (e.g., ‘Automatically create your first report’ in SaaS, ‘Complete delivery of your first order and write a positive review’ in e-commerce, ‘Invite three team members and create your first project’ in collaboration tools). As soon as this event occurs, request recommendations via in-product pop-up, push notification, or email

1.2.2. Satisfaction: Convert NPS/CSAT data into triggers

Reinterpreting the Net Promoter Score (NPS): NPS surveys aren't just a satisfaction metric; they're the most powerful automated trigger tool for instant customer segmentation and next action.

Automation logic:

1. Send NPS surveys at key milestones in the customer journey (30 days after purchase, 3rd month of subscription)
2. System analyzes responses in real time
3. Promoters (9-10 points): Core customers with the highest willingness to recommend. Immediately leads to the referral program landing page with the message “Thank you for recommending our service! Do you have a friend you would like to share this experience with?”
4. Passives (7-8 points): Satisfied but not enthusiastic. “What can be improved to get a better 9?” Request feedback
5. Detractors (0-6 points): Customers at risk of dissatisfaction and churn. Respond quickly (prevent churn) by automatically creating a warning or support ticket to the customer support team/CSM immediately upon response

1.2.3. Achievement: Inducing trust-based B2B recommendations

• Special characteristics of B2B: In a B2B environment, recommendation is not a matter of ‘discount’ but is directly related to the ‘reputation’ of the recommender. Customers hesitate to recommend solutions until they are ‘confident’ that the solutions they recommend will deliver real results to their colleagues/third parties.
• trigger: Not immediately after signing up, but at the point when data confirms that clear business results have been achieved (e.g., ‘maintain long-term subscription for more than 6 months’, ‘check ROI report generated by product use’, ‘organic mention of our solution at conferences’)
• execution: It is more effective to have a personalized message directly from the CSM or sales team rather than an automated email. 91% of B2B purchasing decisions are influenced by word of mouth

Part 2: Recommendation Workflow Automation and Technical Implementation

This is a technical blueprint to turn the strategy designed in Part 1 into an executable ‘engine’. In this step, we build a system that reacts to customer actions (triggers) in real time to automatically guide, track, and reward the recommendation process.

2.1. Automation workflow design: build logic for each scenario

A successful referral program shouldn't be managed manually. From registering customers, issuing unique links, tracking sharing activity, confirming conversions, and distributing rewards, the entire process should be automated by the system.

2.1.1. Essential Tool Stack Definition

For reliable recommendation engine automation, at least four key components must be organically linked:

1. Referral Platform (Command Center): Setting up referral logic, generating unique codes/links, tracking referral attribution, providing dashboards, and approving and managing rewards (e.g. Referral Rock, GrowSurf, ReferralCandy, Friendbuy)
2. Email Software (Communication): Sending event-based automated email/message sequences (e.g. MailChimp, ActiveCampaign, Brevo)
3. CRM (Customer Data): Store all customer information, purchase history, NPS score, CSAT score, etc. and perform customer segmentation (e.g. HubSpot, Salesforce)
4. Integration Tool: Acts as a glue that connects the three core systems above and existing websites/apps (e.g. Zapier, own API integration)

2.1.2. Scenario 1: Automatic registration of advocates for ‘satisfied customers (Promoters)’

• Trigger: Customers are tagged with an ‘NPS score of 9-10’ or achieve an ‘Aha Moment’
• Action (Automation):
  1. Automatically send a “special gift for VIP customers” personalized email sequence to the relevant customer
  2. When you click on the CTA (“Read more”) in the email, you will be taken to the ‘Advocate landing page’
  3. Click the ‘Program Participation/Agree’ button
  4. Immediately generate and issue unique referral codes and referral links through API calls
  5. Instantly redirect to your personal dashboard

2.1.3. Scenario 2: Tracking the inflow and conversion of ‘Friends’

• Trigger: New visitors click on a ‘referral link’ (e.g. https://brand.com/?ref=DAVIDP4821)
• Action (Automation):
  1. Website tracking script immediately recognizes the URL's recommended parameters (ref=DAVIDP4821)
  2. Identify the visitor as a recommended person of 'DAVIDP4821' and save the information in a cookie.
  3. Immediately display the 'Friend landing page' to visitors (e.g. "David gave you a 20% discount!")
  4. Referred person signs up for membership or completes first purchase (conversion event occurs)
  5. Tracking script detects conversion event and sends it with referral code stored in cookie
  6. ‘1 conversion (pending)’ recorded as a result of recommendation of ‘DAVIDP4821’

2.1.4. Scenario 3: Reward automatic payment and notification

• Trigger: Recorded ‘Conversion (Pending)’ meets predefined ‘success’ criteria (e.g. 30-day refund period has passed, abuse review has passed)
• Action (Automation):
  1. A notification is automatically sent to the recommender saying, "Congratulations! Your friend's purchase has been confirmed and a 29,000 won reward has been paid."
  2. Send a confirmation notification to the referred person saying “20% discount on first purchase has been successfully applied”
  3. Automatic payment of rewards (discount coupon code, account credit, points) set to both parties

2.2. Recommendation code issuance and tracking logic design

A reliable referral tracking system is the heart of your recommendation engine. If code issuance is duplicated or recommendation performance is missing, customers will immediately lose trust in the program, which will directly lead to program failure.

2.2.1. Referral Code vs. Recommended Links: What Should I Use?

• Referral Code: (Example: DAVIDP4821) A unique identifier that is easy for users to remember and enter directly. Advantageous when sharing offline or verbally
• Referral Link: (Example: .../?ref=DAVIDP4821) The recommendation code is included in the URL parameter. As the recommended information is automatically transmitted to the system as soon as the recommender clicks, it is absolutely advantageous for online sharing, with no user errors and very high tracking accuracy.
• Optimal design: Provide both, but use the referral link as the default. The best UX is designed so that when the referred person clicks the referral link and reaches the landing page, the referral code is already automatically entered in the 'Coupon Code' field on the membership registration/payment page.

2.2.2. Unique referral code creation logic (Technical Logic)

Best Practice:

1. Database-dependent approach: The system already assigns every user a unique identifier (e.g. auto-increment User ID 12345)
2. Encode this unique ID 12345 with Base62, etc., or combine it with part of the user's name (e.g. DAVID + 12345) and simple hash it to generate an unpredictable yet unique code (e.g. DAVIDK3X).
3. Since this method does not require any DB lookup when generating code (lookup-free), performance is very fast and system scalability is guaranteed.

2.3. System Integrity: Strategies to Prevent Abuse and Referral Fraud

core: Strong incentives (especially cash rewards) inevitably attract fraud and abuse attempts. If an abusing defense system is not built in from the recommendation engine design stage, it may degenerate into a cost engine rather than a growth engine.

2.3.1. Identify major types of fraud

• Self-Referral: The referrer creates another email address or account to recommend himself and steals both parties’ rewards.
• Duplicate Accounts: A single fraudster creates dozens or hundreds of accounts using multiple fake emails, different IP addresses, virtual devices, etc. to steal only referral rewards.
• Coupon Spam: The recommender posts his/her unique recommendation code on a coupon discount site visited by an unspecified number of people other than 'friends', damaging the original purpose of the program, 'trust-based recommendation', and incurring enormous costs.
• Bot activity: Mechanically repeat sign-ups and recommendations using automated scripts

2.3.2. Technical Defense Logic

The system must have the following multi-layered defense logic:

• IP address monitoring: If the referral IP address of the recommender and the recommended person is the same or is from an already known data center/proxy IP, compensation payment will be automatically withheld.
• Device Fingerprinting: Delete cookies on the same device (browser, mobile device) and identify and block abusers who attempt to sign up with different emails.
• Velocity Checks: If a single referrer attempts an unusually large number of referral successes (e.g. 100) or reward refunds within a short period of time (e.g. 1 hour), the account is automatically flagged and placed under administrator review.
• Required Certifications: When new referrals sign up, they must go through email authentication or SMS OTP (mobile phone authentication) to fundamentally prevent bot activity and the creation of fake accounts.

2.3.3. Policy Defense Logic

Technical defense alone has its limits, and a legal basis is required in the event of a dispute. All referral programs must have clear and legally binding provisions to which participants must agree. Terms of Use (T&C)Must have.

Required provisions:

• Eligibility: Clearly define who can participate in the program (e.g., only existing customers with a purchase history are eligible as referrals, excluding employees)
• Definition of a ‘Qualified Referral’: Specify in detail the criteria for 'success' for which compensation is paid (e.g., when the referred person completes the first purchase of KRW 72,500 or more after signing up and the 30-day refund period has elapsed)
• Prohibited Conduct: Abusing practices such as self-recommendations, spamming (posting on coupon sites, sending spam emails), and use of bots are explicitly prohibited.
• Company Rights Notice (The “Sole Discretion” Clause): The most important clause. If fraud, abuse, abuse, or violation of the Terms of Use is suspected, the Company reserves the right, at its sole discretion, to refuse compensation or disqualify you from participating in the Program.

Part 3: Guidelines for creating key deliverables

If you have prepared the strategy in Part 1 and the technology in Part 2, you now need to design the key touchpoints to meet customers. The quality of the design and copywriting of these deliverables determines the actual share rate of customers and the conversion rate of new customers.

3.1. Recommended landing page design to maximize conversion rate

Most businesses make the mistake of running a single referral page, but high-converting programs have completely different goals. Two separate landing pagesoperates strategically.

3.1.1. Page 1: Advocate landing page (Goal: Maximize shares)

Target: ‘Existing customers’ (referrals) who are already using our products/services

Core components:

1. Clear Value Proposition (Headline): Make it clear what you can get as soon as you enter the page (e.g. “Gift 29,000 won to a friend and get 29,000 won!”)
2. A concise how-to for participation: Provides a simple visualization of the process for customers to follow in 3 steps (1. Share link → 2. Sign up/purchase as a friend → 3. Obtain rewards)
3. Key CTAs (Sharing Tools): Most prominently place a button to copy your personalized referral link/code. Provides buttons for direct sharing via KakaoTalk, email, SMS, and social media
4. My Recommendation Status Board (Personal Dashboard): Gamification elements are added by visually showing the number of friends I have invited, how many of them have joined, and how much compensation has been accumulated to date.

3.1.2. Page 2: Friend landing page (Goal: New conversions)

Target: ‘New visitors’ (referred people) who came in after clicking on the referral link

Core components:

1. Personalized Trust: The most important factor. You should remind these visitors that they came from an invitation from a 'friend' and not from an advertisement (e.g., "[Referrer's name] shared a 20% discount offer with you.") Reduce bounce rates by building immediate trust
2. Clear Benefits (The Offer): Make it clear what you can get from this page (e.g. “Get 20% off your first purchase immediately”)
3. Key product/brand introduction (Context): These new visitors may know nothing about our brand or products. Concisely include the product's core value proposition, representative features, customer reviews, etc. so that the recommender can understand why the product was recommended.
4. Powerful CTA (Conversion): Place strong call-to-action buttons so there’s no reason to hesitate (e.g. “Get started with a discount”, “Get my 20% off”)

3.1.3. Best practices for landing page design and copywriting

• design: Designed to be mobile-first. Most social sharing happens on mobile. Minimize user cognitive load by using a single-column layout, clear buttons that are easy to click, and sufficient white space between elements.
• Copywriting: Focus on ‘benefits’ rather than ‘functions’. Directly present a positive outcome for the customer, such as “Get a free massage” (O) instead of “Join a referral program” (X)
• Social Proof: Citing strong testimonials from other customers and highlighting key points is a proven way to maximize conversion rates.

3.2. Message templates that inspire immediate action

Customers are busy and ignore most marketing messages. Your referral request message should be clear, personalized, and very easy to act (share) right away.

3.2.1. Email templates: contextual subject and body

Subject Line Formula:

• The Gift: “A gift for your friend (and 29,000 won) has arrived”, “Share the love (and receive 29,000 won)”
• The Personal: “[Customer Name], your friends will appreciate this news.”
• The Clear: “Recommend a friend and get 36,250 won”, “How to get your next purchase for free”

[Template] First invitation email to satisfied customers (NPS 9-10 points):

3.2.2. DM / KakaoTalk message template

KakaoTalk can send promotions to customers through 'Friend Talk' (advertising messages), and the KakaoTalk Message API supports pre-defining and using various templates suitable for recommendation purposes, such as feed type, list type, and commerce type.

The key is to prepare a message (Preshared Message) in advance for the recommender to ‘copy and paste’. This message should be brief, informal (since you're sending it to a friend), and include a link.

[Template] KakaoTalk/DM sent by recommender to friend:

Part 4: Performance Measurement and Optimization Framework

As Peter Drucker said, “If you can’t measure it, you can’t manage it,” a clear dashboard and KPI (key performance indicator) are needed to determine whether the built recommendation engine is working as intended and whether it is a ‘growth’ engine or a ‘cost’ engine.

4.1. Define and calculate key performance indicators

4.1.1. KPI 1: Referral Rate among New Inflows

• definition: The percentage of ‘customers brought in through referral programs’ among ‘all new customers’ generated during a specific period (month/quarter). The most basic indicator of the overall contribution and efficiency of the program
• Calculation formula: Referral Rate (%) = (Number of new customers brought in through referrals / Total number of new customers) × 100
• example: If the total number of new customers in the month of June 2025 was 20,000, and 400 of them came through referral links, the referral rate for that month would be 2.0%.

Benchmarks (as of 2025):

• Global average for all industries: 2.35%
• SaaS (B2B/B2C): Trust-based recommendations are active, so 3% to 5% above average is considered a healthy benchmark.
• meaning: If this percentage is significantly lower than the benchmark (e.g. 2%), it is a sign that customers are not aware of the program's existence (lack of promotions and triggers) or that the incentives are not attractive.

4.1.2. KPI 2: Lead time until first referral (Time to First Referral - TTFR)

• definition: Average time (days or weeks) from when a customer signs up (acquisition) to generating ‘first successful referral’ (i.e. receiving reward)
• Calculation formula: TTFR = [by specific cohort] Average (first referral success date - customer subscription date)

Benchmark:

• B2C (based on Aha Moment): The goal is ‘shortening’. If customer value is delivered quickly, set a goal of having the first referral within 7 to 10 days after signing up.
• B2B (trust-based): The goal is ‘occurrence’ itself. It is natural for B2B customers to withhold recommendations until sufficient trust has been built (e.g., more than 6 months) due to reputation issues with the recommender.
• meaning: In B2C, if your TTFR is unnecessarily long, it's a strong signal that your Aha Moment or NPS trigger is firing too late, or that your customers aren't aware of the program's existence.

4.1.3. KPI 3: Recommendation Spreading Speed ​​(Viral Coefficient: K-Factor)

• definition: A ‘viral’ metric that represents the number of new customers each customer brings on average. If the K-Factor exceeds 1, it theoretically means that the service can grow exponentially (Organic Viral Growth) without marketing costs.
• Calculation formula: K = i × c
  • i = Average number of invitations sent per customer (e.g. 1 customer sends invitations to an average of 5 people → i=5)
  • c = Conversion rate per invitation (e.g. 1 out of 5 invitees sign up → c=0.2 or 20%)
• example: If a customer sends invitations to an average of 5 people (i=5) and 20% of the invitees sign up (c=0.2), the K-Factor is 5 × 0.2 = 1

meaning:

• If i is low: Customer leaves the ‘Sharing’ stage. This means that the UX of the recommender landing page is complicated or the message template is not attractive.
• If c is low: I get a lot of invitations, but I don't get any 'conversions'. This means that the recommended person’s landing page is not persuasive or the recommended person’s benefits are weak.

4.2. Recommendation Reward Redemption Rate Dashboard Design

core: Redemption Rate is one of the most important indicators of the ‘health’ and ‘ROI’ of a referral program. This tracks the ratio of rewards actually ‘used’ by the customer (activation/repurchase) compared to rewards ‘issued’ by the company to the customer (cost/liability).

4.2.1. The importance of ‘compensation consumption rate’ (Why it matters)

• Measuring program attractiveness: If the burnout rate is significantly low, it means that customers do not feel the value of using the reward even though they received it (e.g., the discount rate is too low, the usage conditions are strict), the process of using the reward is complicated (e.g., the expiration period is too short), or they have forgotten the existence of the reward.
• Proof of real ROI: A high burnout rate means that customers are ‘re-visiting’ or ‘re-purchasing’ the service to use their rewards. Powerful data proving that the referral program directly contributes not only to the acquisition of new customers, but also to the engagement and retention of existing customers.
• Financial health management: From an accounting perspective, points or credits issued to customers (but not yet used) are recognized as a ‘liability’ of the company. Tracking burn rates is essential for predicting and managing when and at what cost this debt will be realized in the future.

4.2.2. Burnout rate calculation formula and benchmark

• Calculation formula: Redemption Rate (%) = (Quantity or amount of rewards used (exhausted) within the period / Total quantity or amount of rewards issued within the period) × 100
• example: 145,000,000 points were issued as referral rewards in one month, and if customers used (used up) 36,250,000 points during the same period, the burnout rate was 25%.

Benchmark:

• Regular email coupon: 1% to 5%
• Referral Program Reward: 5% to 20%
• Loyalty program rewards: 10% to 30%
• meaning: If your referral program’s rewards burn rate is less than 5%, it likely has attractiveness, complexity, or awareness issues and needs immediate review.

4.2.3. Recommendation engine KPI dashboard core composition plan