From 45 Days to 8 Minutes: Anatomy of a Process Transformation
The number that started it
In 2016 I was Transformation Manager at Transbank, the company that processed essentially every card payment in Chile. A small business that wanted to accept cards filled out a paper application, and then waited. The system that generated the commitment date promised activation in 46 calendar days. Our measurements said the promise was honest: half of merchants were live in 17 business days, 85% within 27, and the unlucky tail stretched past 38.
Two years later a merchant could complete the same onboarding online in 8 minutes: application, credit evaluation, contract signature, and installation scheduling, in one sitting. This is the story of how, and of the one diagnostic finding that made the whole redesign obvious.
Where 45 days actually goes
The first thing we did was a value-stream map of the existing process, with a stopwatch. The finding that reframed everything: across those 45 days, the actual work performed on a typical application added up to about 29 minutes. Downloading the request, pulling the credit and compliance reports, evaluating, creating the merchant in the core systems, scheduling. Half an hour of touch time, six weeks of elapsed time.
The other 44-plus days were structural waiting:
- Nine teams across four companies. Call center, sales, sales operations, risk, legal, pricing, activation, data entry, and field installation, spread across Transbank and three outsourced providers. Every handoff was a queue.
- Paper. The signed contract traveled by motorcycle courier. A folder of client documents physically moved between areas, and 14 to 20 of the waiting days were just that: the file sitting in someone's inbox.
- Batch processing. Applications were worked in lots, so an application that arrived right after a batch closed waited for the next one.
- A self-fulfilling promise. Because the system committed to 46 days, 46 days was, by definition, acceptable performance. Nobody was failing.
That last point matters more than it looks. When the official promise is six weeks, every team's internal SLA comfortably fits inside it, and the process is "healthy" while the customer waits a month and a half. The first casualty of the redesign was the promise itself.
The redesign rule: no human on the happy path
The new process was built around one rule: a standard merchant should never wait for a person. Humans handle exceptions; the happy path is automated end to end.
In the redesigned flow, the merchant sits down at a web portal and:
- Enters their tax ID, and the form pre-fills company data and legal representatives from the credit bureau's online service, instead of asking the merchant to retype what the bureau already knows.
- Gets evaluated in real time. Business rules built with the risk team return approve, reject, or "gray zone" while the merchant is still on the page. Only the gray zone, a small minority of cases, routes to a human analyst.
- Signs digitally. A certified e-signature replaced the printed contract and the motorcycle courier. This was as much a legal negotiation as a technical one, and it removed the single longest wait in the old process.
- Schedules their own installation against the real availability of field technicians and terminal stock, instead of waiting for a scheduling desk to call back.
- Is activated automatically. The portal triggers merchant creation and activation in the core systems, and dispatches the installation order to the field provider, with no re-typing by a data-entry team.
Behind that simple sequence sat the real work: integrating the portal with the CRM, the credit bureau, the certified-signature provider, the document management system, the internal fraud blacklist, the scheduling system, and monitoring. Each integration was a "lever" we sequenced deliberately, because each one removed a specific number of minutes of manual work. Our capacity model tracked it: integrate the portal and per-case handling drops from 29 to 21 minutes, automate the core-systems creation and it drops to 16, add digital signature and it drops to 13, add self-scheduling and the residual human work on a standard case approaches four minutes, none of it on the critical path the merchant experiences.
Designing the exception path is the actual job
The happy path was the easy 80%. The design effort went into what happens when the flow breaks: the gray-zone evaluation that needs a human decision, the legal representative who must sign in person, the failed installation visit, the merchant who abandons mid-application. Each exception got its own modeled subprocess with explicit re-entry points into the main flow.
That is the part I would underline for anyone attempting this: an automated process is only as good as its worst exception. If the exception path is an email to a shared inbox, you have not removed the queue, you have hidden it.
The ramp, honestly
We did not flip a switch. The online channel launched as a pilot handling ~310 onboardings a month, around 10% of volume, in late 2016. We ramped it as the integrations matured: 37% by early 2017, 87% by mid-2017, and 100% by January 2018, by then around 3,200 onboardings per month against a yearly volume of roughly 32,000. Running the old and new processes in parallel for 14 months cost money and patience, and it was the right call: every increment of rollout surfaced exceptions we had not modeled, while the legacy process caught what the new one could not yet handle.
The business result: onboarding cost per merchant fell by roughly three quarters, worth about $2M per year, and the lead time for a standard merchant went from 45 days to 8 minutes.
What transfers
I have since built AI systems at a hyperscaler, and the lessons from this pre-AI transformation transfer almost embarrassingly well:
- Measure waiting, not work. Touch time was 29 minutes; elapsed time was 45 days. Optimizing the 29 minutes (faster analysts! better macros!) could never matter. Every meaningful gain came from deleting queues.
- Kill the comfortable promise. A generous SLA makes everyone compliant and the customer miserable. The redesign target came from physics (what must actually happen?), not from benchmarking the status quo.
- Sequence integrations as levers with known value. We knew the minutes each integration would remove before we built it, which made prioritization boring, in the best way.
- The exception path is the product. Then and now, with workflows or with AI agents: the standard case automates easily, and the design quality shows in what happens when it does not.