TL;DR: you start automation with processes that are simple, mature and proven - not with a big system, and not with a process that limps. Automating a bad process just produces more trouble, faster. In most companies the best first step is the inbound flow of customer enquiries: leads from forms, e-mail and industry portals land in one place by themselves, and the customer gets a reply within a minute instead of waiting for someone to get back to their desk - the average company takes over 40 hours to respond, and a sizeable share never responds at all. Then the next process, and the next - step by step. Once a process matures and grows, the natural next stage is often a custom application: we show this with a deployment where e-mails about product samples grew into an internal “sample shop”.
Where to start? With customer enquiries
Because that is where automation earns its keep fastest and is easiest to measure. Enquiries arrive from several places at once - the website form, e-mail, phone, industry portals - and in most companies they land in different people’s inboxes. Some wait, some get lost. The scale of the problem is well studied: in an audit of 2,241 companies, the average response time to a web enquiry was 42 hours, and 23% of companies never responded at all. A fresher test of 1,000 B2B companies suggests little has changed: only 365 of them replied, on average after more than a day (a study by a speed-to-lead tool vendor, so we cite it with that label).
And time really does the work here: firms that contact a customer within an hour of the enquiry qualify the lead nearly seven times as often as those that reach out even an hour later - and more than 60 times as often as those that wait a day.
A simple automation fixes this without replacing a single system. Tools like Make or n8n wire the form, the inbox and the CRM together without writing code: every enquiry, wherever it came from, lands on one list by itself - in the CRM, though to begin with even a well-set-up spreadsheet will do. The customer immediately gets a reply by e-mail or SMS following a prepared template: confirmation the enquiry arrived, what happens next and when you will be in touch. Everything triggers itself, by agreed rules - at 10 pm on a Sunday too. In the morning the salesperson does not dig through an inbox; they see every enquiry in one place, with history.
Rule number one: automate only what works
Before you automate anything, ask one question: does this process, done by hand, work well? There is a rule in the industry attributed to Bill Gates - automation applied to an efficient operation magnifies the efficiency, and automation applied to an inefficient one magnifies the mess. To be honest: there is no trace of these words in Gates’s own writing; it is received wisdom - but the observation is accurate and we see it in practice. Automating something that works badly only means the trouble now produces itself automatically.
That is why we automate only processes that are mature and proven - ones the team has done by hand for a long time, whose exceptions it knows, and where everyone can tell a good outcome from a bad one. A process invented on the fly, “while we’re automating anyway”, is a straight road to a system that does the wrong things quickly. If the process limps - fix it and rehearse it manually first, automate second.
How it grows: from sample e-mails to an internal shop
This is how it played out at one of the manufacturing companies we work with. The starting point looked familiar: salespeople need samples for customers, so they e-mail the person responsible for producing and shipping them. With a few people and a dozen requests a month, that works. At larger scale the overload begins: requests get lost in the inbox, nobody knows what is in stock, shipping status has to be chased, and all the knowledge about the process lives in one person’s head.
Today that process is closed inside an internal application - written entirely from scratch, for this one process. It works like a shop: a salesperson orders samples the way you buy online. Inside there are warehouses synchronized in real time, full order handling with shipment tracking numbers, and a separate path for non-standard samples that first have to be produced. The whole thing is encrypted, and employees sign in with their existing company credentials - zero new passwords to remember, zero new accounts to administer.
What changed: reaction time and shipping time got shorter. Salespeople order samples more willingly, because the road from need to shipment is short and predictable. The number of unfulfilled orders fell - because stock levels are visible live, before you order. And for the first time the company has full transparency over the process: you can see who orders, what moves and where things get stuck - without involving people in every single order.
Step by step, not big bang
That story shows the healthy order of things. You do not start with a “system for everything” - you start with one process, automate it, measure it, and only then take the next one. Simple automations on ready-made integrations are cheap and fast - they are how you test whether a process is truly mature. When the process grows and integrations stop being enough, you go one level up: a dedicated application, tailored to that one process. How to build it without blowing the budget, we cover in the post on specification in AI-built software.
The same “one process, a number before and after” rule applies to adopting AI - we wrote about it here. That is no coincidence: automation and AI are two tools for the same job - taking work off people that no human needs to be doing.
Security and personal data - from the start, not at the end
Automations process real data: customer enquiries, addresses, order history. So tool selection starts from your organization’s security and data protection policy, not the other way round. Depending on the company’s maturity and internal guidelines, you pick solutions that fit within them - from where data may be processed to how people sign in. An example from the deployment above: signing in with existing company credentials instead of new passwords is not a convenience gadget - it is a smaller attack surface and easier off-boarding when someone leaves.
Is your process fit for automation? A quick test
Run through five questions - the more “yes” answers, the better the candidate:
- Is it repetitive? Same steps, same decision pattern, many times a week.
- Is it mature? The team has done it by hand for a long time and knows the exceptions - you did not invent it last week.
- Does it work well manually? If it limps - repair first, automate second.
- Do errors cost money? A lost lead, an unshipped sample, a mistyped order line - the more expensive, the bigger the return on the automation.
- Does it have an owner? Someone who knows the process inside out and will receive the automation as relief, not threat.
Want to check which process in your company to automate first? We will run this list against your actual processes during a free consultation.
Frequently asked questions
Does automation require replacing the systems we use? No - simple automations are precisely about wiring together the tools you already have: the form, the inbox, the CRM, spreadsheets, the ERP. Replacing a system is a last resort, not a starting point.
At what scale does automation pay off? From the first repetitive process with real volume. Lead flow pays off even in a small company - because the cost of a late reply is disproportionately high, and an automation always answers, outside working hours too.
When do ready-made integrations stop being enough? When the process has logic of its own: stock levels, statuses, exception paths, many users with different roles. Then the natural step is a dedicated application - and thanks to AI the barrier to entry is lower than ever.
Key takeaways
- Start with customer enquiries - aggregating leads plus an automatic reply is the fastest, measurable return; the market average response is over 40 hours, so the bar hangs low.
- Automate only mature, proven processes - automating a mess produces trouble automatically.
- Step by step, not big bang - one process, measurement, the next process.
- Ready-made integrations first, an app from scratch once the process grows - that is how sample e-mails became an internal shop with live stock and tracking.
- You choose security at the start - tools must fit your company’s data policy; sign-in with existing credentials instead of new passwords.
- Automation should remove work, not add systems - if people click more after the rollout, something went wrong.