# The State of F&B Customer Retention in Southeast Asia 2026

Malaysia, Singapore and Vietnam — what a returning customer is worth, and why loyalty programmes keep failing

Restaurants across Southeast Asia are closing at the fastest rate in a generation while delivery platforms grow at double digits. This edition sets out what retention is worth in each market, why most loyalty programmes decay, and what a retention system looks like instead.

Singapore · Published 2026-09-15 · Last reviewed 2026-09-15
Author: Alvin Koh, Founder, PEKO
Publisher: Hungry Bear Internet Pte. Ltd. (PEKO)
Canonical: https://heypeko.com/reports/f-and-b-retention-sea-2026
Licence: https://creativecommons.org/licenses/by/4.0/

## How to read the evidence tags in this report

[M] Measured, third-party — a published figure from a government body, central bank, research house or news organisation, cited by number in the Sources section.

[P] PEKO merchant data — measured across PEKO venues in Vietnam, January 2024 to June 2026. These figures are Vietnamese. They are not extended to Malaysia or Singapore as measurements.

[E] Estimate or model — a worked example, projection or inference built by PEKO from [M] and [P] inputs. Assumptions are stated where the figure appears.

Merchant-level measurements in this edition are drawn from Vietnam. Every Malaysian and Singaporean number in this report is [M] or [E]; merchant counts by market are not published. The 2027 edition will add measured figures from all three markets.

Delivery-platform commission rates are not published by any platform in these markets. Every commission figure here is an independent estimate or a third-party report, cited as such, and may not reflect any specific operator's contract. Third-party names and marks belong to their owners. Corrections: research@heypeko.com.

## Executive summary

Southeast Asia's restaurants are closing at the fastest rate in a generation while the delivery platforms that sit between them and their customers grow at double digits. Those two facts are connected. This report is about the connection — and about the one lever most independent operators still control: whether the customer who came once comes back.

#### Ten findings, each tagged by evidence type:

1. The shakeout is real and simultaneous. Vietnam lost more than 50,000 F&B outlets in the first half of 2025 alone, a 7.1% contraction, with Hanoi and Ho Chi Minh City down more than 11% [M-1, M-2]. Singapore recorded 3,047 closures in 2024, the most since 2005, and 2,431 more between January and October 2025 [M-3, M-4]. Malaysia's operators face the same cost stack without the same headline count.
2. Delivery platforms grew through it. Regional food delivery GMV rose 18% to US$22.7 billion in 2025; Malaysia and Vietnam each grew 18–19% [M-5]. Vietnam's platform GMV reached US$2.1 billion, split almost evenly between ShopeeFood and GrabFood at 48% each [M-6].
3. Commission is unpublished and negotiated everywhere. No platform in the three markets publishes a rate card. Independent estimates cluster at 25–30% in Malaysia, where a third-party summary of ShopeeFood's merchant FAQ reports a standard 30% [M-7]; 25–30% is commonly cited in Singapore; 20–30% in Vietnam. The variance itself is a cost — operators cannot price against a number they cannot see.
4. Most first-time customers do not return. Across PEKO venues in Vietnam, 60–75% of first-visit customers had not returned within 90 days [P]. The pattern is structural, not a Vietnamese peculiarity: the customer had no bad experience and no reason to remember the venue.
5. The identity layer has arrived, and it is a payment rail. DuitNow QR volume more than doubled to 3 billion transactions in 2025 across 2.6 million acceptance points [M-8, M-9]. VietQR runs roughly 15 million transfers a day, nearly half of NAPAS volume [M-10]. Singapore's digital payment adoption stands at 92% [M-11]. For the first time, a small venue can identify a returning guest without asking for a phone number.
6. Cash still breaks the data. Payment-linked loyalty captures only the guests who pay digitally. In kopitiams, mamaks, hawker stalls and street-front quán, that gap can be half the room. Receipt capture and in-venue QR close it; card-linked loyalty does not.
7. Loyalty programmes fail for two structural reasons, in every market. Enrolment depends on a cashier remembering to ask, and point balances are invisible to the customer between visits. The two compound into a loop the venue cannot exit by changing the reward [P, E]. A working retention system removes staff from the loop entirely and adds mechanics that create the next visit — check-in streaks, referral rewards, visible tiers, one-tap feedback — rather than only recording the last one (Section 4).
8. Cashback is not retention. Cashback is a merchant-funded discount on every bill, usually delivered through a platform that owns the customer relationship. It moves margin from the venue to the platform and does not identify who is about to stop coming.
9. Prediction beats reminders, and an agent beats a dashboard. Venues using cadence-based lapse prediction and automated win-back saw repeat rate rise 8–15 percentage points within 90 days [P]. Rule-based 'inactive for 30 days' triggers fire after the customer has already been lost. The operators sustaining these results are not building segments by hand; they describe the campaign in a sentence and approve what an AI agent drafts.
10. Formalisation is accelerating the shift. E-invoicing mandates in all three markets, Vietnam's new Personal Data Protection Law, and Malaysia's amended PDPA are pushing small operators onto digital rails whether or not they wanted retention tooling. The venues that use the rails for identity — not just compliance — will keep more of their regulars.

## 1. Why retention, why now

### 1.1 The squeeze

Three forces are compressing independent F&B margins across the region simultaneously: input costs, rent and labour; a consumer who is spending more carefully; and intermediaries who take a share of every transaction they route.

In Vietnam, 69% of F&B businesses reported pressure from rising raw-material costs in 2025, and the market grew only 5.5% in revenue to about VND 726.5 trillion while outlet count rose 2% to roughly 329,500 — meaning the closures of the first half were backfilled by new entrants who now face the same economics [M-12]. In Singapore, closures averaged 307 a month in early 2025, up from 254 in 2024 and roughly 230 in 2022 and 2023, with operators citing rent, utilities, wages and weaker spending [M-3]. Full-service restaurant revenue fell 5.6% year on year in June 2025 while cafés and food courts held roughly flat [M-13] — the value formats are holding, the higher-cost formats are not.

Malaysia does not publish a comparable closure count, but its operators sit on the same rails: commissions negotiated outlet by outlet, GrabFood at roughly two-thirds of delivery share [M-14], and a digital-payments shift that has already made the economy cash-light [M-9].

### 1.2 Where the growth went

While venues closed, the platforms above them grew. Regional food delivery GMV, after two years of 5% growth in 2022 and 2023, re-accelerated to 13% in 2024 and 18% in 2025 to reach US$22.7 billion [M-5, M-15]. Momentum Works describes the sector as having entered 'a more mature and structurally clearer phase' with platforms 'doubling down on affordability-focused initiatives' [M-5]. Platform affordability initiatives are financed from platform revenue, of which merchant commission is a component.

| Market | 2025 delivery GMV growth | Leading platform share | Commission (independent estimates) |
| --- | --- | --- | --- |
| Malaysia | ~18–19% [M-5] | GrabFood ~65% [M-14] | 25–30%, negotiated; third-party summary of ShopeeFood FAQ reports 30% standard [M-7] |
| Singapore | Returned to growth after flat 2023 [M-5, M-16] | GrabFood ~63% [M-14] | 25–30% commonly cited [E] |
| Vietnam | ~19% to US$2.1B [M-6] | ShopeeFood 48%, GrabFood 48% [M-6] | 20–30%, negotiated [E] |

### 1.3 The retention gap

Delivery platforms are effective at one thing: putting a venue in front of a customer who has never heard of it. They are not designed for the second thing: making that customer remember the venue rather than the app. In the standard merchant arrangement, the customer's contact details and order history are held by the platform and typically not shared with the venue in a form the venue can use to reach the customer directly.

This is why retention — not acquisition — is the lever left in the operator's hand. Acquisition is increasingly rented. Retention is still owned, if the venue has the means to identify the customer and a reason for them to return.

## 2. The retention arithmetic

### 2.1 The rule that still holds

The foundational finding on retention economics — that a five-percentage-point improvement in customer retention lifts profit by 25% to 95% depending on industry — comes from Reichheld and Sasser's 1990 Harvard Business Review analysis for Bain & Company [M-17]. The mechanism is that a retained customer costs nothing to re-acquire, spends more over time, and refers others. Food service sits toward the higher end of the range because the purchase is frequent and the acquisition cost, once you count commissions and discounts, is high relative to ticket size.

### 2.2 What one lost regular costs — worked examples [E]

The examples below are models. They use conservative assumptions stated in full so an operator can substitute their own figures. None of these are PEKO measurements in the market shown.

| Assumption | Malaysia (RM) | Singapore (S$) | Vietnam (₫) |
| --- | --- | --- | --- |
| Average order value (café / casual) | RM18 | S$18 | 150,000₫ |
| Visits per month for a regular | 4 | 4 | 4 |
| Regular's expected tenure | 18 months | 18 months | 18 months |
| Gross margin on food & drink | 60% | 60% | 60% |
| Lifetime revenue of one regular | RM1,296 | S$1,296 | 10,800,000₫ |
| Lifetime gross margin of one regular | RM778 | S$778 | 6,480,000₫ |
| Commission on 4 platform orders at an illustrative 28% rate | RM20, with the customer relationship held by the platform | S$20, with the customer relationship held by the platform | 168,000₫, with the customer relationship held by the platform |

The point of the table is not the totals. It is the last row. Re-acquiring a regular through a platform is cheap per order and lasting per relationship: the venue pays commission again on every subsequent order that customer places through the platform, because the relationship is held in the app rather than by the venue.

### 2.3 The 'one more visit' lever [E]

A venue with 50 customers a day and 30% of visits from regulars does roughly 450 regular-visits a month. If retention tooling brings one in ten regulars back one extra time a month — a lift of about 2.6 percentage points in repeat rate — that is roughly 45 additional visits, or RM810 / S$810 / 6,750,000₫ a month in incremental revenue at the assumptions above. Measured lift in Vietnam ran higher than this (Section 7), but the conservative case already covers the cost of most retention software several times over.

> **What this section does not claim**
> It does not claim these lifts are guaranteed. It does not claim Malaysian or Singaporean venues will match Vietnamese results. It claims that the arithmetic of retention favours the operator strongly enough that the question is not whether to invest, but how to do it without a cashier's memory as the single point of failure.

## 3. Why loyalty programmes fail in Southeast Asia

Every market in this report has a graveyard of loyalty programmes: the paper stamp card in the drawer, the points module in the POS nobody looks at, the cashback app the owner stopped funding. The failures look different but share two causes.

### 3.1 The cashier bottleneck

In a POS-native programme, enrolment happens when the cashier asks for a phone number. Vendor documentation for the region's leading POS systems states this plainly: enrolment happens at the register, in the moment, and staff must be trained to answer balance questions [M-18]. At peak, with a queue, the cashier drops the ask to clear the line. After enough declines, the cashier stops asking altogether. The programme stops acquiring members and the owner does not find out, because nothing on the dashboard says 'nobody asked today'.

### 3.2 Blind points

In the same systems, the point balance is displayed on the staff-facing sales screen and redemption is performed by the cashier. The customer has no way to check their own balance, tier or vouchers between visits. They are playing a game whose score they cannot see, so they stop playing. A reward they do not know they are close to cannot pull them back.

### 3.3 The death spiral

The two failures compound. Customers who cannot see their points see no reason to join; staff who are declined stop asking; a programme with no new members and no visible rewards produces no return visits; the owner concludes that 'loyalty doesn't work here' and cancels. The programme's failure is then cited as evidence that the customer base is not loyal — when the customer base was never given a working instrument.

### 3.4 Two newer failure modes

#### Cashback as loyalty

Several regional platforms sell cashback as a retention product. The venue funds a percentage of every bill — commonly 5–15% — which is returned to the customer as credit inside the platform's app. The economics are unambiguous: it is a discount on every transaction, funded by the merchant, whose redemption often happens on the next visit regardless of whether that visit was at risk. The diagnostic is the redemption rate. When it approaches 100%, the programme has stopped discriminating between the customer who was leaving and the customer who was coming anyway — it is a discount scheme, not a retention scheme.

#### Platform-owned loyalty

When the loyalty balance lives inside a consumer app operated by a third party — a delivery platform, a cashback app, a coalition network — the venue is building the platform's customer relationship at its own expense. The customer is loyal to the app that holds the balance. If the venue leaves the platform, the balance and the relationship stay behind.

> **The test for any loyalty instrument**
> 1. Can a guest join without a staff member doing anything? 2. Can the guest see their own balance on their own phone? 3. Does the venue own the member list and can it export it? 4. Does the system tell the venue who is about to stop coming, before they stop? An instrument that fails the first two will hit the death spiral. One that fails the third is renting its customers. One that fails the fourth is a reminder service, not a retention system.

## 4. A retention system, not a loyalty programme: the PEKO model

This section describes how PEKO is built, because the design answers the failures in Section 3 directly and because 'loyalty platform' is the wrong category for it. A loyalty programme records the last visit. A retention system manufactures the next one. The difference is in the mechanics, and in who has to operate them.

### 4.1 Design principle: nothing depends on the counter

Every mechanic below runs without a staff member doing anything. That is not a convenience feature; it is the whole point. The cashier bottleneck and blind points (Section 3) are both consequences of routing the programme through the till. Take the till out of the loop and both disappear.

| Failure mode (Section 3) | What causes it | How PEKO removes it |
| --- | --- | --- |
| Cashier bottleneck | Enrolment only when staff remember to ask | Guest scans an in-venue QR or photographs the receipt. Anyone at the table can join — not only the payer. OCR reads receipts from any POS, so no integration is needed. |
| Blind points | Balance visible only on the staff sales screen | Balance, tier, stamps and vouchers live in the guest's WhatsApp (Singapore, Malaysia) or Zalo (Vietnam). The guest checks it between visits without asking anyone. |
| Death spiral | No new members + no visible reward = no return visits | Streaks, referrals and tiers give the guest a reason to return that exists before the next visit, not only after it. |
| Cashback trap | Merchant-funded discount on every bill | Rewards are earned on behaviour the venue chooses — visits, streaks, referrals — and released on the venue's rules, not on every transaction. |
| Platform-owned customer | Balance lives in a third-party app | The venue owns the member list and can export it at any time. The messaging channel is the guest's own WhatsApp or Zalo, not a marketplace. |
| Reminder, not prediction | Fixed 'inactive 30 days' rules fire too late | Cadence model learns each guest's rhythm and flags lapse risk 3–7 days before the usual return window closes. |

### 4.2 The mechanics that create the next visit

#### In-venue check-in with streaks

Any guest sitting in the venue scans a table QR to join and receives a joining reward. Scanning on consecutive days earns a streak bonus. This does two things a points scheme cannot: it enrols the whole table rather than the one person who paid, and it converts frequency itself into the reward. For high-frequency formats — kopi, cafés, billiard halls — the streak is the habit made visible.

#### Customer referral programme

Every guest has a personal referral code. When a friend joins through it, both are rewarded, and the venue sees exactly who referred whom. Referral turns the existing regular into an acquisition channel with a known cost — the reward — rather than an unknown one — the commission.

#### Membership tiers the guest can see

Tiers, perks and progress to the next tier are shown on the guest's own phone. A guest who can see they are two visits from the next tier behaves differently from one who has to ask a cashier.

#### Stamp cards in the messaging app

Buy-nine-get-one, but the card lives in WhatsApp or Zalo. Nothing to lose, no faked stamps, and every stamp is a logged visit that feeds the cadence model.

#### One-tap post-visit feedback

A single question after the visit. An unhappy guest reaches management immediately, before they reach a review site; a happy guest is prompted to leave a public review. This closes the loop most programmes leave open: the venue learns why a regular is drifting while there is still time to act, and converts satisfied regulars into the social proof that acquires the next one.

#### Ten automated guest-care messages

Welcome, points about to expire, voucher about to expire, tier reached, streak at risk, and the rest — ten message types with defined triggers, approved once by the owner and then sent without further attention. These are the messages a good manager would send if they had the time and the list. Most have neither.

#### Zero-commission direct ordering

A branded ordering page reached by QR — no app install — for pickup, dine-in and the venue's own delivery arrangement. Every order placed here carries no commission, and every customer who orders here is a known member rather than a platform record.

#### Fraud controls for receipt capture

Letting guests photograph their own receipts to earn points requires guardrails: rate limits, receipt-age and value thresholds, duplicate-image matching across the brand, failed-scan detection, and risk-tiered actions. These run automatically; the owner sees exceptions, not a queue.

### 4.3 Prediction and win-back

The cadence model learns each guest's personal visit rhythm from the logged visits above and flags lapse risk 3–7 days before that guest's usual return window closes. A win-back message is then drafted for that guest — name, last order, preferred channel, an offer sized to their history — and sent on the channel they read. Messages are capped per guest per month to protect deliverability and the relationship. The measured result across PEKO venues in Vietnam was a repeat-rate lift of 8–15 percentage points within 90 days [P].

### 4.4 The AI agent: prompt, approve, done

Every mechanic above has settings. In a conventional platform those settings are dozens of screens, dropdowns and reports — the reason so many programmes are configured once and never touched again. PEKO ships an AI agent inside the product that takes those actions on the merchant's behalf. The owner describes what they want in plain language; the agent drafts it in the same screen they would otherwise configure by hand; the owner clicks Approve. Nothing goes live without that click.

| The owner says | The agent does |
| --- | --- |
| “Give 20% off laksa to regulars who haven't ordered in three weeks, WhatsApp only, cap 500 redemptions.” | Creates the voucher with type, value, cap, eligibility, schedule, channel and audience filled in. Owner approves. |
| “Show me VIPs in Tiong Bahru who spent over S$200 last quarter and haven't visited in ten days.” | Builds and names the segment, shows the count, offers to save it or campaign to it. |
| “Win back the at-risk regulars this week with a free kopi.” | Picks the audience from the cadence model, writes the message per guest, sets the send time. Owner approves. |
| “Which day was slowest last month, and what did the regulars who usually come then do instead?” | Pulls the visit heatmap and cohort report and answers in a paragraph, with the numbers. |
| “Change the stamp card to buy-eight-get-one and tell members.” | Updates the loyalty rule and drafts the announcement message for approval. |

The agent works across every module — vouchers, segments, campaigns, loyalty rules, menus, points, reports. The practical consequence is who can run retention: a venue with no marketing hire and an owner who would never open a campaign builder can now operate a system that, a year ago, required a specialist. This is what makes self-serve viable for a small operator, and it is why PEKO does not need to route a new merchant through a sales call before the product is useful.

### 4.5 What PEKO is not

- Not a POS. It runs alongside whatever till the venue already uses, from cloud POS to a paper pad. (LOOP, from the same company, is the POS for venues that want one.)
- Not a payment processor. It reads the rails the venue already accepts and never holds funds.
- Not a delivery marketplace. It takes no commission and carries no orders; delivery remains the venue's own arrangement.
- Not a consumer app the venue's customers share with other brands. The member list is the venue's, exclusively.

> **The loop, end to end**
> Guest joins by QR or receipt without a staff ask → balance, tier and stamps appear in the guest's own messaging app → streaks, referrals and tiers give a reason to return before the next visit → a one-tap survey captures sentiment after it → the cadence model flags the guest who is drifting → a win-back message goes out on the channel they read → the repeat-rate lift is measured against subscription cost. The owner operates all of it by describing what they want to an agent and approving the draft.

## 5. Malaysia

### 5.1 Market structure

Malaysia's food delivery market is the region's fourth largest by GMV and one of its fastest-growing, expanding 18–19% in 2025 [M-5]. GrabFood held about 65% share in the most recent public breakdown, foodpanda about 30% [M-14]. ShopeeFood is the price challenger. No platform publishes a Malaysian rate card; commission is negotiated per outlet and independent estimates cluster at 25–30%, and a third-party summary of ShopeeFood's merchant FAQ reports a standard 30% covering delivery, marketing and platform [M-7]. Some third-party guides cite lower tiers of 15–22% for subscribed merchants [M-20]; treat those as the negotiated floor, not the default.

### 5.2 The identity rail

Malaysia has the region's most complete state-built payment rail for small merchants. DuitNow QR reached 2.6 million acceptance points by end-2024 [M-8] and transaction volume more than doubled to 3 billion in 2025, with Bank Negara describing the country as 'increasingly operating as a cash-light economy' [M-9]. Digital payment adoption exceeds 80% [M-21]. Account-to-account payments are forecast to reach 40% of online and 16% of point-of-sale value by 2030 [M-8].

For retention, this matters because a DuitNow QR payment carries a payer reference. A venue that consents its customers properly can recognise a returning guest at the moment of payment without a phone-number ask at the counter. The limit is cash: in mamak, kopitiam and hawker formats a material share of transactions remain cash, and any loyalty system that only sees the payment rail is blind to them.

### 5.3 The messaging channel

WhatsApp is the default communication channel for Malaysian consumers and small businesses alike, used by roughly nine in ten internet users [M-22]. This has two consequences for retention. First, SMS-based loyalty — still the dominant model among incumbent local vendors — is competing against the channel the customer actually reads, and charging per message to do so. Second, the WhatsApp Business API has a per-conversation cost that is small but real; operators should require any vendor to state whether messaging is included, passed through at cost, or marked up.

### 5.4 Regulatory backdrop

The Personal Data Protection Act 2010, as amended in 2024, now carries mandatory breach notification and data protection officer obligations for organisations processing personal data at scale, with the amendments phased in from 2025. Any loyalty programme collecting phone numbers or payment references needs documented consent and a one-tap opt-out. LHDN's e-invoicing mandate is being phased through 2026 by revenue band, with the smallest businesses currently exempted; POS vendors have made e-invoice compliance a primary hook, which is pulling the last cash-only venues onto digital tills.

### 5.5 Vendor landscape

| Vendor | Model | Buying motion | Structural note |
| --- | --- | --- | --- |
| StoreHub (Membership / Beep Cashback) | POS with loyalty add-on; cashback mechanic | Demo-gated; no self-serve signup | Loyalty available only to StoreHub POS customers; Beep delivery at 9% commission [M-23] |
| Qashier Treats | POS + payments; payment-linked points add-on ~RM40/mo | Self-serve, free Lite tier | Earns on card / QashierPay transactions; cash customers not captured [M-24] |
| SimpleLoyalty (Mulah) | Loyalty-only; SMS-first points and broadcasts | Quote-based; pricing not published | Confirm messaging fees and POS connection terms in writing |
| Advocado | Loyalty CRM with human Growth Consultant | Quote-based, Malaysian entity | High-touch model; strong comparison content |
| Eber | Enterprise loyalty CRM, USD-priced | Scheduled call | From US$69/mo annual; multi-store adds US$89/mo [M-25] |
| Loyverse | Free POS with basic points | Self-serve | No campaigns, no prediction; recommended by StoreHub's own restaurant guide for very small single-outlet venues [M-23] |
| PEKO | Retention system: QR/receipt self-enrolment, streaks, referrals, tiers, surveys, 0% ordering, AI churn prediction, prompt-driven AI agent | Self-serve, free to 300 members | Founding cohort open in Malaysia; merchant counts by market not published |

### 5.6 What is distinctive about Malaysia

- The market leader has, in its own published guide, pointed very small single-outlet venues toward a free POS alternative rather than its own product, and suggested that a loyalty programme is not the reason regulars return [M-23]. Whatever the merits of that view for StoreHub's product, it leaves the largest addressable segment in the country without a retention instrument from the vendor most likely to reach it.
- Retention is sold as an add-on line item — RM40 here, cashback there — never as the product. No vendor with volume positions prediction as the offer.
- Language is under-served. Incumbents publish Bahasa Malaysia content on generic SME topics (licences, how to open a shop). Nobody publishes BM content on retention. Tamil and Chinese-language customer messaging is claimed by several vendors and demonstrated by few.
- Cash matters more than in Singapore and less than in Vietnam. A retention system that ignores cash guests will under-count the regulars in exactly the formats that most need help.

## 6. Singapore

### 6.1 Market structure

Singapore is the region's most saturated and most expensive F&B market. Around 13,959 F&B establishments generated S$9.1 billion in operating revenue in the most recent Department of Statistics industry profile, with purchases, remuneration and rent together accounting for roughly 80% of business costs [M-26]. Closures reached 3,047 in 2024, the highest since 2005, and 2,431 between January and October 2025 against 3,357 openings [M-3, M-4]. One analysis of tax filings suggests a majority of closed outlets never recorded a profit [M-27]. Prime-area rents averaged about S$8,500 a month in 2023 [M-13].

GrabFood held about 63% of delivery share and foodpanda 28% in the last public breakdown [M-14]; the market was flat in 2023 and returned to growth thereafter [M-5, M-16]. Commission rates are not published; 25–30% is commonly cited by operators. A distinctive Singaporean feature is the direct-ordering middle layer: platforms that let a venue take orders off the aggregators but still charge a percentage per order or per queue entry plus enrolment fees. Operators who leave GrabFood for one of these have reduced, not eliminated, their per-order cost.

### 6.2 The identity rail

Digital payment adoption stood at 92% in 2025 [M-11]. PayNow and SGQR are universal; SGQR overlays multiple schemes on one code, and PayNow now connects cross-border to India's UPI, Thailand's PromptPay, Malaysia's DuitNow and Indonesia's QRIS [M-28]. The consequence for retention is that a Singaporean venue has the cleanest payment-based identity signal in the region — and the smallest cash gap, though hawker centres remain a partial exception.

### 6.3 Grants and the buying motion

The Productivity Solutions Grant subsidises pre-approved digital solutions for eligible SMEs, and several loyalty vendors sell primarily on their PSG status. Two cautions. First, PSG eligibility is a property of a specific pre-approved solution at a specific time; it is not transferable and should be verified on the GoBusiness portal at the point of application, not taken from vendor collateral. Second, a grant that halves the price of a programme that fails the four-question test in Section 3 halves the cost of a programme that will not work.

### 6.4 Regulatory backdrop

The Personal Data Protection Act 2012 governs consent, purpose limitation and the Do Not Call registry; a loyalty programme sending marketing messages to a phone number needs clear consent and must honour opt-outs. InvoiceNow e-invoicing is being phased in for GST-registered businesses, adding another compliance-driven push toward digital records.

### 6.5 Vendor landscape

| Vendor | Model | Buying motion | Structural note |
| --- | --- | --- | --- |
| Advocado | Loyalty CRM; phone-number enrolment; human advisory | PSG-listed paid plans | Incumbent by outlet count; no free tier |
| Eber | Enterprise loyalty CRM, USD-priced | Scheduled call | Best fit for multi-brand groups with a CRM team [M-25] |
| Oddle | Direct ordering + reservations + CRM | From ~S$300/mo on annual commitment | Usage fees beyond plan caps; ordering channel is Oddle's |
| Stampede | Guest WiFi capture + marketing | Self-serve per-outlet pricing | Strongest where WiFi is the capture surface (pubs, bars) |
| PEKO | Retention system: QR/receipt self-enrolment, streaks, referrals, tiers, surveys, 0% ordering, AI churn prediction, prompt-driven AI agent | Self-serve, free to 300 members; not PSG-listed | Founding cohort open in Singapore; merchant counts by market not published |

### 6.6 What is distinctive about Singapore

- The cost base is so high that a retention programme's payback is measured in single-digit orders, not months. At S$18 average order and S$49 a month, three recovered visits cover the software.
- The buyer is time-poor and sophisticated. Singaporean operators will pay for setup rather than do it themselves, and they discount vendor claims by default. Reference deployments matter more than feature lists.
- The grant distorts the market. Vendors compete on PSG status rather than outcome, which suppresses honest comparison. An operator's first question should still be 'will my customers see their own points', not 'is it subsidised'.
- Cross-border customers are real. Johor Bahru residents working in Singapore and Singaporeans dining in JB show up in both markets' data; a venue near the Causeway should expect frequency benchmarks to look odd unless those guests are segmented.

## 7. Vietnam

### 7.1 Market structure

Vietnam is the region's smallest delivery market by GMV and among its fastest-growing, reaching US$2.1 billion in 2025, up 19%, with ShopeeFood and GrabFood each at 48% and beFood at 4% [M-6]. The wider F&B market reached about VND 726.5 trillion in 2025 revenue, up 5.5% [M-12]. The outlet base is enormous and fragmented: roughly 329,500 outlets at end-2025, of which fewer than 5,000 belong to modern chains [M-12, M-2]. The first half of 2025 saw more than 50,000 outlets close — a 7.1% contraction, the second major shakeout since early 2024 — with Hanoi and Ho Chi Minh City each down more than 11% [M-1, M-2]. Consumers did not stop eating out: 54% intended to hold their F&B budgets and 68% of operators still planned expansion [M-1]. The market is filtering weak models, not shrinking demand.

### 7.2 The identity rail and the channel

VietQR is the interbank QR standard operated by NAPAS. NAPAS processes about 35 million transactions a day, nearly half of them VietQR, and QR payments grew 62% in volume and 151% in value in 2025 according to the State Bank of Vietnam [M-10, M-29]. In 2025 NAPAS began rolling out a person-to-merchant QR standard, which is the rail most relevant to F&B [M-10].

The messaging channel is Zalo, used by more than 75 million Vietnamese, with Zalo Official Accounts and ZNS (Zalo Notification Service) providing a templated, consented transactional messaging layer at a per-message cost well below SMS. Zalo Mini Apps allow a venue to run ordering and loyalty inside the app without an install. Vietnam is therefore the one market in this report where the messaging channel and the loyalty surface can be the same thing.

### 7.3 Regulatory backdrop

Decree 13/2023 established Vietnam's first comprehensive personal data protection rules; the Law on Personal Data Protection passed in 2025 and effective from 1 January 2026 supersedes it with a fuller consent, transfer and enforcement regime. Separately, Decree 70/2025 required household businesses above a revenue threshold to issue e-invoices from cash registers from 1 June 2025 — a change that has pushed hundreds of thousands of quán and cafés onto digital tills for the first time, and that operators in the iPOS/Nestlé survey now count alongside labour and tax as part of the cost of a formalising market [M-12].

### 7.4 What PEKO's Vietnamese data shows [P]

The figures in this section are measured across PEKO venues in Vietnam between January 2024 and June 2026. They are the only merchant-level measurements in this report. They are Vietnamese, and we do not present them as expectations for Malaysia or Singapore.

| Measure | Value | Basis |
| --- | --- | --- |
| First-visit customers not returning within 90 days | 60–75% | Venue-level range across the sample |
| Repeat-rate lift within 90 days of activating cadence-based win-back | +8–15 percentage points | Venue-level range; platform-level claim |
| Lead time at which lapse risk is flagged | 3–7 days before the customer's usual return window closes | Model output |
| Customers recovered per month, mid-sized venue | 30–45 | Count of guests returning after a win-back message, not a rate |
| Transaction history needed before prediction reaches target precision | ~14 days | Model calibration |
| Venues onboarded | 3,000 | Cumulative signups, including free tier |

Two observations from the data that we believe generalise, subject to measurement in the other markets:

- Lapse is personal, not calendrical. A customer who comes every nine days and has not come for eighteen is at risk; a customer who comes monthly and has not come for eighteen days is not. Fixed 'inactive for 30 days' rules, which most rule-based systems use, fire too late for the first customer and too early for the second.
- Enrolment without a staff ask changes the base. Venues where guests self-enrol by QR or receipt capture accumulate members at peak hours — exactly when cashier-dependent programmes stop enrolling.

### 7.5 What is distinctive about Vietnam

- Scale and fragmentation. Three hundred thousand outlets, almost all independent, most on cheap POS with a loyalty module nobody uses. The retention opportunity is broad and shallow.
- The channel is native. Zalo gives Vietnam a messaging-plus-loyalty surface that WhatsApp cannot yet match in Malaysia or Singapore.
- Billiard halls and entertainment venues behave differently from cafés: higher visit frequency, group guests who are not the payer, and no incumbent retention vendor. In-venue QR check-in, where anyone at the table can join, outperforms payment-linked enrolment in these formats.
- Formalisation is the tailwind. The e-invoice mandate did more to digitise small F&B in one year than a decade of POS marketing.

## 8. Cross-market benchmarks

Every cell is tagged. Empty cells mean no reliable public figure exists; we would rather leave a gap than invent a number.

| Metric | Malaysia | Singapore | Vietnam |
| --- | --- | --- | --- |
| Delivery GMV growth 2025 | ~18–19% [M-5] | Positive, below regional average [M-5] | ~19% to US$2.1B [M-6] |
| Leading platform share | GrabFood ~65% [M-14] | GrabFood ~63% [M-14] | ShopeeFood 48% / GrabFood 48% [M-6] |
| Typical commission | 25–30% [M-7, E] | 25–30% [E] | 20–30% [E] |
| Digital payment adoption | >80% [M-21] | 92% [M-11] | QR payments 41% of e-commerce, 29% of POS value [M-30] |
| National QR rail | DuitNow QR — 3B txns 2025, 2.6M points [M-8, M-9] | PayNow / SGQR [M-11] | VietQR — ~15M transfers/day [M-10] |
| Primary messaging channel | WhatsApp | WhatsApp | Zalo (75M+ users) |
| Outlet closures, latest period | Not published | 3,047 in 2024; 2,431 Jan–Oct 2025 [M-3, M-4] | >50,000 in H1 2025, −7.1% [M-1] |
| First-visit non-return, 90 days | Not measured | Not measured | 60–75% [P] |
| Repeat-rate lift from predictive win-back, 90 days | Not measured | Not measured | +8–15 pts [P] |
| Data protection regime | PDPA 2010, amended 2024 | PDPA 2012 | PDP Law 2025, in force 2026 |
| E-invoicing | LHDN phased through 2026 | InvoiceNow phased | Decree 70/2025 from June 2025 |

## 9. Channel economics: what each route to the customer really costs

Operators tend to compare channels on commission alone. The fuller comparison includes who owns the customer afterwards, because that determines whether the next order costs commission again.

| Channel | Per-order cost | Who owns the customer | Retention implication |
| --- | --- | --- | --- |
| Delivery aggregator | 20–35% commission — independent estimates; rates unpublished and negotiated [M-7, E] | Platform | Repeat orders placed through the platform incur commission again; customer contact details typically not shared with the venue |
| Direct-ordering middle layer (SG) | Percentage or per-unit fee plus enrolment | Shared; channel is the vendor's | Lower cost, still per-order; customer data terms vary |
| Cashback platform | Cashback funded by venue (5–15%) plus platform fee (~1–2%) | Platform app | Discount on every bill; balance lives in the app |
| POS loyalty module | Included or add-on subscription | Venue, but staff-mediated | Fails on cashier bottleneck and blind points |
| SMS-based loyalty | Subscription plus per-message | Venue | Channel the customer may not read; cost scales with sends |
| Own direct ordering + WhatsApp/Zalo loyalty | Flat subscription; messaging at cost | Venue | 0% per order; requires self-enrolment to reach scale |

The direct route is not free — it costs a subscription, the venue's own delivery arrangement, and the discipline to put a QR on every table and every bag. What it does not cost is a share of every future order from a customer the venue has already won.

## 10. Five shifts that define 2026

### 1. Identity moved to the payment rail

DuitNow QR, PayNow and VietQR each carry a payer reference. With consent, that reference recognises a returning guest without a counter ask. This removes the cashier bottleneck for digital payers — and leaves cash payers behind unless a second capture method exists.

### 2. Loyalty moved into the messaging app

The stamp card the customer cannot lose is the one that lives in WhatsApp or Zalo. Balance visibility — the cure for blind points — is now a message, not a portal. Vendors still on SMS are paying more to be read less.

### 3. Prediction replaced the reminder

Rule-based triggers ('inactive 30 days') are being replaced by cadence models that learn each customer's rhythm and flag deviation. The measured difference in Vietnam was a repeat-rate lift of 8–15 points within 90 days [P]. The mechanism is timing: the message arrives while the habit is still warm.

### 5. The AI agent replaced the campaign builder

The owner who cannot spend Monday building segments now describes the campaign in a sentence and approves a draft. This matters less for what it does than for who can do it: a venue with no marketing hire can now run retention that previously required one.

### 6. Compliance became the on-ramp

E-invoicing mandates in all three markets and new data protection law in Vietnam and Malaysia are forcing digital records on the smallest venues. The operators who treat those records as an identity layer — not just a tax file — get retention as a by-product of compliance.

## 11. What operators should do

### 11.1 Single outlet, under 500 regulars

1. Put enrolment on the table, not the counter: a QR any guest can scan, and a receipt-capture path for cash payers.
2. Make the balance visible in the channel the customer reads — WhatsApp or Zalo — not on the staff screen.
3. Start free. A programme that cannot prove a second visit inside 60 days on a free tier will not do it on a paid one.
4. Measure one number: second-visit rate among newly enrolled guests. Everything else is downstream.

### 11.2 Small group, 2–10 outlets

1. Unify identity across outlets before adding mechanics. A member who earns at one branch and cannot redeem at another has been given a reason to leave.
2. Replace fixed inactivity rules with per-customer cadence. Ask any vendor to show, on your data, who they would message this week and why.
3. Keep the POS. The cost of switching POS to get loyalty exceeds the cost of a POS-agnostic layer in every case we have modelled.
4. Track recovered gross margin after reward and message cost, by outlet, with a holdout group.

### 11.3 Chain, 10+ outlets

1. Decide who owns the customer before choosing a platform. Coalition and cashback networks bring reach and take exclusivity.
2. Budget the team if the platform needs one. An enterprise CRM without an operator is shelfware with a renewal date.
3. Insist on export. Member profiles, consent timestamps, balances and campaign outcomes, in a common format, at any time.
4. Pilot in one cluster for two full purchase cycles before rolling out; separate new, active, at-risk and lapsed cohorts in the readout.

### 11.4 The four questions to ask any vendor

Can a guest join without staff doing anything? Can the guest see their own balance on their own phone? Do I own and can I export the member list? Does the system tell me who is about to stop coming, before they stop? A vendor who cannot answer all four with a demonstration on your own data is selling a reminder service or a discount scheme, whatever the brochure calls it.

## 12. Methodology, provenance and limitations

### 12.1 How this report was built

Market and platform figures were drawn from published sources — central banks, statistics departments, Momentum Works, iPOS.vn/Nestlé Professional, and named news organisations — and are cited by number. Vendor facts were taken from each vendor's own published pages at the date of writing; where a figure comes from a competitor's analysis it is labelled as third-party and unverified. PEKO merchant figures are measured across PEKO venues in Vietnam from January 2024 to June 2026 and are reported as ranges rather than point estimates because venue-level variance is large and the sample is not stratified by format.

### 12.2 Limitations

- Merchant-level measurements in this edition are drawn from Vietnam. Malaysian and Singaporean retention figures are estimates or modelled examples with stated assumptions. Merchant counts by market are not published.
- Commission rates in all three markets are unpublished and negotiated. Ranges are independent estimates and may not reflect any specific operator's contract.
- The Vietnamese sample over-represents cafés, casual dining and billiard venues in Ho Chi Minh City and Hanoi; it under-represents fine dining and East Malaysia-equivalent secondary cities.
- 'Venues onboarded' counts cumulative signups including free-tier accounts; it is not an active-venue count and should not be read as one.
- PEKO is a vendor in the category this report describes. We have listed competitors' strengths and our own limitations, and we do not publish a ranking of ourselves. Readers should treat Sections 5.5, 6.5 and any PEKO-favourable inference with the scepticism they would apply to any vendor-authored research.

### 12.3 What the 2027 edition will add

Measured first-visit non-return, repeat-rate lift and recovered-visit counts from Malaysian and Singaporean venues; cash-versus-digital enrolment shares by format; and a public holdout-group methodology so that lift claims from any vendor, including us, can be reproduced.

### 12.4 How to cite

PEKO (2026). The State of F&B Customer Retention in Southeast Asia 2026: Malaysia, Singapore, Vietnam. Hungry Bear Internet Pte. Ltd., Singapore. Available at heypeko.com/reports. Licensed CC BY 4.0.

## Questions this report answers

**How many F&B outlets closed in Southeast Asia in 2025?**

Vietnam lost more than 50,000 F&B outlets in the first half of 2025 alone, a 7.1% contraction, with Hanoi and Ho Chi Minh City down more than 11% [M-1, M-2]. Singapore recorded 3,047 closures in 2024, the most since 2005, and 2,431 more between January and October 2025 [M-3, M-4]. Malaysia's operators face the same cost stack without the same headline count.

**What commission do food delivery platforms charge in Malaysia, Singapore and Vietnam?**

No platform in the three markets publishes a rate card. Independent estimates cluster at 25–30% in Malaysia, where a third-party summary of ShopeeFood's merchant FAQ reports a standard 30% [M-7]; 25–30% is commonly cited in Singapore; 20–30% in Vietnam. The variance itself is a cost — operators cannot price against a number they cannot see.

**Why do restaurant loyalty programmes fail?**

Loyalty programmes fail for two structural reasons, in every market. Enrolment depends on a cashier remembering to ask, and point balances are invisible to the customer between visits. The two compound into a loop the venue cannot exit by changing the reward [P, E].

**What percentage of first-time restaurant customers never return?**

Most first-time customers do not return. Across PEKO venues in Vietnam, 60–75% of first-visit customers had not returned within 90 days [P]. The pattern is structural, not a Vietnamese peculiarity: the customer had no bad experience and no reason to remember the venue.

**How much does one lost regular customer cost a restaurant?**

A venue with 50 customers a day and 30% of visits from regulars does roughly 450 regular-visits a month. If retention tooling brings one in ten regulars back one extra time a month — a lift of about 2.6 percentage points in repeat rate — that is roughly 45 additional visits, or RM810 / S$810 / 6,750,000₫ a month in incremental revenue at the assumptions above. Re-acquiring a regular through a platform is cheap per order and lasting per relationship: the venue pays commission again on every subsequent order that customer places through the platform, because the relationship is held in the app rather than by the venue.

## Data

- https://heypeko.com/data/fnb-retention-sea-2026-findings.json
- https://heypeko.com/data/fnb-retention-sea-2026-findings.csv
- https://heypeko.com/data/loyalty-failure-modes-2026.json
Licensed CC BY 4.0. Attribution: PEKO (Hungry Bear Internet Pte. Ltd.), 2026.

## Cite this report

PEKO (2026). The State of F&B Customer Retention in Southeast Asia 2026. Hungry Bear Internet Pte. Ltd. https://heypeko.com/reports/f-and-b-retention-sea-2026

## Downloads

- Download PDF: https://heypeko.com/reports/peko-f-and-b-retention-sea-2026.pdf
- Download Word (DOCX): https://heypeko.com/reports/peko-f-and-b-retention-sea-2026.docx

---

## Sources

<!-- BEGIN SOURCES -->

[M-1] iPOS.vn and Nestlé Professional, Vietnam F&B Market Report H1 2025, October 2025, as reported by Vietstock, 14 October 2025.
[M-2] B&Company, 'The trend of mass closure of F&B stores in the first half of 2025', citing iPOS.vn.
[M-3] Reuters, 'The bill's too high: why Singapore's F&B scene is losing flavour', 10 April 2025 (closures per month, 2022–2025); Malay Mail, 20 July 2025 (3,047 closures in 2024).
[M-4] Ministry of Trade and Industry Singapore data, January–23 October 2025, as cited by Funding Societies and Indonesia Business Post: 3,357 openings, 2,431 closures.
[M-5] Momentum Works, Food Delivery Platforms in Southeast Asia 6.0, January 2026, as reported by TechNode Global, 28 January 2026.
[M-6] Momentum Works, as reported by The Investor (Vietnam), 29 January 2026: Vietnam delivery GMV US$2.1B, ShopeeFood 48%, GrabFood 48%, beFood 4%.
[M-7] Eats365, 'Food Delivery Platforms in Malaysia 2026', citing ShopeeFood merchant FAQ (30% standard commission) and independent estimates for GrabFood and foodpanda.
[M-8] Global Payments / Worldpay via Fintech News Singapore, 'Southeast Asia Payment Methods in 2026': DuitNow QR 2.6M acceptance points end-2024; A2A forecast 2030.
[M-9] Bank Negara Malaysia, Governor's keynote, as reported by TechNode Global and Fintech News Malaysia, May 2026: DuitNow QR volume more than doubled to 3 billion in 2025.
[M-10] VnEconomy, 'Opportunities for QR payment growth', 2025: NAPAS ~35M transactions/day, ~15M VietQR transfers/day; P2M QR standard rollout.
[M-11] PwC Singapore, Payments' State of Play 2026: digital payments adoption 92.0% in 2025.
[M-12] B&Company, 'F&B franchise in Vietnam: why 2026 is becoming a more selective expansion phase', citing iPOS.vn: 2025 revenue VND 726.5T, outlets ~329,500, 69.38% reporting raw-material cost pressure.
[M-13] Market Research Singapore, 'Pressure on the Plate', citing Department of Statistics Singapore F&B Services Index June 2025 and Knight Frank / Zipdo cost data.
[M-14] Momentum Works, Food Delivery Platforms in Southeast Asia 4.0, January 2024, as reported by Katadata: platform shares by market, 2023.
[M-15] Momentum Works, Food Delivery in Southeast Asia 5.0, 2025: 2024 GMV US$19.3B, +13%.
[M-16] Momentum Works, Food Delivery Platforms in Southeast Asia 4.0 announcement: Singapore topline flat in 2023.
[M-17] Reichheld, F. F. and Sasser, W. E. (1990). 'Zero Defections: Quality Comes to Services', Harvard Business Review, September–October 1990.
[M-18] Public help documentation of KiotViet and iPOS (Vietnam) on loyalty enrolment and balance display, as catalogued in PEKO's Loyalty Transparency Standard, 2026.
[M-19] Advocado, 'Advocado vs Simple Loyalty (Mulah Rewards) Malaysia', April 2026 — competitor-authored comparison; consulted for market context only, no figures from it are relied on in this report.
[M-20] klikit, 'Food Delivery Commission Fees in Malaysia (2026)': subscription-tier commission ranges.
[M-21] Tazapay, 'Local Payment Methods and E-Wallets in Southeast Asia', 2026: Malaysia digital payment adoption >80%.
[M-22] DataReportal, Digital 2025: Malaysia and Singapore — WhatsApp usage among internet users aged 16–64.
[M-23] StoreHub, 'POS System for Restaurant Malaysia' (blog) and storehub.com/my/pricing, accessed September 2026.
[M-24] Qashier, qashier.com/my/software-pricing and F&B pricing pages, accessed September 2026.
[M-25] Eber, eber.co/pricing, accessed September 2026.
[M-26] Department of Statistics Singapore, Food & Beverage Services industry infographic (reference year 2021, updated March 2023).
[M-27] Rubbish Eat Rubbish Grow, 'Why so many F&Bs are closing down in Singapore', December 2025, citing tax-declaration analysis — secondary source; treat the 82% figure as indicative.
[M-28] Antom / AMRO, Southeast Asia Payment Trends 2026: regional QR interoperability.
[M-29] State Bank of Vietnam, 2025 QR payment growth (62% volume, 151% value), as reported by Fintech News Singapore.
[M-30] Digital in Asia, Asia Digital Payments Tracker 2026: VietQR share of e-commerce and POS value, 2025.
[P] PEKO merchant data, Vietnam, January 2024 – June 2026. Methodology at heypeko.com/methodology.
[E] PEKO estimates and worked models; assumptions stated at the point of use.

<!-- END SOURCES -->

## Related answers

- [Best restaurant loyalty software in Singapore](https://heypeko.com/answers/best-restaurant-loyalty-software-singapore) — Vendor-by-vendor comparison of the loyalty platforms operating in Singapore, with pricing models and the trade-off each one makes.
- [Why loyalty programmes fail](https://heypeko.com/answers/why-loyalty-programs-fail) — The mechanism behind the retention decay this report measures, explained end to end.
- [Loyalty without asking the cashier](https://heypeko.com/answers/loyalty-without-asking-the-cashier) — Why enrolment at the counter caps programme reach, and what removes that cap.

## Related research

- Báo cáo Giữ chân Khách hàng F&B Việt Nam 2026: https://heypeko.com/vi/bao-cao/giu-chan-khach-hang-fnb-viet-nam-2026
- Loyalty Transparency Standard 2026: https://heypeko.com/loyalty-transparency-standard

---

PEKO is operated by Hungry Bear Internet Pte. Ltd..
https://heypeko.com/reports/f-and-b-retention-sea-2026