
Many businesses pick a complex process to automate first, assuming a tangled workflow will bring real change. The problem: that process is often unstable, the way it is done shifts week to week, and the person running it relies on personal judgment, so the system is outdated the moment it ships. MADIAD opens every Automation workflow project with a current-state survey, mapping which process repeats in a stable, frequent pattern before any tool gets proposed. This piece walks through how to pick the right first process, and when to automate now versus wait.
Common mistakes when picking the first process to automate
A common mistake is automating a complex process first, assuming complexity means a bigger payoff. In practice complexity usually just means instability: many exceptions, many people involved, handling that shifts with every situation. When a system gets built around this month's way of working, and that way changes next month, the system needs rebuilding. A hypothetical example to illustrate: a 5-step order approval process, from stock check to delivery confirmation, changes its handling three times in three months because the sales team keeps adding a new check whenever an unusual case comes up. Automating that process from day one means every change in handling forces a rewrite of the logic. A second mistake is automating a task that just launched, before anyone has settled into it, on the assumption that acting early avoids bad habits. But a new task usually lacks clear rules yet, so early automation just freezes a way of working that is still being tested. MADIAD addresses this by surveying current operations before proposing an Automation workflow, choosing only a process stable enough to automate without constant rewrites.
Survey current operations before choosing a process
Before picking which process to automate, a current-state survey comes first: record every step of the process as it actually runs, not as it appears on paper. The first step is listing out every manual action, from who receives the request, who enters the data, who checks it, to who sends the final result. Many processes look like 3 steps on paper but run as 7 or 8 in practice, because people add their own cross-checks to avoid mistakes. The next step is counting how often this process repeats in a week, in a month: a task repeating a few times a month is a different design problem than one repeating dozens of times a day, both in priority and in how the system gets built. The last step is identifying who does the work and how many manual steps it takes each time: someone copying data across 3 different tools for every order is a clear sign the process is ready for automation. MADIAD's AI & Automation services start from this exact survey step, not from picking a tool first. Only after the survey is done does it become clear which process to automate first and which to leave for later.
Why automation comes before AI
Automation comes first because it handles the part with clear rules: data entry, syncing data between systems, sending notifications under fixed conditions. These tasks do not need language understanding or reasoning, just correct execution of pre-set logic, running steadily year-round at low operating cost. AI is only needed for what is left: understanding a customer question written in free form, summarizing a long response, or classifying a situation that fixed rules cannot fully cover. When the order gets reversed, meaning AI gets used for something a rule could have written, operating cost rises while the result does nothing a simple piece of code could not already do. A hypothetical example: a workflow that sends an order confirmation email only needs to fill a customer's name and order code into an existing template, fully doable with a conditional automation script, no AI call needed per email. In contrast, a workflow that reads customer feedback to classify urgency level does need AI, because the wording customers use always differs and no fixed rule covers every way of phrasing it. MADIAD designs each setup around this boundary: which part runs on stable automation, which part needs an AI content assistant (Trợ lý AI quản lý nội dung) or an Agentic Workflow, avoiding AI calls for tasks a simple piece of logic already solves.
When to automate the repeated task, when to wait
A process is ready to automate when it meets two conditions: it is stable, and there is data recording how often it repeats. Stable means the way of working no longer changes constantly, the steps are agreed among the people doing them, and it is no longer the case that each person follows a different method. Having data means knowing exactly how many times this process runs in a week or a month, not just a feeling that it repeats a lot. These two conditions travel together: a process that repeats many times a day but still changes depending on who does it is not ready yet, because the system would need fixing the moment the method shifts again. The case for waiting is a process that still changes week to week, for example a marketing content review workflow where the approval criteria are still being debated and revised inside the team. Automating a process like that is like building a house on a foundation that has not set yet. The right move is to let the process run manually a while longer, record the method the team agrees on, then automate the settled part. MADIAD uses the data from the current-state survey itself to decide which process qualifies for automation now and which one should keep running with the team a while longer before coming back to it.
Start with a current-state survey with MADIAD
A current-state survey with MADIAD starts by sitting down with the people who actually run the process, not just the manager. The session covers three parts: listing every real step of the process as it runs today, including steps that are not in the official process documentation; estimating how often it repeats per week and per month based on available data, not gut feeling; and pinpointing where manual steps pile up for the person doing the work, for example having to re-enter data between two different pieces of software. The outcome of the survey is a clear picture of which task should get automated first, which one needs to wait until its method settles, and which one needs AI to handle the language-understanding part instead of plain automation. The survey does not require the business to prepare complex documents beforehand; it just needs the person who does the work to show up and describe how things actually run. From that picture, MADIAD proposes the next step, in the right order: survey first, automation next, AI last when needed.
Conclusion
Picking the right first process to automate matters more than picking the right tool: the order is always survey current operations first, automate the part that is already stable, then bring in AI for the part that needs language understanding. Keeping a process that still changes its method every week running manually a while longer avoids having to redo the automation every time the method changes. One step to try this week: pick a process that repeats daily, write down the exact manual steps the person doing it follows, then count how many times it runs over the next seven days. MADIAD runs current-state surveys for businesses that want to pin down the right process to automate first. Book a current-state survey with MADIAD to get started.
Frequently asked questions
Does an Automation workflow fit a small business without its own tech team?+
Yes. Once the process is stable enough, MADIAD can apply the automation workflow right away. An Automation workflow does not require the business to have its own coders, since MADIAD handles the setup and testing during deployment. What is needed is a process where the way of working has already been agreed on, not a standing tech team. The business still needs someone who knows the process well to join the current-state survey, since that person provides accurate information about how the work actually runs. Contact MADIAD to book a current-state survey before deciding.
How long does MADIAD take to deploy an Automation workflow? What is the process?+
Timing depends on how complex the process is, but it always follows the same sequence. MADIAD starts with a current-state survey to understand how the process runs today, then designs the architecture and picks the tools that fit, then sets up and tests with real data before moving to live operation. A process that is already stable with few exceptions tends to move through these steps faster than one with many variants. Because every process differs, the specific timeline gets set during the current-state survey itself. Book a survey with MADIAD to get a timeline estimate for your specific process.
What determines the cost of deploying an Automation workflow?+
Cost depends on the scope of work, not on a fixed price for every business. The main factors are: how many steps need automating, how many systems or tools need connecting to each other, how much the logic needs customizing to fit the business's own way of operating, and how much data gets processed each day. A process connecting only two tools differs from one connecting five systems with many branching conditions. See the factors that affect cost at madiad.com/en/pricing, or contact MADIAD for a quote based on the actual scope of the Automation workflow.
After deploying an Automation workflow, what does MADIAD support, and what can the business handle on its own?+
After deployment, MADIAD monitors the system through the early operating period, handles cases the process has not been tested against yet, and updates the setup when the business changes its way of working. MADIAD also trains the people who use the system directly, so they can handle common tasks themselves, like checking run status, reading error logs, or adjusting a few simple parameters. Contact MADIAD for details on support for your specific process.
What should a business prepare before MADIAD deploys an Automation workflow?+
No complex documents needed, but someone who knows the running process well should join the current-state survey. That person should know the specific manual steps, which systems or software are in use, and the exceptions that come up often in the process. If the business already has data on how many times the process runs per week or per month, bringing it along speeds up the survey, but not having it is fine too since MADIAD can estimate it together with the business during the session. Book a current-state survey with MADIAD to start preparing for an Automation workflow.