Workflow automation for small business: automate the chain, not one task

What a workflow is, and why the chain beats single tasks

A workflow is the route one piece of work travels from arrival to completion. At an installation company: enquiry in, follow-up call, quote out, customer signs, job scheduled, engineer on site, job sheet back, invoice. Eight steps, four systems, three people.

Automating single tasks picks off one of those steps. The quote gets drafted faster. Fine, but the work still sits there until someone checks and sends it. You have made the fastest step faster and left the waiting untouched.

The gain is in the transitions. If the signed quote creates the order itself, that order lands in the schedule itself, and the completed job sheet drafts the invoice itself, nobody has to pick anything up. That is also the honest answer to what it returns: rarely a saved head count, usually an invoice that goes out weeks earlier.

Where time and mistakes actually leak away

Almost every chain we map shows the same leaks. They appear on no timesheet, because they hide inside work everyone considers normal.

• Retyping between systems. The same customer details in email, in the CRM, in the scheduling package and in the accounts. Four entries is four chances of a typo, and it usually surfaces at the customer.
• Waiting for approval. A quote above a certain amount has to go past the director, who is on a roof or at a customer. The work stops because nobody knows it is ready.
• Duplicate entry. An order that sits in the ERP and in a spreadsheet. While both agree you notice nothing; the moment they differ, you get an argument about which one is right.
• Work that stalls when someone is off sick. At a wholesaler one person usually knows how supplier confirmations are processed. If that person is away a week, the pile grows.
• The last metre. Job sheets that are finished but still in the engineer’s van. Between “done” and “invoiced” there is often more time than in the job itself.

Mapping your workflow: you can do this yourself

You do not need an agency for this. An afternoon and the people who actually do the work is enough. Do it before you talk to anyone about automation, including us.

1. 01Pick one type of work and follow it end to end. Not “the admin”, but: a repair job for a regular customer.
2. 02Write every action on its own line, including “copy it out of the email” and “ring to check”. Those are exactly the steps no procedure mentions.
3. 03Add who does it, in which system, and how long it takes. Estimate generously; people underestimate their own manual work.
4. 04Mark every point where work passes from hand to hand or system to system. Those are your handovers.
5. 05For each handover, add how long the work sits there. Not the action, the waiting. Usually the most sobering number of the afternoon.
6. 06Count your exceptions for a month instead of guessing: rush jobs, customers with their own arrangements, orders without an order number.
7. 07Measure one number before you change anything, for example the days between job completed and invoice sent. Without a baseline you will never know afterwards whether it helped.

Fixed rules or AI: which do you use where

Every chain contains two kinds of step. One follows a rule you can write down. The other needs a judgement. Mixing the two up is the most expensive mistake in this work.

A fixed rule is: if an order goes above the agreed amount, the director has to approve it. You handle steps like that with ordinary integrations and rules. No AI needed, and that is good news: cheaper, faster, and it never does anything unexpected.

You need AI as soon as a step can only be completed by reading or weighing something — an email in which a customer describes a fault in their own words. No fixed rule catches that. In practice it is a combination: the AI reads the incoming post and prepares a proposal, the rules decide what happens next. The AI does not decide.

• Fixed rules: routing, creating, copying, scheduling, checking an amount or a date, chasing, escalating when something stalls.
• AI: reading free text, extracting from documents, sorting by subject, summarising, drafting something a human checks.
• Neither: anything genuinely negotiated, and anything someone personally signs off.

What you deliberately keep manual

Automating a chain completely sounds better than it is. In almost every workflow we keep one or two steps in human hands on purpose.

• The final check before money or a customer is affected. One wrongly sent invoice wipes out a month of time saved.
• Exceptions that rarely occur. Something that happens a few times a year costs more in maintenance than it saves.
• Steps where the customer expects a person. A complaint handled automatically feels like a complaint that was not heard.
• Work where the rules are not settled yet. If you are still arguing internally about how it should go, it is too early for software.

Workflow automation compared with the alternatives

The rule we work by: automate the transport, not the judgement. There are roughly four ways to fix a chain that keeps jamming.

• Buy a new package that does everything. Works if your software genuinely falls short and you are willing to fit your process to the package. Costs a migration, a learning curve and usually a year.
• Hire someone. Immediate and flexible, but you are buying capacity, not reliability: the same handovers and the same typos, with more hands.
• Automate single tasks with a no-code tool. A fine start and often cheap. It starts to hurt once you have fifteen automations nobody oversees, breaking at the next update.
• Take on the whole chain. More expensive to start and you have to map your process first. But it addresses the waiting and the handovers, and that is usually where the money is.

How a project runs with us

We do the fourth, and we connect to what you already have: accounts, CRM, ERP, scheduling, Office. If a standard package is enough, we will say so. You can stop after any of the steps below.

1. 01Free AI scan. Five minutes online, with questions that adapt to your business. You get a score, concrete opportunities and a first price indication.
2. 02Core session on site. A half-day (4 hours) at your office or shop floor, €596 excluding VAT and travel. We follow one chain end to end and count the actions. Full day: €1,192.
3. 03Core report with a go or no-go. Per opportunity: what it returns, what it costs and whether we would do it. The report is yours, even if you do nothing with it.
4. 04Building in short cycles. €110 per hour for straightforward work such as integrations and configuration, €165 for complex development. Fixed hours per month, paid in advance from a build budget.
5. 05Going live on the floor. We only hand over when it runs in your practice. For the first weeks a human checks every outcome; only then do you let go.

What workflow automation costs and what it returns

Do not count the return as “time saved” in general, because saved minutes disappear into the day. Pick one or two numbers you already track: days between job completed and invoice sent, or orders that had to be redone because something was copied across wrongly. Measure them before you start and a few months later.

Against that sit three cost lines, plus a fourth that appears on no invoice: your own people’s hours to explain, sit in on tests and check the results.

• Build cost, one-off. Investigating, configuring, integrating and testing at €110 or €165 per hour, plus the €596 core session.
• Running cost, monthly. Licences, integration fees some packages charge separately, hosting and the AI model’s own usage.
• Maintenance, ongoing but irregular. A supplier changes an integration, your process changes, someone finds an edge case. Budget for maintenance every year.

Where it goes wrong

Workflow automation rarely fails on the technology. These causes come back every time.

• The process is still changing. If you are still moving who does what this quarter, you are fixing a moving target in software.
• The exceptions are half the work. At some businesses every order is slightly different; then the standard route is the minority. Fix the process first, not the technology.
• There is no owner. Without one person on your side who gets time for it and watches the first weeks, it fizzles out. The most predictable cause of failure there is.
• Automating a messy process as it stands. You get the same mistakes, only faster. Take the pointless steps out first.
• The data is not in order. Duplicate customers, products spelled three ways, orders without a number. Cleaning up is dull manual work up front and it appears in no demo.
• Too much at once. One chain that genuinely runs is worth more than three that are half finished.

Frequently asked questions

What we do for this

Further reading

Other topics

Start with one chain, not with a plan

Take the free five-minute AI scan, or book a core session on your shop floor. We follow one workflow end to end and tell you honestly what there is to gain — including when the answer is: not much.