Cash on delivery does not describe every way a customer pays

A World Bank survey shows why sellers should separate payment for a delivery from phone & card use in stores.

Original Influence IQ analysis of 4 World Bank Global Findex country releases, based on interviews in 2024.

Abstract

Choosing cash on delivery does not necessarily mean someone avoids digital payments elsewhere. Within the survey’s account-screened group, our analysis of 2024 responses found a weighted share reporting in-store phone or card payment of 32.1% in Indonesia, 55.0% in Malaysia, 34.3% in the Philippines & 67.6% in Thailand. Each figure concerns respondents who bought something online for delivery, usually paid cash on arrival, & answered yes or no to the in-store payment question. These are percentages of people, not orders or spending.[1][2][3][4]

The questions came after an account, debit-card or mobile-money screening step, so the results do not describe all online shoppers. We analyzed each country separately using the World Bank’s Global Findex 2025 release, which contains interviews from 2024. We also examined weighting & considered payment frequency where the smaller groups supported it. For businesses & creators selling products, the implication is practical: ask about the purchase in front of the customer before treating a payment choice as a description of that person.

A small business receives an order marked “cash on delivery.” The label is useful for arranging payment. It becomes less useful when the team turns it into a customer profile: someone who does not use a card, needs a lesson in mobile payments or will reject every online payment option.

That is an illustrative situation, not an observed customer story. It captures the question behind this analysis: how much can a seller infer from the way someone usually pays for delivered online purchases?

We looked for a direct answer in responses from the same people. Among those who reported usually paying cash when an online purchase arrived, did any also report using a phone or card to pay in a store? In each of the 4 country samples, they did. The overlap gives sellers a reason to keep those 2 contexts separate.

2 purchases, 2 payment contexts

A parcel & illustrated cash sit beside a front door; a shopping bag, payment terminal, phone & card appear alongside them.
The payment chosen for a delivery & the payment used in a store are separate questions. Conceptual illustration, not survey data.

Who is included in the cash-on-delivery comparison?

The comparison begins with a specific sequence of questions. Respondents said they had personally used a phone or computer to buy something online that was delivered during the previous 12 months. They then said they usually paid in cash when it arrived. Purchases arranged solely through a phone call were outside the online-shopping question.[5]

“Usually” matters. The survey did not establish that these people always used cash. People who answered “both” to the usual-payment question formed a separate category & are not included here. That answer also does not establish an equal split between cash & online payments.

There is another important boundary before those questions. The original questionnaire used responses about an account, debit card or mobile-money service to decide who received them. Our group therefore comes from that screened population. We cannot use these results to describe people without accounts, or to estimate the share of all cash-on-delivery customers who use digital payments.

For the main calculation, we kept the usual-cash respondents who answered yes or no when asked whether they had made an in-store payment by phone or card in the previous 12 months. This creates a comparison between 2 answers from the same person, rather than combining unrelated national statistics.

Many usual-cash respondents also used phone or card payments in stores

The main results show the share reporting at least 1 in-store phone or card payment during the previous 12 months. They are weighted using the final survey weight supplied for each country.[1][2][3][4][7]

Cash on delivery & in-store phone or card use

In-store phone or card use among screened respondents who reported a delivered online purchase, usually paid cash on arrival & answered yes or no to the in-store payment question.

Data table: Cash on delivery & in-store phone or card use
CountryWith survey weightsWithout survey weightsRespondents answering yes or no
Indonesia32.1%28.2%209
Malaysia55.0%51.6%157
Philippines34.3%21.8%197
Thailand67.6%69.8%139
2024 interviews. Known-answer coverage was complete in Indonesia, Malaysia & Thailand, & more than 99% by survey weight in the Philippines. Weighted & unweighted estimates use the same respondents. The count is the number giving a yes-or-no answer, not a country population or a transaction count.

The method section shows how unknown answers affect the estimate. Those missing-answer bounds are separate from sampling uncertainty.

The sample sizes in the first table count people answering the in-store question, not a country’s cash-on-delivery population. Someone who bought once & someone who bought repeatedly each provide 1 response. The survey weight determines their contribution to the weighted calculation; it does not turn their answers into a count of purchases.

A person can appear in both groups without any contradiction: usually paying cash for delivered online purchases & having paid by phone or card in a store. For a seller, the useful finding is the overlap itself. The survey did not ask that person to choose between 2 permanent identities, “cash customer” & “digital customer.”

These figures do not say how many online orders were paid in cash, how much money changed hands, or whether a particular buyer would switch payment methods. We also do not use them as a country league table. The estimates describe separate survey groups with different samples & fieldwork periods.

Why show the weighted & unweighted results together?

A survey respondent does not necessarily represent the same number of people as another respondent. The World Bank’s final weights account for unequal selection probabilities & adjust the sample using population information. We used those supplied weights without trimming or pooling them across countries.[6]

The unweighted calculation asks a simpler question: what fraction of the responding people in the released sample answered yes? Showing it alongside the weighted estimate makes the role of weighting visible.

Both percentage columns in the main results table use the same respondents & the same yes-or-no answers. Only the contribution of each respondent changes.

The Philippines illustrates why this check belongs beside the headline. Its weighted estimate is 12.5 percentage points above its unweighted estimate, using the unrounded calculations. The overlap exists under both calculations, but its estimated size is sensitive to how respondents are weighted.

We did not analyze which demographic characteristics explain that weighting difference, & it does not describe a change in behavior over time. When quoting the Philippines result, a seller or researcher should retain the weighted label & the accompanying sensitivity check, rather than treating 34.3% as an exact share of today’s customers.

Using a payment method once is different from using it regularly

The in-store question covers any use during the previous 12 months. A follow-up asked those users how often they paid by phone or card in a typical month. This helps distinguish occasional use from a more frequent routine, within the smaller group that reported in-store use.[5]

How often did in-store users pay by phone or card?

Typical-month frequency among usual-cash respondents who reported in-store phone or card use & gave a known frequency answer. This denominator is smaller than the main comparison.

Data table: In-store phone or card payment frequency
CountryWeeklyMonthlyLess than monthly
Indonesia21.9%28.1%50.1%
Malaysia51.5%30.8%17.8%
PhilippinesNot displayedNot displayedNot displayed
Thailand55.2%25.0%19.8%
Weighted percentages from 2024 interviews; rounding can make totals differ from 100%. The source codes a “never” frequency answer within “less than monthly.” The Philippines frequency breakdown is not displayed because the subgroup did not meet the predefined respondent-count & weight-only effective-size criteria.

That last detail affects the reading of the first row. Indonesia’s 50.1% in the lowest category should not be described as a group with a regular monthly digital-payment habit. The category is broad, & the typical-month question differs from the earlier question about any use over a year.

For these supported breakdowns, frequency-answer coverage was complete in Indonesia & Thailand & more than 99% by survey weight in Malaysia. Malaysia’s outward-rounded missing-answer bounds are 51–52% weekly, 30–32% monthly & 17–19% less than monthly. Each category has its own bound; their upper endpoints do not form a distribution. With complete answers, Indonesia’s & Thailand’s bounds coincide with their displayed estimates.

The Philippines remains in the main analysis. Only its frequency breakdown is withheld from comparison because that smaller subgroup failed our prespecified respondent-count & weight-only effective-size criteria. A blank frequency result means the evidence was too thin for that breakdown, not that nobody used those payment methods.

What should a seller do with this finding?

Begin with the information a payment choice actually supplies. It may tell the team how to collect payment for an order. It does not, on its own, explain the customer’s reason, confidence, financial access or experience elsewhere.

For a creator launching physical merchandise or a business selling through social media, this distinction matters when writing order instructions. Explain which methods the business really accepts, when payment is due, & what happens at delivery. Avoid adding a story about the customer’s skills or motives unless the customer has supplied it.

The same distinction can improve customer research. Instead of asking “Why don’t you use digital payments?”, try “What made cash on delivery the right choice for this purchase?” The second question leaves room for a person who already pays digitally in another setting. Ask a separate question if you need to understand that other setting.

Possible explanations can become questions for a seller’s own customers. They are not findings from this study. These survey fields do not identify trust in a particular store, a delivery concern, available funds or a reason for choosing cash. Reporting one of those explanations as the cause would add information the data do not contain.

The analysis also does not choose a checkout policy. Keeping, removing or changing a payment option involves the business’s own costs, operations & customer evidence. Our result supports a better starting question, not a promise that a change will increase sales.

Observe the choice, ask about the purchase, clarify the instructions

A parcel, cash & clipboard connect to a shop with speech bubbles, then to a sheet showing payment symbols beside a pencil.
  1. Record the choice

    Describe the payment selected for this order without assigning a wider customer label.

  2. Ask about this purchase

    Use an optional question to understand why the customer chose that method for this order.

  3. Clarify the instructions

    Explain the methods actually offered, when payment is due & what happens at delivery.

A practical seller workflow, not survey data. Separate what an order shows from what the customer explains.

A fictional seller puts the distinction into practice

Imagine a creator selling printed notebooks who offers payment through a checkout or cash on delivery. In this fictional example, an order record shows that a customer selected cash.

The seller’s first draft of a customer note says: “This buyer does not use online payments.” The order record does not support that statement. A more accurate note is: “Cash on delivery selected for this notebook order; reason not yet known.”

The seller then asks an optional question: “What made this payment option work best for this order?” In the fictional reply, the customer says that somebody else will receive the parcel & handle payment. The seller records that explanation as the customer’s account of this purchase. It is not turned into a rule about other buyers.

A separate, optional conversation establishes that the same fictional customer pays by card in shops. That answer can sit beside the delivery choice. It does not cancel it, prove willingness to prepay for the next order, or establish which checkout explanation will persuade the customer.

The seller confirms the delivery arrangements & corrects the unsupported customer label.

Copy this payment-context worksheet

Use these fields in an existing customer-research or order-review document. No sign-up is needed.

1. Purchase context: [product] · [sales channel] · [delivery arrangement]. Distinguish this order from a customer’s usual behavior.

2. Payment observation: [method selected, or customer’s description of their usual method] · [source of this information].

3. Business information: [methods genuinely offered] · [when payment is due] · [delivery instructions given to the buyer].

4. Customer explanation: [answer to an optional question about this purchase]. If there is no answer, write “reason not known.”

5. Other payment contexts: [relevant information the customer chooses to share]. Do not infer it from this order.

6. Next action: [specific clarification or operational follow-up] · [owner] · [review date]. Link the action to the information above.

Before using the notes to brief a campaign or change an order page, ask whether each sentence describes an observation, a customer explanation or an untested idea. If several customers raise the same issue, that is a lead to investigate with a defined sample. It does not turn these 4 national survey estimates into a forecast for your store.

How we calculated & checked the results

We used the 4 versioned country releases of the World Bank’s Global Findex 2025. Gallup conducted the surveys in 2024 among civilian, noninstitutionalized adults aged 15 & older. The country samples used face-to-face interviews, with the coverage qualifications described in the source methodology.[1][2][3][4][6]

We fixed the questions, country set, calculations & support criteria before examining the joint outcomes. The primary calculation divides the total survey weight of in-store phone or card users by the weight of respondents giving a yes-or-no answer, within the usual-cash group. Each country has its own calculation.

Unknown answers are not recoded as “no.” We assess how much of the group answered the question & calculate bounds that assign every unknown answer first to “no” & then to “yes.” These bounds concern missing answers only; they are not confidence intervals. Small response groups & details that could reconstruct them are withheld from the public tables.

How much could unknown answers change the result?

Missing-answer bounds for in-store phone or card use among usual-cash respondents. These are not confidence intervals.

Data table: Missing-answer bounds for in-store phone or card use
CountryIf unknown answers were all “no”If unknown answers were all “yes”
Indonesia32.1%32.1%
Malaysia55.0%55.0%
Philippines34.0%35.0%
Thailand67.6%67.6%
The Philippines lower bound is rounded down & its upper bound up to whole percentage points to protect small response groups. With complete answers, both endpoints show the same 1-decimal estimate. These bounds do not address sampling error or missing answers to the earlier questions that defined the group.

The sequence also explains why the figures use different groups:

1. Original responses about an account, debit card or mobile-money service determine which questions are asked.

2. A delivered online purchase in the prior 12 months & usual payment in cash define the group of interest. The separate “both” answer is excluded.

3. Known yes-or-no answers about in-store use form the main percentage’s denominator.

4. Known frequency answers among in-store users form the smaller frequency denominator.

All 4 main country groups passed the predefined checks for sample support, response coverage & concentration of survey weights. The frequency breakdown passed for Indonesia, Malaysia & Thailand. Those checks are safeguards for presenting a descriptive result; they do not establish statistical precision.

The released files do not supply the full sampling-design information needed for our conditional estimates’ design-based confidence intervals. We therefore do not attach significance tests or claim country rankings. The surveys also used different fieldwork windows & asked respondents to recall the preceding year.[6]

A second calculation independently reproduced the aggregate results. The downloadable method & code explain the definitions & calculations; individual survey records are not redistributed. These are historical survey findings, not a live measure of payment behavior in 2026.

No sign-up required. Tables contain aggregate historical results with small-cell protections. The reproduction package contains translated method explanations & the original code, not individual survey records.

Sources & notes

  1. Development Research Group, Finance and Private Sector Development Unit (World Bank). Indonesia: The Global Findex Database 2025: Connectivity and Financial Inclusion in the Digital Economy, 2024 interviews, reference IDN_2024_FINDEX_v02_M. Downloaded October 19, 2026. Percentages in this article are Influence IQ’s analysis of the released responses, not figures quoted from the World Bank report.
  2. Development Research Group, Finance and Private Sector Development Unit (World Bank). Malaysia: The Global Findex Database 2025: Connectivity and Financial Inclusion in the Digital Economy, 2024 interviews, reference MYS_2024_FINDEX_v02_M. Downloaded October 19, 2026.
  3. Development Research Group, Finance and Private Sector Development Unit (World Bank). Philippines: The Global Findex Database 2025: Connectivity and Financial Inclusion in the Digital Economy, 2024 interviews, reference PHL_2024_FINDEX_v02_M. Downloaded October 19, 2026.
  4. Development Research Group, Finance and Private Sector Development Unit (World Bank). Thailand: The Global Findex Database 2025: Connectivity and Financial Inclusion in the Digital Economy, 2024 interviews, reference THA_2024_FINDEX_v02_M. Downloaded October 19, 2026.
  5. World Bank, Global Findex 2025 questionnaire, pages 11–12; codebook; & country question mapping. Questions FIN26B & FIN27 define the online-purchase/usual-cash group. FIN25E2 measures in-store phone/card use, & FIN25E3 measures typical-month frequency. The original account/debit-card/mobile-money screen is not equivalent to filtering on the release’s constructed account indicator.
  6. World Bank, Global Findex 2025 survey methodology. Fieldwork in 2024: Indonesia, August 21–September 30; Malaysia, August 2–September 23; Philippines, August 29–November 6; Thailand, September 7–December 13. Thailand excludes Narathiwat, Pattani & Yala & some other districts for security, covering less than 4% of the population. Full-country margins of error in the methodology are not margins of error for these conditional groups.
  7. Influence IQ analysis, October 2026. The method was fixed before the joint outcomes were examined; independent calculations reproduced the aggregate results. The package contains the method, disclosure-protected aggregate outputs & reproduction code, not individual survey records. The source collectors, distributors & funders are not responsible for this analysis or its interpretations.

The article & table labels are available in every article language. The technical reproduction package includes translated explanations; the source questionnaire remains in English. The 2024 survey period & country scope are unchanged across languages.

Related research