Marketing Automation
Automation multiplies your process, including the broken parts
Automation does not repair a lifecycle. It repeats one, faster and at greater volume, which is why a confusing journey becomes a confusing journey nobody can keep up with. We map what actually happens to a person, fix the data the workflows will run on, and build orchestration somebody can still read in two years.
The honest starting point
What automation does to a process that was already unclear
Nothing below is an argument against automating. It is an argument for knowing what you are about to multiply, because the platform will do exactly what it is told at a scale no one is reviewing.
- A workflow from two years ago is still emailing people.
- It was built for a launch that finished, the person who built it has moved on, and it keeps enrolling anyone who matches a condition nobody remembers writing. Every month it quietly contacts people no human has reviewed. It is not switched off because switching it off requires understanding it, and understanding it takes an afternoon nobody has spare.
- The scores are ignored, and everyone knows it.
- Marketing hands over records above a threshold, sales works whichever ones look promising by eye, and the model becomes a report nobody reads. Almost always the cause is that the model was never checked against deals that closed, so it rewards behaviour common to researchers, competitors and students as readily as behaviour common to buyers.
- One person exists three times.
- A record from the website form, another from a webinar list, a third created by a sales rep with a personal address. Each has a different stage, a different owner and a different history, and the automation treats them as three separate people. The result is duplicated sends, contradictory branching and reporting that counts one buyer as three.
- Nobody can describe what happens after the third branch.
- The journey was reasonable for the first two decisions and then grew by accretion. By the fourth branch there are combinations that were never intended, including paths where somebody receives nothing at all and paths where they receive the same message twice. Complexity in a builder looks like sophistication and behaves like a bug.
- The stages mean different things to different teams.
- Marketing calls somebody qualified when they have downloaded enough. Sales calls somebody qualified when they have a budget and a date. The automation moves people between stages using the first definition and reports on it using the second. Every performance conversation then starts by arguing about the vocabulary rather than the result.
A distinction worth being precise about
Email is a channel. Automation is what decides who gets what
These two are sold as the same thing often enough that budgets get allocated to the wrong one. They fail differently, they are fixed by different work, and they involve different people.
| Dimension | Email and lifecycle marketing | Marketing automation |
|---|---|---|
| What it is | A channel: messages sent to an address you are entitled to use | The layer that decides who enters a journey, what happens next and where it goes out |
| The unit of work | A campaign, a broadcast or a sequence | A journey that can branch across email, SMS, ad audiences, sales tasks and alerts |
| Where the work happens | Inside the sending platform, and in DNS | Between the CRM, the website, the ad platforms and everything else that can send |
| What typically breaks it | Authentication, list health, complaint rates, rendering | Duplicate records, undefined stages, fields nobody maintains, workflows without owners |
| What good looks like | Mail that reaches the inbox and earns a reply or a click | A person moves through something coherent and nobody had to press send |
| Who has to be in the room | Marketing, and whoever controls the sending domain | Marketing, sales operations, and whoever owns the CRM data model |
Sequence
Why the tool is chosen fourth, not first
Map the lifecycle as it exists today
Every message, task and alert a person can currently trigger, drawn out including the ones marketing forgets it owns. Most of the opportunity appears as silence on this map rather than as a bad message, and most of the risk appears as a path with no exit.
You get: A current-state lifecycle map with gaps and dead ends marked
Audit what is already running, and turn things off
Every live workflow gets an owner, a stated purpose and a count of who is inside it. Anything that cannot be explained is paused. This step usually reduces the number of active journeys substantially, and it is the one clients are most relieved to have done.
You get: Workflow inventory with a retirement list
Fix the data the journeys will branch on
Duplicates, ownership, required fields, stage definitions and the consent basis on the record. We are not trying to make the whole database perfect. We are making the specific fields that control routing reliable enough to trust.
You get: Agreed data model and a cleaned set of routing fields
Agree the definitions before the build
What a qualified lead is, who decides, what happens when sales disagrees, and how a record gets handed back. Written down and signed off by both teams. Skipping this is the most common reason a technically correct build produces an argument every month.
You get: Written stage and handover definitions
Build a small number of journeys, then instrument them
Welcome and onboarding, a genuine re-engagement path, and the handover to sales. Each documented, each with a review date. Scoring goes in only after there is enough closed data to check it against, never on day one.
You get: Live journeys with documentation and review dates
Score the scoring model
Take deals that closed and lost, replay what the model would have said about them at entry, and remove the rules that did not separate the two. Repeat on a cadence, because the behaviour that predicted a buyer last year will not predict one indefinitely.
You get: A validated scoring model with the working shown
Lead scoring
The check almost nobody runs on their own model
Lead scoring is usually built the same way. A group of people sit in a room, list the behaviours that feel like buying signals, assign points to each, agree a threshold, and switch it on. It is a reasonable first draft and it is presented as a finished system.
The problem is that the room is guessing. Opening three emails, visiting the pricing page and attending a webinar all feel like intent. They are also exactly what a competitor does, what a student writing a dissertation does, and what somebody who is mildly curious and will never buy does. Points get awarded for curiosity because curiosity is easy to observe.
The check is simple and rarely performed. Take the deals that closed over the last year and the ones that were lost. Look at what the model would have said about each at the moment it entered the pipeline. Then ask whether the score actually separated the two groups. If the closed deals and the lost ones scored roughly the same, the model is measuring engagement, not intent, and sales has been right to ignore it.
What tends to survive that check is unglamorous: repeat visits from more than one person at the same company, a request that costs the person something to make, and a match against the profile of accounts that have historically bought. What tends to fail is anything a single anonymous visitor can do in one sitting.
There is a second point that matters commercially. A validated model does not just find better leads, it lets you stop sending the rest to sales. The time saved by not working records that were never going to close is usually worth more than the incremental deals the model finds, and it is far easier to see in the first quarter.
The work
What an automation engagement actually covers
Scope depends heavily on what already exists. A business with a clean CRM and no automation needs almost the opposite of a business with six years of accumulated workflows and a data model nobody has looked at.
- A current-state map of every automated message, task and alert your systems can trigger
- An inventory of live workflows with an owner, a purpose and a population count for each
- A retirement list, and the confidence to actually switch things off
- Duplicate detection and merge rules, with a policy for how duplicates get prevented afterwards
- The specific record fields that control routing, defined, populated and maintained
- Consent basis and its source stored on the record, so a journey can branch on entitlement
- Written definitions of every lifecycle stage, agreed by marketing and sales together
- Orchestration across channels rather than inside one, including sales tasks and internal alerts
- Lead scoring built after there is closed data to validate it against, then validated on a cadence
- Suppression that every sending system honours, including the CRM and individual sales users
- Documentation written for the person who inherits this, not for the person who built it
- A quarterly review where journeys are removed as readily as they are added
Questions
What operations teams ask before they start
How is this different from email marketing?
Email is a channel. Automation is the layer that decides who is in a journey, what happens next, and which channel the next thing goes out on. A well-run email programme can exist with no automation platform at all, and a well-built automation programme sends through email, SMS, advertising audiences, sales tasks and internal alerts.
In practice the two overlap heavily, because email is where most automated journeys do their work. The distinction matters when you are deciding what to fix. If messages are not reaching the inbox, that is a channel problem. If the right people are receiving the wrong thing at the wrong moment, that is an orchestration problem, and no amount of channel work will touch it.
Our CRM data is a mess. Should we clean it first or start building?
Clean it first, at least to the point where a record can be trusted to say who somebody is and what stage they are at. Automation acts on fields. If the field is wrong, the workflow is confidently wrong for every person it touches, and it does not stop to check.
That does not mean a year of data projects before anything ships. It means agreeing the handful of fields the journeys will actually branch on, fixing duplicates and ownership on those, and building the first workflows only on the parts of the record you can defend.
Does lead scoring actually work?
It works when it is validated and it becomes noise when it is not. The test is straightforward: take the deals that closed in the last year, look at what the model would have said about them at the point they entered the pipeline, and see whether the score separated the winners from the rest.
Most models fail that test the first time, usually because they reward activity that any curious person performs rather than behaviour that only a buyer performs. That is a useful finding, not a failure. The rules that survive the check are the ones sales will start trusting.
How many workflows should we have?
Fewer than you have now, in almost every case we look at. The number that matters is not how many exist but how many somebody can explain, and the gap between those two figures is where the risk lives.
A good rule is that any live workflow needs a named owner, a written purpose, and an answer to the question of who is currently inside it. Anything that fails those three gets paused and reviewed rather than left running because turning it off feels risky.
Can you work with the platform we already own?
Usually yes, and we would rather do that than sell you a migration. The platforms differ less than their marketing suggests, and most of the value is in the lifecycle design, the data model and the discipline around what gets built.
We will recommend a move only where the current tool genuinely cannot hold what you need: stage and consent data on the record, a single suppression list honoured by everything that sends, and results written back to the system your sales team lives in. If it fails those, we will say so and explain what the move costs.
Find out what your platform is currently doing without you
The most useful first piece of work is usually an inventory: every live workflow, who owns it, and who is inside it right now. It takes days rather than weeks and it changes the conversation about what to build next.
If we don't deliver the work we agreed to deliver for reasons within our control, you don't pay for the undelivered work. Read our guarantee
Related
Where to go next
- the email channel itselfAutomation decides who gets what. Email is one of the ways it arrives.
- the tracking a scoring model runs onA scoring rule is only as good as the event behind it.
- text messaging as a channel in the journey
- the growth practice this belongs to
- raising the standard of enquiriesScoring is only worth building where quality is the problem.
- what happens on the page before the workflow starts
Last updated · Reviewed by Zubair Afzal