Learning from leading companies

How the companies that do this best rolled out A.I.

Six lessons from companies that went first, documented in public sources. Not to copy, but to scale down for a business of a few dozen people in Vietnam. We describe methods, not figures, because their numbers are not yours.

01

Start from the outcome, written down before building

Amazon · "Working backwards"

What they did

Before building any product, Amazon teams write a mock press release and FAQ describing the finished thing as if it already existed. Important meetings begin with the room silently reading a six-page narrative memo, not slides.

Why it worked

Writing in prose forces the proposer to think it through. Fully describing the outcome before building kills most vague projects at the start, and gives everyone a standard to compare against when done.

How sonnhat applies it for Vietnamese businesses

Every sonnhat automation project starts with a one-page description: when the workflow finishes, who receives what, when, what it looks like, and how it is measured. You approve that page before we build. If that page cannot be written, the workflow should not be built yet.

02

Using A.I is the default; asking for people is the exception

Shopify · CEO Tobi Lütke's memo, 2025

What they did

In early 2025 Shopify's CEO sent a company-wide memo, later made public: skilled A.I use is a baseline expectation for every employee; before asking for more headcount or budget, a team must show why A.I cannot do the job; A.I usage is part of performance reviews.

Why it worked

It flips the question. Instead of "what can A.I do for us", each person must answer "why can a machine not do this yet". When the question changes, behaviour changes without any big programme.

How sonnhat applies it for Vietnamese businesses

In the leadership workshop, sonnhat helps you write a similar rule at your company's scale: for example, every request to hire for office work must include one line, "has this task been tried on a machine, and what happened". That one line is enough to change how the whole company thinks about work.

03

The machine stops on abnormality; the person decides

Toyota · Jidoka, "automation with a human touch"

What they did

In the Toyota Production System, machines are designed to detect a defect and stop immediately, halting the whole line if needed. Workers have the right and the duty to pull the stop cord. Fixing the root cause matters more than keeping the line moving.

Why it worked

Automation without a stop point multiplies errors at machine speed. A pre-defined stop condition plus a human decision turns every error into a process fix instead of an accident.

How sonnhat applies it for Vietnamese businesses

Every workflow sonnhat builds has a "stop cord": abnormal conditions defined up front (amount above a threshold, a customer on a special list, A.I not confident) stop the flow and notify a person. Any step touching money, contracts or personal data is machine-proposes, human-approves. No exceptions, even when you ask to remove the approval step for speed.

04

Automate the task, not the relationship

Klarna · A.I customer-service assistant, 2024–2025

What they did

In 2024 Klarna announced an A.I assistant handling most customer-service chats and sharply reducing staffing needs. More than a year later, the CEO publicly admitted service quality had fallen because cost had been over-prioritised, and the company hired people back for customer care, keeping A.I in a supporting role.

Why it worked

Repetitive work (lookup, sorting, drafting) is done better by machines. But customers notice when no real person is on the other end, and that loss does not appear in the cost sheet until it is too late.

How sonnhat applies it for Vietnamese businesses

sonnhat separates two layers in every customer-care workflow: what the machine does (read, sort, draft, log to CRM, remind) and what people keep (the final reply to an unhappy customer, high-value customers, every promise). In the first phase we decline to build flows where A.I talks directly to customers without human approval.

05

Small, continuous improvements proposed by the people doing the work

Toyota · Kaizen

What they did

Toyota does not improve through large top-down projects. Every worker is encouraged to propose small improvements to their own step, implement quickly, measure, and keep what works. Thousands of small improvements add up to an advantage nobody can copy.

Why it worked

The people doing the work daily know which task wastes time absurdly. Big top-down projects usually automate the wrong task. Small bottom-up improvements automate the right one, and the people who use them keep them alive.

How sonnhat applies it for Vietnamese businesses

This is why sonnhat does training as well. The class does not teach A.I; it makes each person automate one task of their own. Afterwards the team has the habit of asking "what did I stop doing by hand this week", and sonnhat only builds what exceeds the team's reach, never what the team can do itself.

06

Small pilot, measure, then scale, with a champion in each department

Copilot rollouts at large Microsoft customers

What they did

When bringing A.I assistants to tens of thousands of employees, large companies do not switch it on for everyone at once. They choose a pilot group, build a network of "champions" in each department, measure usage and results weekly, then expand in waves with training. Microsoft packaged this as an adoption framework for customers.

Why it worked

Tools do not spread by themselves. Colleagues showing each other spread them. A small pilot makes mistakes cheap; weekly measurement tells you whether to expand or stop; a champion in the department makes using A.I the normal thing in that department.

How sonnhat applies it for Vietnamese businesses

For a business of a few dozen people, sonnhat scales this down to three things: choose one department and one task to go first; choose one person in that department as operator and champion; run a two-week trial with a scorecard. Pass, and move to the next department; fail, fix or stop. Never switch on for the whole company at once.

The common thread

Six lessons, one line.

Describe the outcome before building. Treat the machine as the default for repetitive work. Keep the stop cord and the decision with people. Do not automate relationships. Let the people doing the work propose. Pilot small, measure, then scale.

None of these needs a large budget or a technical team. They need something harder: the discipline to pick one task, write it down, measure it, and not drop the human approval step once the machine seems to run fine. That is the part sonnhat does with you.

Leading companies do not have special tools. They have the discipline to describe the work before handing it to a machine.

sonnhat's principle

Start with one task

Tell sonnhat about one task your team still does by hand.

The first call is 30 minutes, online, free. By the end you know how far that task can be automated, with what, and how long it takes. If it is not a fit, we stop there. No commitment.