ASAAS Cut Invoice Response Time 85% with Birdie

About ASAAS

ASAAS is the financial and operational technology layer for small and medium businesses in Brazil. A Payment Institution and Direct Credit Society authorized by the Central Bank of Brazil, headquartered in Joinville, it centralizes billing, Pix instant payments, receivables advances, invoice issuance, a digital account, and ERP functionality in one platform. More than 3.5 million accounts have been created on it, over R$148.9 billion has moved through it, and more than 487.5 million charges have been paid.

The structural fact that shapes everything about its customer experience: an ASAAS customer is not a casual user. When something breaks on the platform, their business stops. The invoice does not go out, the supplier does not get paid, revenue does not arrive. ASAAS holds the RA1000 seal on Reclame Aqui, Brazil's largest public consumer complaint platform, and runs a multichannel CX operation across chat, WhatsApp, phone, email, bot, and a formal ombudsman channel.

The Situation and Challenges

ASAAS had a listening problem of scale, not of intent. Between January and May 2026, 236,825 interactions moved through its relationship funnel. In the support journey alone, 16,091 unique customer signals arrived across more than six channels: support tickets, NPS free text, the ombudsman channel, Reclame Aqui, Consumidor.gov, and social media.

No manual structure reads that volume. Without technology, a company of that size decides by sampling, by anecdote, or by whichever customer shouts loudest. Operational metrics made the problem worse by looking like an answer: response time and CSAT tell you how much it hurts, never where or why.

That gap hid the signals that actually cost money. A bot that "resolved" tickets customers had abandoned. An invoicing failure driving silent churn among business accounts. A canned reply turning NPS neutrals into detractors. Before the improvement cycle began, the support area carried 60.6% NPS detraction. And 70% of all ombudsman cases originated on Reclame Aqui, which meant every untreated pain became public brand discourse, putting both the RA1000 seal and institutional NPS at risk.

The cost of that blindness was concrete. One business customer reported four months unable to resolve invoice issuance, issuing manually the whole time, and openly considering a return to a traditional bank.

The question ASAAS needed to answer was not "how satisfied are our customers." It was: which specific failure, in which journey, for which customer segment, is costing us retention right now?

How Birdie Helped

Birdie became the listening infrastructure underneath every CX and product decision at ASAAS, built as a dedicated workspace with a taxonomy designed for fintech journeys rather than a generic feedback schema.

Full coverage, no sampling

Birdie captures every channel at once: Zendesk support tickets, NPS free text, ombudsman records, Reclame Aqui, Consumidor.gov, and social mentions. Generative AI then classifies each one by area, opportunity, root cause, segment, and sentiment. In a single journey over a single quarter, that meant 16,091 unique signals analyzed at 100% coverage. Every mention carries five sentiment levels plus root cause plus segment (business or individual, plan, company size), so a pattern can be sliced by who it affects, not just how often it appears.

A living taxonomy instead of a static category tree

Categories at ASAAS are named, traceable opportunities: "difficulty reaching a human," "delay in resolution," "invoice support." Each one points back to the tickets that created it. That is the difference between a report and a diagnosis. "Login issues" as a category tells a team nothing; the specific failure points underneath it tell them what to build.

The same signal feeding two altitudes

One customer sentence feeds both the tactical view (an effort by impact matrix that ranks what to fix) and the strategic radar (the funnel from support to ombudsman to Reclame Aqui). Six journeys ran on evolving monthly reports: Support, Card Anticipation, Statement, Bill Pay, Security (login, token, MFA), and Onboarding. Security alone contributed over 6,000 signals across three prioritized improvement opportunities. Onboarding contributed over 5,000, producing three opportunities with two already in development.

Near real-time sentiment as an early warning system

Because classification runs continuously rather than on a survey cycle, the sentiment curve became a live instrument. On the most critical day of the January crisis, it registered nearly 12 negative mentions for every positive one, with an equally high volume of neutral signals: business owners undecided about the future of their own operation. That reading is what told ASAAS the crisis was a trust problem, not just a queue problem.

From Insights to Action

Birdie's reports are co-produced monthly with the ASAAS CX Ops team, and their output is not a document. It is an agenda with an owner and a deadline, carried into forums with product, support, security, onboarding, and the executive board. Each report opens with the points that need to become decisions, includes full customer cases with ticket and protocol numbers, and closes with an effort by impact matrix where every proposed action answers a real customer sentence.

‍

The governing principle: no action enters the matrix without customer evidence behind it, and no recurring pain goes without an owner. Customer voice stopped being opinion and became traceable evidence. Forums stopped debating impressions.

Closed-loop decisions ASAAS shipped

Then the system got stress tested

On January 1, 2026, Brazil's tax reform (LC 214/2025, which unifies five taxes into two with a transition running to 2033) made a new electronic invoicing standard mandatory. Issuance migrated from roughly 2,000 municipal systems to a single National Portal, with new fields, codes, and rules. Within days, access volume brought the national system down. APIs failed. 216 million fiscal documents hit the portal in its debut month.

‍

"What we saw in January was a scenario of absolute chaos, much worse than any initial projection."

Christophe T. Chavey, CEO, Nota Gateway

‍

Inside the ASAAS operation, tickets rose 208% at the first week's peak, 188 a day, 4,011 created in January. CSAT fell 26.2 points, from 87.1% to 60.9%, with resolution rate under 10% at the worst moment. The backlog went 8x, from 200 to 1,622 tickets in days, a deficit of over 1,000 a day. Average first response time hit 15h29, up 237% from December's roughly 4 hours.

Every crisis forces the same question: do you protect the operation or the customer? ASAAS decided it did not have to choose, and, more consequentially, that it could not wait for the National Portal to stabilize on its own. That decision was made on data: the sentiment thermometer and the root cause categorization coming out of Birdie.

The response ran on four layers at once. Operationally, extended hours triggered the day the backlog hit 1,200, a hybrid task force of CX plus invoicing specialists, N2 agents moved to N1 to lift first-interaction resolution, four active backlog sweeps, and channel redistribution that pushed voice up 60%. Tactically, a contingency plan on January 5 that cut SLA targets 80%, bulk root cause categorization on January 8, 777 tickets answered by automation in one day, 309 duplicates merged by AI, support AI in Zendesk, and automated reporting every three hours. In communication, mass outreach before customers called, segmented by error type, a pop-up for invoice issuers inside the platform, a standardized FAQ and internal knowledge base, and posts across every social network including the CEO publicly owning the problem before a fix existed. And in contact, an operation that had been 100% reactive started calling impacted customers, identified by cross-referencing Birdie's segmentation with platform data, prioritizing large accounts on critical SLA.

Results

Crisis response, 20 days from the January 5 contingency plan

• 31 hours to normalize the invoice SLA at N2, which had been in breach for 5 days
• Backlog down 69%, from 1,622 to 501 tickets in under two weeks
• Bot retention at 47% on January 11, up from 14.7% at the collapse peak, a 220% recovery
• Internal escalations down 85%, from 280 to 42, with support AI in Zendesk
• 777 tickets answered automatically in a single day (January 8), 19% of crisis volume, plus 309 duplicate tickets merged by AI

The quarter that consolidated the turnaround, Q1 2026

"Five minutes with the right specialist solved what months of standard support could not. That is respect for the customer in practice: not what a company declares, but what it chooses to do when it costs something."‍‍

‍Klaa Advocacia, ASAAS business customer contacted proactively during the crisis

The program in steady state, month over month

• "Difficulty reaching a human" down 38%, from 4,086 signals in February to 2,542 in April
• "Delay in response and resolution" down 47%, from 2,764 in February to 1,478 in April
• "Anticipation, limit and blocking" down 52%, from 471 in January to 215 in April, following the new March 23 policy
• 97.7% of customer pain resolved before it escalated. Of 236,825 interactions between January and May 2026, only 5,429 reached the ombudsman channel, an escalation rate of 2.29%

Beyond the Numbers

The crisis ended. What it built stayed.

ASAAS came out of January with an invoice crisis playbook written from scratch, with triggers, a response rite, and defined roles, ready for the next regulatory wave. The tax reform runs to 2033, so ASAAS now keeps an annual regulatory calendar and anticipates each milestone instead of reacting to it. The AI duplicate merging deployed as an emergency measure on January 9 became permanent operational structure. The knowledge base written under pressure went into Zendesk Copilot with the most frequent invoicing errors, and the documentation consolidated during the crisis is what took the bot from 32% retention to 86.2%.

Two structural shifts matter more than any of that.

The first is that a 100% reactive support operation learned to make outbound contact, prioritized by data. That capability did not exist at ASAAS before January 2026 and it did not go away after.

The second is governance. The changes outgrew CX: a strategic OKR for the tax crisis was approved as a corporate commitment, which is what it looks like when the customer at the center is a governance structure rather than a slogan. Customer voice at ASAAS is not a report that supports decisions. It is the operating system the decisions run on.

FAQ

How did ASAAS cut invoice response time by 85%? By treating the January 2026 invoicing collapse as a diagnosable problem rather than a queue to drain. Birdie's Customer Intelligence categorized root causes in bulk on January 8 and tracked sentiment in near real time, which told ASAAS which errors to communicate proactively, which tickets to automate, and which customers to call first. First response time went from 15h29 in January to 1 hour by the end of Q1, with the invoice backlog down 93%.

How fast can a CX team act on customer signals? Fast enough to matter, if the listening layer is already running. ASAAS normalized a breached invoice SLA in 31 hours and cut its backlog 69% in under two weeks. That speed came from infrastructure built before the crisis, not improvised during it: full-coverage classification, a named taxonomy, and a monthly ritual that had already trained the organization to turn signals into owned decisions.

How do digital banks and fintechs prevent complaints from reaching regulators? By resolving pain upstream of the escalation path. At ASAAS, 70% of ombudsman cases originated on Reclame Aqui, so the funnel from support to ombudsman to public complaint became a monitored metric. Between January and May 2026, only 5,429 of 236,825 interactions reached the ombudsman, a 2.29% escalation rate, meaning 97.7% of pain was resolved before it became a public or regulatory case.

Want this level of visibility into your own customer signals? Book a demo.

Explore how Birdie works for financial services and see the full customer context platform.