Perspective / 04

Automation makes a broken sales process worse

Automation removes time and friction. In a sales process that was never fully designed, some of that friction is where unresolved judgement lives.

Perspective

Independently created. Contains no employer or client implementation detail, internal names or figures.

The thesis

Automation removes time and friction from a process. In a sales process that was never fully designed, some of that friction is carrying judgement nobody wrote down. Remove it without redesign, and the process does not improve. It gets faster at being wrong.

The familiar advice is right, but imprecise

“Do not automate a bad process” is good advice that rarely changes anything. Nobody sets out to automate a bad process. The process looks reasonable on a slide, the delays are obvious, and automation promises to remove them. The advice does not say which part of the process is bad, or why removing the delay would make it worse.

A more useful version starts from what automation actually does. It removes time between steps and removes the effort of moving work from one person to the next. Both are usually welcome. The question is what that time and effort were doing.

Some friction is doing work

In a sales process that has grown rather than been designed, manual handoffs often contain checks nobody documented. The inside sales rep who reads a lead before passing it on notices that the company is already a distributor’s customer. The manager who signs off a quote looks at the margin and asks a question. The order desk that types the order into ERP spots that the delivery address belongs to another legal entity.

None of these checks appears in the process map. They live in the friction: in the minutes a person spends looking at the work before moving it. Automate the movement and the looking disappears with it. The process gets faster, and some of the judgement that kept it honest is gone without anyone deciding to remove it.

Routing, transformation and judgement are different work

It helps to separate three kinds of work that broken processes tend to mix in one step. Routing moves work to the right owner: assigning a lead, notifying an approver. Transformation computes, copies or formats: a price from a list, a document from a template, a record from a form. Judgement decides something under ambiguity: whether this deal is worth pursuing, whether this exception is acceptable, whether this customer is the one we think it is.

Routing and transformation are good candidates for automation, provided they are tested and observable. Judgement is different. It either stays with a person, at a deliberate point in the flow, or it is made explicit enough to become a rule, with someone accountable for the rule. What does not work is leaving judgement implicit inside a step and then automating the step.

Exceptions and irreversible actions

Informal processes handle exceptions informally. Someone calls someone. Automation cannot do that. An exception the flow does not model is either dropped, forced through the standard path, or left waiting in a state nobody watches. The faster the process, the sooner these cases accumulate.

It is also worth asking which actions can be undone. Updating a field is reversible. Sending a quote to a customer, creating a customer in ERP or confirming a delivery date is not, or only at a cost. Irreversible actions deserve an intentional stop: a point where a person decides, with the evidence in front of them. A human step placed there is not a failure of automation. It is part of the design.

Failure needs the same attention. When a step fails halfway, what state is the work in, who knows, and can it be retried without creating a second order or a second customer? Manual processes answer these questions by improvising. Automated ones need the answer built in.

What to redesign before automating

Before automating a sales process, I would make a few things explicit. The states the work moves through, and what must be true to leave each one. The decisions inside the process, with an owner for each. The exceptions that matter, and where each one goes. Which steps are routing, which are transformation and which are judgement. Which actions are irreversible and therefore need a deliberate stop. And how failures are detected and recovered without duplicating work.

With that in place, automation becomes the easy part. It accelerates a process whose assumptions have been examined, instead of accelerating the assumptions nobody looked at.