Decisions that cost revenue first and earned trust later

Putting e-invoice submission into every Malaysian plan, the free one included

In early 2024 we had to decide how Rimbunara would support e-invoicing in Malaysia. The Inland Revenue Board was phasing it in by annual turnover, beginning with the largest businesses in August 2024. Many of our 121,000 paying Malaysian customers fell into later phases; others sat below the exemption threshold but sold to larger companies already in scope, which were asking when their suppliers could issue validated e-invoices. Each invoice would have to be submitted to the MyInvois system and validated before it counted.

Our chief revenue officer and chief financial officer proposed a compliance add-on at RM 25 a month, forecast at US$6.2 million of annual revenue once adoption matured. Their case was reasonable: we would carry real integration and support costs, and competitors were charging. I argued that customers would read a charge for obeying the law as a charge for staying in business, and that the businesses most likely to leave were the small ones we most wanted to keep. I proposed submission in every plan, including the free plan up to twenty invoices a month. Our chief executive backed me on condition that we tracked the revenue declined against the churn avoided.

The price was not only the add-on. We moved 31 engineers and product staff onto the compliance layer for three quarters, which delayed our bookkeeping assistant by two quarters, and we absorbed about US$1.1 million a year of extra support cost. When the timetable was revised, we had already emailed 90,000 customers with dates that no longer applied, and the support queue doubled for a fortnight.

By June 2026 Malaysian customers were submitting about 2.6 million e-invoices a month through Rimbunara. Monthly churn among paying Malaysian businesses fell from 2.3 per cent to 1.6 per cent, the share of free users moving to a paid plan within a year rose from 6.1 per cent to 8.4 per cent, and paying businesses grew to 168,000. Compliance is now a standing product line with its own budget, not a project.

I would do two things differently. I would never send a reminder tied to a date without saying that the authority could change it, and I would have formed the country compliance team a year before the first deadline rather than six months.

Rebuilding the mobile app for a weak signal

In mid-2022 our research team spent three weeks with customers in East Java, Sulawesi, the Visayas and Mindanao. What they brought back was uncomfortable. Our mobile app was a web application inside a native wrapper. It needed a connection to save an invoice, weighed 61 MB and slowed badly on phones with 2 GB or 3 GB of memory. Outside the main metropolitan areas of Indonesia and the Philippines, 38 per cent of invoice-creation sessions were abandoned before the invoice was saved, and those customers were leaving at almost twice the rate of customers in Jakarta and Metro Manila.

I proposed rebuilding invoicing, receipt capture and stock counts as an offline-first Android application, with iOS to follow. Our chief technology officer agreed with the diagnosis. Our chief revenue officer and two board members disagreed with the timing: competitors were announcing automated bookkeeping, and our plan for that year promised new payment features on mobile. The rebuild meant nine months without a significant new mobile feature and deferred about US$2.1 million of planned upsell revenue. The board accepted it after I committed to reporting the abandonment figure monthly instead of a launch date.

The hardest problem was invoice numbering. Tax authorities in our markets expect invoices to be numbered in sequence, and three phones issuing invoices offline cannot see one another's numbers. In our first Philippine pilot, two devices gave the same number to different customers. We stopped the pilot, reserved blocks of numbers for each device and made the rules for unused numbers visible to accountants before restarting.

The new app shipped in Indonesia in March 2023 and in the Philippines in June. The Android download fell to 17 MB, abandoned sessions outside the metropolitan areas fell to 9 per cent and crash-free sessions rose from 97.1 per cent to 99.6 per cent. Weekly active businesses in those regions grew 44 per cent over the following year. The iOS version arrived five months later than planned.

I would have asked our tax advisers to review the numbering design before any synchronisation code was written, and I would have launched in one country, not two.

Pricing by the business, not the seat, in Indonesia and the Philippines

Until 2023 Rimbunara priced every market the same way: per user, per month. In Indonesia our cheapest paid plan cost Rp 149,000 a month for one user, a real sum for a warung or market stall earning a few million rupiah. Our research found that 47 per cent of Indonesian businesses that tried a paid plan and left gave price as the reason, and that many owners shared one login with relatives and staff to avoid paying for a second seat, which meant nobody could tell who had issued an invoice.

I proposed per-business pricing in bands based on invoices issued, free staff logins and a Mikro plan at Rp 49,000 a month in Indonesia and ₱199 in the Philippines, payable weekly or monthly from an e-wallet or in cash at a convenience-store counter. Our chief financial officer opposed the size of the cut; he modelled a fall of about 30 per cent in average revenue per business and asked for a smaller step. Our Indonesia country manager feared that existing customers would move down. Both were right to worry. We agreed a one-quarter test in East Java before a wider launch, and a rule that existing customers could change plan only at renewal.

The cost arrived as forecast. In the first six months about 9 per cent of existing Indonesian paying customers moved to Mikro, taking about US$420,000 of annualised revenue with them, and average revenue per paying business in Indonesia fell by about a quarter. Weekly prepayment also produced more failed renewals than we had planned for, because owners forgot to top up.

Paying businesses in Indonesia grew from 58,000 in mid-2023 to 131,000 in June 2026, and in the Philippines from 39,000 to 84,000. Indonesian subscription revenue rose from US$6.4 million in 2022 to US$10.1 million in 2025. Because every member of staff now has a login, owners can see who issued each invoice, and our research can see how work is shared in a small business.

I would test weekly prepayment separately from the price change. Bundling them made it impossible to say which caused the extra failed renewals, and we spent two quarters untangling the effect.

Pausing instant invoice financing after a fake-invoice ring

In 2024 we launched a feature that let customers receive an advance from one of three partner lenders against unpaid invoices issued through Rimbunara. We earned a referral fee; the lenders made the credit decisions and carried the losses. By early 2025 about 14,000 businesses used it, and in the Philippines it had become one of the reasons contractors chose us.

In April 2025 a lending partner noticed that a cluster of Philippine accounts had drawn advances against invoices to buyers who did not exist. Our investigation found 43 linked accounts, opened over five months, which had drawn about ₱38 million, roughly US$660,000, before they disappeared. Our fraud model had been trained mostly on Malaysian data and did not recognise the pattern. I had approved the launch with that model, and I said so to the Audit and Risk Committee.

I recommended pausing instant offers in the Philippines and Indonesia for six weeks while we added three controls: a confirmation link the buyer must open before an invoice can be financed, a minimum account age of ninety days, and matching of the payout account's name against the registered business. Teodora Malabanan, our Philippines country general manager, argued for keeping the feature open to established customers, because contractors relied on it to pay their crews. Two lenders asked us to share losses so that volumes could continue. I declined the second request and accepted a version of the first: an exception for 600 long-standing customers, each reviewed by hand.

Financing referral revenue fell by about US$1.3 million in 2025, approved advances fell 27 per cent for the year, and some contractors moved to a competitor that had kept instant offers. One lender reduced its limit with us.

Fraud losses at our partners fell from 0.9 per cent of advanced value to 0.08 per cent. By the first quarter of 2026 volumes had recovered to 92 per cent of their earlier level, and buyer confirmation had an effect we had not expected: confirmed invoices were paid on time 74 per cent of the time, against 61 per cent before.

I would put every new feature that moves money through a trust-and-safety review in each market before launch, using that market's own data, not after the first loss.