Strategy
Part of Marketing automation strategy: platform control tests to run
A marketing automation ninety day plan that puts your platform to the test
A marketing automation ninety day plan in six dated phases that tests your platform, from HubSpot to Braze, on suppression, rollback and measurement first.
This marketing automation ninety day plan aims to reach a defensible continue, change or stop decision. It does not assume that a live journey will launch. It also works as a hands-on review of the platform you run or are trialling: by the end of the build phase at day 60 the platform has either produced evidence that it can suppress, roll back and measure your first journey, or it has failed in your own environment, which tells you more than any vendor comparison. Dates are planning windows, not a claim that every organisation can work at the same speed.
What to take away
- The plan reaches a defensible continue, change or stop decision rather than assuming a live journey will launch.
- Each phase has an accountable owner role, defined inputs and a written output before the next stage begins.
- Stop conditions include failed suppression, uncontrolled audience expansion and measures that cannot be reproduced.
- No threshold in the guide is a benchmark, so each organisation must set its own limits from evidence.
- A limited live test is authorised only after every required gate passes.
Days 1 to 15: define the decision
Owner: process lead. Inputs: workflow evidence, service contacts, data notes and current messages.
Decision brief and uncertainty log
- Write observable problem
- Define eligible audience
- List exclusions
- Map activity to output
- List assumptions
- Sign decision brief
- Log uncertainty
Write one observable problem, eligible audience and exclusion list. Map the proposed activity to its output and intended outcome, then list the assumptions in between. HM Treasury's current Magenta Book recommends developing evaluation around the intended causal route, existing evidence and reasons the intervention may not work.
Output: signed decision brief and uncertainty log.
Stop if no owner can change the process or the problem cannot be located in existing evidence.
Days 16 to 30: establish the baseline
Owner: analyst. Define each measure's event, numerator, denominator, eligible population, period, exclusions, time zone and system. Include one outcome measure, a guardrail for harm or workload and diagnostics for rule failures.
The GOV.UK Service Standard says teams should define success and use performance data from all relevant channels in its measurement standard. Apply the measurement principle, not government-specific mandatory KPIs, to the marketing plan.
Output: metric dictionary, baseline extract and data-quality note. Pause when the denominator cannot be reproduced.
Days 31 to 45: design the rule and gates
Owner: operations lead. Specify trigger, required fields, exclusions, timing, frequency, suppression, exception queue, manual override, pause switch and rollback. Data protection and legal reviewers document the purpose, lawful basis under UK GDPR and the Data Protection Act 2018, PECR channel analysis, retention and preference route.
For email or SMS, use the ICO's April 2026 electronic-mail guidance as a current input. It cannot approve the organisation's particular facts.
Output: versioned rule card and named gate log. Do not configure a live audience while any required decision remains blank.
Days 46 to 60: build a controlled path
Owner: technical lead. Use synthetic or appropriately governed test records. Exercise duplicate triggers, missing data, conflicts, opt-outs, failed delivery, manual review and rollback. Confirm that logs explain why each record was included or excluded.
Platform choice does not change these phases. The same gates apply whether the organisation runs HubSpot, Marketo, Klaviyo, Mailchimp, ActiveCampaign, Braze, Dotdigital, Adobe Campaign, Salesforce Marketing Cloud or Customer.io; what changes is the build, because each platform's own journey logic, suppression and consent settings, integration behaviour and limits must be exercised and evidenced in its own environment.
If an AI component proposes a score, segment or message, define its permitted input, human authority, error route and monitoring. The ICO AI guidance collection provides data-protection risk and explanation resources. Qualified reviewers must assess the actual use.
Output: test evidence and unresolved-defect register. Stop if the team cannot reconstruct an action.
What the named platforms change in the build
The gates in days 46 to 75 are the same for every vendor, but the evidence each gate needs is not. Before the bounded test, exercise the trigger, the suppression and consent settings, the integration or data feed, the frequency limits, the exception queue and the pause or rollback switch inside the platform the organisation actually runs.
- HubSpotjourneys run from the CRM contact record, so ownership of contact properties, lifecycle stage and subscription type decides who enters and who must be suppressed.
- Marketo (Adobe)smart lists and engagement programmes set entry, so the smart-list logic and the data it reads have to be tested as one route.
- Klaviyobuilt on e-commerce profile, order and catalogue events, so the store integration, not the message editor, is where segmentation and suppression are proved.
- Mailchimpan email-led tool where audience tags, groups and journey settings carry the exclusions; free and paid tiers differ in automation depth, so test the tier actually licensed.
- ActiveCampaignemail automation and a CRM share the contact record, so deal stage and automation entry must be tested together to avoid contacting the same person twice.
- Brazea cross-channel engagement platform for consumer apps, so per-channel subscription state and frequency capping need their own test before the first send.
- Dotdigitala UK email and cross-channel automation provider, so consent and preference data held in its address books carry the suppression test.
- Adobe Campaignenterprise campaign management inside Adobe Experience Cloud, so workflow templates and delivery configuration define the exclusions that must be evidenced.
- Salesforce Marketing CloudJourney Builder and data extensions drive the route, so the data extension query and the entry and exit criteria are the artefacts to evidence.
- Customer.iobehavioural messaging fed by product events and attributes, so the event stream that triggers a message needs the same reconstruction test as the message itself.
Each platform is judged here only on evidence produced in its own environment, not on a vendor's feature list. A tool that cannot demonstrate suppression, rollback and reproducible measurement fails the plan at day 60, whatever else it offers.
Record the platform review in the same gate log, one line per test:
| Test | What the platform must show | Verdict |
|---|---|---|
| Entry rule | The record that triggered the journey and why | Pass or fail |
| Suppression | An opted-out or excluded record never entering | Pass or fail |
| Consent and preferences | A withdrawal reaching the platform and stopping sends | Pass or fail |
| Frequency limit | A second trigger held back within the limit | Pass or fail |
| Pause and rollback | The journey stopped and in-flight records handled | Pass or fail |
| Reproducible measure | The same count rebuilt from the platform's export | Pass or fail |
A platform that passes every line is fit for the bounded live test; one that fails any line is the reason to change the build, the licence tier or the platform before day 76.
Days 61 to 75: test the full customer route
Owner: quality lead. Check the message, link, landing page, form, validation errors, confirmation and alternative contact route. The W3C WCAG 2.2 standard provides technical success criteria, but a complete assessment requires human judgement and the applicable legal context.
Run preference, withdrawal and suppression tests through connected systems. Ask support staff to rehearse exception handling. Output: journey test report, accessibility findings and release recommendation.
Days 76 to 90: bounded live test or stop
The accountable executive may authorise a limited test only after every required gate passes. Fix the population, period, concurrent changes, reporting frequency and stop conditions before release. Review outcome, guardrail and diagnostic measures together; never replace a missing outcome with a convenient activity count.
Stop immediately for a failed suppression, material accessibility barrier, uncontrolled audience expansion, irreproducible measure or severe exception backlog. Otherwise, end the period with one recorded decision: retire, redesign, continue within the boundary or prepare a separately approved expansion.
No threshold in this guide is a benchmark. Each organisation must set its own limits from evidence, capacity, customer risk and qualified review.
Before you act
- Name one owner who can change the process.
- Write the observable problem, audience and exclusions.
- Define each measure's event, numerator and denominator.
- Document purpose, lawful basis and retention before configuration.
- Test suppression, opt-outs, rollback and exception handling.
- Fix population, period and stop conditions before release.
Common questions
What must be in place before a limited live test can start?
The accountable executive may authorise a limited test only after every required gate passes. Before release, the team must fix the population, period, concurrent changes, reporting frequency and stop conditions. Review outcome, guardrail and diagnostic measures together, and never replace a missing outcome with an activity count.
When should the team stop the work entirely?
Stop if no owner can change the process or the problem cannot be located in existing evidence. Pause when the denominator cannot be reproduced. Stop if the team cannot reconstruct an action. Stop immediately for a failed suppression, material accessibility barrier, uncontrolled audience expansion, irreproducible measure or severe exception backlog.
How should AI components be handled in the build phase?
If an AI component proposes a score, segment or message, define its permitted input, human authority, error route and monitoring. The ICO AI guidance collection provides data-protection risk and explanation resources, but qualified reviewers must assess the actual use. Output is test evidence and an unresolved-defect register.
