Campaigns and automation

Audit what already runs before building anything new

The first question on an automation project is never what to build. It is what is currently running, and in most accounts nobody can answer it.

Automation accumulates quietly. A flow built for a campaign two years ago is still sending because nobody was sure it was safe to turn off. Three lists overlap and a contact sits in all of them, receiving three versions of the same message in a week. A scoring rule adds points for a page visit that no longer exists. None of this shows up as an error, so it persists.

So we start by mapping what runs: every active flow, its trigger, its audience, and when it last did anything. That map routinely finds flows that can simply be switched off, which is the cheapest improvement available and frequently the largest. It also tells you whether your problem is that you need more automation or that you need less of it, and the answer is less more often than anyone expects.

---

Entry and exit criteria, in writing, before the build

The two questions that prevent most automation disasters are: who enters this flow, and what takes them out of it. Both sound obvious and both are routinely left implicit.

A flow with no exit criterion keeps sending to someone who has already bought, already replied, or already asked to be left alone. A flow with a vague entry criterion picks up contacts it was never meant for, which is how a nurture sequence written for new prospects ends up in the inbox of a customer of six years.

We also insist on suppression as a first-class part of the design rather than an afterthought: who must never receive this, under what circumstances, and how that is enforced across every flow rather than remembered per campaign. And every flow gets a documented off switch, so that stopping it in a hurry is a known operation rather than an emergency.

---

Lead scoring that sales will actually use

Most scoring models are built entirely inside marketing and are therefore ignored entirely inside sales. The model assigns points for email opens and page views, sales discovers the scores do not predict anything, and within a quarter nobody looks at them.

A model that gets used is built the other way round. Start with a set of deals that closed and a set that did not, and ask sales which behaviours actually distinguished them. Usually it is a small number of unglamorous signals: a pricing page visit, a repeat visit within a short window, a specific form, a job title. Behavioural signals generally beat demographic ones for timing, and demographic signals beat behavioural ones for fit, which is why the two are worth scoring separately rather than collapsing into one number.

Then the threshold has to be agreed with the people who will act on it, and the model has to be reviewed on a schedule, because a scoring model built once and never revisited is describing a business that no longer exists.

---

Segments with one owner and one definition

The failure mode here is duplication. Someone needs a list of enterprise prospects, cannot find the existing one, and builds a new one with slightly different criteria. Six months later there are four enterprise segments, all in use, none matching, and every report built on any of them disagrees with the others.

The fix is administrative rather than technical: a naming convention, one named owner per segment, and a written definition stored where people can find it. Dynamic membership rather than static lists wherever the platform allows it, so a segment stays true as records change instead of decaying from the day it was built.

It is unglamorous work. It is also what makes the difference between a platform your team trusts and one they work around with exports to spreadsheets.