Experimental Design
Experiment 1 was a customer-account-level natural field experiment conducted among 427,199 early-arrears accounts at a large U.S. gas utility. Eligible accounts had recently fallen into arrears and had entered the utility’s collections process, but had not yet reached later-stage warning or disconnection steps. Accounts were allocated to one of five groups: business-as-usual control, basic email reminder, active-choice email reminder, autopay email reminder, or basic SMS reminder. The interventions were delivered on November 17, 2020. The experiment was designed to test whether one additional reminder increased repayment, whether reminder framing affected repayment, and whether SMS outperformed email. Outcomes were measured from administrative billing and customer-service records and included cumulative payments over the following four weeks, indicators for any payment and full repayment of the overdue balance, and secondary operational outcomes including customer-service contact and payment extensions.
Experiment 2 was a customer-account-level natural field experiment in the same broad early-arrears setting, with two nested randomized components. First, among 151,022 overdue accounts with balances above $25, the utility randomized whether the next regular bill included an additional reminder sentence at the top of the bill; this created a two-arm comparison between a standard bill and an amended bill. Second, among 95,051 accounts with both an email address and a mobile phone number on file, the utility randomized assignment to one of four reminder conditions delivered four days after bill issuance: no additional reminder, email reminder, SMS reminder, or automated outbound-dial reminder. The experiment was designed to test whether a very low-cost bill-language intervention accelerated repayment and whether reminder channel affected repayment in this setting. Administrative outcomes included cumulative payments over the subsequent four weeks, indicators for any payment and full payment, and secondary outcomes including customer-service calls, autopay registration, and payment extensions.
Experiment 3 was a customer-account-level natural field experiment conducted among 120,209 early-arrears accounts after the utility had already incorporated lessons from the earlier experiments into its standing reminder workflow. Accounts were allocated across five groups: the status-quo reminder sequence and four alternative reminder-content conditions. The four treatment conditions were: a more direct day-4 text, a day-4 text that added the exact amount owed, a day-4 text that added an autopay prompt, and a more direct day-8 email that left the day-4 text unchanged. The experiment was designed to test whether message simplification and more direct wording improved repayment within an already-active reminder sequence, and whether the placement and specific content of those changes mattered. Outcomes were measured from administrative records and included payments one, four, and ten weeks after the day-4 reminder, any payment within ten weeks, and secondary outcomes including later gas usage, autopay enrollment, payment extensions, and receipt of the day-8 reminder.
Experiment 4 was a customer-account-level natural field experiment conducted among 41,650 late-stage arrears households at a large U.S. gas utility. Eligible households had overdue balances of at least $400 and had reached the utility’s collect-or-close stage between May and July 2024. All households remained on the standard collect-or-close workflow, which already included a mailed disconnection-warning notice. Households were allocated to one of four groups: control, a text message requesting a minimum payment of $200, a text message requesting a minimum payment of $400, or a text message requesting payment of the full amount owed. The experiment was designed to test whether varying the requested minimum payment affected repayment behavior in a late-stage collections setting. Administrative outcomes included payments and any payment, as well as secondary outcomes including payment extensions and enrollment in California’s Arrearage Management Plan.