What not to automate

The interesting question in a small business is not what can be automated but what should not be, and the answer is more than people expect and in different places than they guess.

The jobs that stay with a person

• Anything where the right answer is an exception. Automation implements the rule, and the value in a small business is frequently in the departures from it.
• The first real conversation with somebody who might become a customer. This is the thing small businesses are actually better at, and automating it discards the advantage.
• Anything that has gone wrong. A complaint handled by a workflow becomes a worse complaint.
• Pricing a job that is not standard.
• Money leaving the business.
• Anything whose output is published under your name without anybody reading it.

Automate around the person, not instead of them

The productive pattern in most cases. The judgement stays human and everything surrounding it stops needing attention.

1. 01The information is gathered automatically and arrives assembled, so the person is deciding rather than collecting.
2. 02The draft is prepared automatically, and a person reads and sends it.
3. 03The routine case runs through and the unusual one is flagged, so attention goes where it is worth something.
4. 04The record-keeping happens by itself after the decision, which is the part people hate and skip.

A customer dealing with this experiences a business that replies quickly and knows what it is talking about, which is the actual goal. A fully automated version is faster and worse.

The quiet costs

• You stop seeing your own work. The person who answered every enquiry knew what customers were confused about; the person whose workflow answers them does not.
• Judgement does not get practised, so the ability to handle the exception erodes along with the need to.
• The process becomes invisible, and an invisible process cannot be improved because nobody knows what it does any more.

Which argues for a habit rather than a policy: do a sample of the automated work by hand occasionally. Read ten of the messages that went out. Handle a few enquiries yourself. It is the only way to find out that the workflow has been producing something slightly wrong since March.

Handing a workflow to somebody else

Automation is frequently how a small business survives its first employee, and it is also how that employee is set up to fail. A process that runs through machinery nobody has explained is not a process they can be responsible for.

1. 01Show them what runs, what it does, and where to look when it has not.
2. 02Give them the one-page list, including which accounts things depend on.
3. 03Tell them what they are allowed to turn off, which should be anything that is producing something wrong.
4. 04Make sure one failure does not require you. A process only you can repair is a process that breaks while you are away.

Taking it back

Sometimes the right answer is to switch a workflow off and do the thing by hand again. Volume dropped, the cases got more varied, customers started complaining about something you cannot quite locate.

That is not a failure of nerve. Doing it manually for a month is the fastest way to find out what the automation was quietly getting wrong, and you can rebuild it afterwards knowing the exceptions. The businesses that get into trouble here are the ones that treat switching something off as an admission rather than as a diagnostic.

The thing customers notice

Automation is felt rather than observed. What a customer notices is whether the reply addresses what they actually asked, whether the second message acknowledges the first, and whether anybody appears to have read anything. A business can be heavily automated and feel attentive, or lightly automated and feel like a machine, and the difference is almost entirely whether a person is reading the awkward messages.