Seven times we told a client not to automate

We turn down a handful of projects a year on the grounds that the client should not spend the money. It is worth writing down the patterns, because they recur.

1. The volume was too low

A firm wanted document extraction automated. They processed around thirty documents a month, taking about eight minutes each. Four hours a month, roughly 150 dollars of loaded cost. The build was quoted at 2,200 dollars, which is a payback measured in years. We suggested a better folder structure and a keyboard shortcut.

2. The process was about to change

A client wanted to automate a reconciliation process that existed because two systems did not talk to each other. They were replacing one of the systems in five months. Automating the workaround would have been obsolete before it paid back. We told them to wait and helped scope the replacement's integration requirements instead.

3. The problem was a policy problem

An approvals process took eleven days on average. The client wanted it automated. Watching it, the delay was almost entirely one person who was a required approver on everything and was in meetings all day. Automation would have delivered requests to that person faster. The fix was a delegation matrix, which cost nothing.

Automating around an organizational problem makes the problem faster, not smaller.

4. The data was not good enough

A lead scoring model was requested. The CRM had a high proportion of duplicated contacts, inconsistent source attribution, and outcome data recorded for only a fraction of closed deals. Any model trained on that would have learned the recording habits rather than the buying behavior. We quoted a data cleanup instead, at a fifth of the price, and told them to revisit scoring in six months.

5. The hours saved had nowhere to go

A team of four would have saved roughly six hours a week between them. Asked what those hours would be spent on, nobody had an answer. The work was not capacity constrained. The saving would have been real and would have produced nothing, and everybody agreed once it was said out loud.

6. The judgement was the job

A client wanted to automate the qualification of complex enquiries. Reviewing a sample, the qualification decision drew on relationship history, sector knowledge, and reading between the lines of how the enquiry was written. It was not a rules problem wearing a disguise, it was actual expertise. We automated the data gathering that fed the decision and left the decision alone.

7. They needed a person, not a system

A growing business wanted to automate their way out of hiring an operations coordinator. Mapping the work, roughly forty percent was automatable and the rest was judgement, relationship handling, and dealing with things going wrong. Automating forty percent of a role you need does not remove the need for the role. We said hire the person, then automate the forty percent so they can do the rest well.

Why we say it

Partly because it is true, and partly because a client who was told no once believes the yes. Every one of these relationships produced work later, on projects that actually made sense.