Start with the complaint: repeated work takes too long, people lack the information they need, or customers want something the current service cannot provide. These lead to different designs and different measures of success.
What does automation change?
It takes work that repeats, follows a describable rule, and hands it to a system. Filing, sorting, entering the same figure into a second system, producing the same report every month.
Useful measures include time per case, errors, waiting time and service consistency. Record the baseline and include review work in the comparison. Automation can improve control as well as reduce effort.
What does automation change besides the hours?
Time savings attract attention, but standardisation can be just as useful. Clear steps and a visible history make it easier to understand what happened and where intervention was needed.
For an illustrative pilot handling a hundred cases, count how many complete correctly, need review or fail. That tells you more than saying the process is automated. Existing systems may already provide part of this baseline.
What does augmentation change?
It leaves the person in charge and gives them a faster way to do the thinking. Drafting, summarising, pulling together what is known about a client before a meeting, checking a long document for the three clauses that matter.
A person reviews and authorises the outcome. Make that responsibility explicit, including who checks factual accuracy and which material must not be entered into the tool. The benefit can be better work, faster preparation or both.
When is augmentation a useful first step?
A bounded drafting or research aid can be easier to test than an unattended process. It still needs safeguards: a person can accept a wrong answer or disclose information inappropriately. “Human in the loop” is a responsibility, not a guarantee.
There is a second reason. Augmentation puts the technology in front of the people who will live with it, early, on work they understand. The objections heard in that first month are the most useful specification available.
What does differentiation mean here?
An offer that was not possible before, rather than a better version of an existing process. A service that adapts to each client, a response time nobody in the market can match, a product built out of data a business already held and could not use.
A new service needs reliable foundations relevant to that service. Personalisation may depend on consistent customer data; faster fulfilment may depend on dependable integrations. It does not require completing every possible automation first.
Is there a fixed order?
No. A training programme may precede integration work, and a tightly scoped new service may be tested before a company-wide clean-up. Follow the actual dependencies rather than a universal maturity ladder.
The useful sequence is local: establish the required inputs, test the risky assumptions and expand only when the result is dependable enough. A small successful pilot is evidence for the next decision, not proof that every process should follow it.
Which approach should you start with?
Compare business value, frequency, data readiness, review effort and consequences of failure. A costly task is not automatically a good automation candidate, and an inexpensive one can still be a useful low-risk pilot.
Choose one outcome that can be tested. Decide in advance what improvement would justify continuing and what result would make you change direction. That is a more useful starting point than choosing a stage.
