What you should not automate (even if you can)
You usually arrive with a list of tasks you want off your plate. Some of them are good candidates. Others are not, and building an automation on top makes the problem worse, hides it, or makes it more expensive to fix. This piece is about those others: how to spot them before you pay for them.
An automation does not fix the process, it repeats it
An automation fixes nothing on its own. It takes what you already do and repeats it tirelessly and without judgement. If the process you hand over has one badly defined step, that step now runs far more often and always the same way. Before, someone handled it and noticed the fault. Now it runs by itself and nobody sees it until a customer complains.
So the first useful question is not which tool to use, but what happens today when the process breaks. If nobody can answer that, there is nothing to automate yet; there is something to tidy up. Tidying it up is usually boring and unglamorous. Write down the steps, decide who decides, drop the exceptions that no longer serve a purpose. Once that is done, whoever builds the flow does not have to guess the rules along the way: there is less to assume and fewer exceptions to patch. When it is not done, what you buy is a software layer on top of the mess.
A process that has not settled yet
Some teams have been doing something a certain way for a very short time and already want it automated. That is understandable: the newest task is the one that annoys you most. But a process that has just been born is still finding its shape. That first stretch is exactly what surfaces the odd cases, the exceptions nobody had anticipated and the steps that are not needed. Freeze it into a flow too early and you freeze the design mistakes with it. Fixing them later costs more, because they no longer depend on a person but on an integration someone has to touch.
The signal is simple. If describing the process brings out phrases like 'it depends', 'we sort that out as we go' or 'last month we did it differently', it is not time yet. Keep doing it by hand a while longer, write down every exception that shows up, and automate afterwards, once the shape repeats on its own. That is not lost time: it is gathering the information anyone building the flow will need.
A process whose rules change every month
A different case: the process works, but its rules change often. Prices that get revised, short campaigns, criteria that depend on a season or on a regulation. Here the automation can indeed be built, but you have to account for the cost of keeping it alive. Every rule change means editing the flow, testing it and taking up someone's time. If that time is more expensive or slower than doing the task by hand, the automation does not hold up, however well it was built.
There is a middle path worth putting on the table: separate what is stable from what is volatile. The part that does not change (picking up the message, identifying the customer, filing the record where it belongs) gets automated. The part that changes moves somewhere you can edit it separately: a sheet, a table, a small panel. That way most rule changes are made by your own team, without touching the flow. It looks less impressive than a flow that decides everything by itself, and it does not force a rebuild every time a rule changes.
When the value is precisely that a person does it
There are conversations where the content of the message matters least. A complaint that has gone badly, a negotiation, the call to a worried patient, the long-standing client who wants to speak with you. The value there is not in the correct answer: it is in the fact that somebody bothered. An agent can write a flawless message and still leave the feeling that nobody was on the other end. That cost appears in no calculation of time saved, but the business pays it.
This does not mean leaving those conversations unsupported. It means putting the machine before or after, not in the middle. A voice agent can take the call in the middle of the night, capture the details and alert the right person; the chatbot can handle opening hours, availability or the status of an order and pass the rest to a human being, with the context already gathered. The practical rule: automate the work around the hard conversation, not the hard conversation.
When a tool for that already exists
In hospitality, real estate, e-commerce or clinics there is already software that covers a good part of what you need: bookings, listings and portals, catalogue and orders, scheduling and records. If your case fits what that tool already does, commissioning custom development to repeat the same thing is money thrown away. Buy the licence, learn to use it properly and keep the budget for what it genuinely does not cover.
What those tools do not usually cover is the joins between them. The order that arrives on WhatsApp and has to be copied into the management system, the booking that comes in by phone after hours, the enquiry from the property portal that nobody looks at until Monday. That is where the useful work sits, and it tends to be narrower than it looks. So before suggesting you replace a program that already works, we look at what can be joined up between the parts that do not talk to each other today.
What an automation costs after it is delivered
It is worth being clear about the other side. A well-built automation is not free once it is delivered either. Integrations break when a provider changes something, the channels you use change on their own, and a voice agent runs into accents, background noise and requests nobody had imagined. All of that can be managed, but somebody has to look at it now and then. If nobody in your business is ever going to look at it, factor that in before you decide.
That is why we hand over training when a project ends. Not out of generosity: a system only its builder understands is a fragile system. If your team knows where to look when something fails, how to change a message and when it is better to switch the flow off, you stop depending on us for the small things. It is worse business for an agency in the short term, and it is what keeps the project standing when we are no longer around.
Ruling things out is part of the job too
If you sit down with someone, tell them all your ideas and they like every single one, what they are doing is billing you hours. Ruling things out is the expensive part of the job, because it forces you to understand the business and to turn work down. A useful assessment sometimes ends in 'you can solve this with what you already have' or 'wait until it settles and then we talk'. That earns nothing that day, but it is what stops you paying for something you were never going to use.
That is how we work from Spain: we look at the process first, then decide whether it should be automated at all. When it should, most projects ship in five to seven working days, and we reply in under twenty-four hours. There is no catalogue price, because it depends on the channels, the volume and the integrations you already have. If you want an opinion before a quote, write to info@zepaiagency.com or call +34 604 14 09 97.
Want to build this in your company?
Half an hour, no sales pitch. And if we think it is not worth it for you, we will say so.
Get a free quote →