Choosing marketing automation tools with the operating problem defined before the demo

Choosing marketing automation tools and providers for 2027 should begin with a controlled business process, not a feature leaderboard. An England-based buyer needs to know which journeys are worth automating, what data may enter them, who can approve a send and how the organisation will leave the service. Only then can a product demonstration answer useful questions.

Research for this 2027 purchasing guide closed on 5 September 2026. It is desk research: no platform, connector, migration or support service was tested.

Supplier pages show what their publishers currently describe and sell, but do not prove fitness, compliance, accessibility, security or results for a particular organisation. Prices, editions and availability can change before a 2027 purchase. The final shortlist requires fresh written quotations and named qualified UK legal, data-protection, security, accessibility and commercial review.

What to take away

Define the operating problem before viewing software

Start with two or three journeys that already have an owner and a defensible purpose. For each, record the trigger, recipient, channel, data inputs and decision rules. Then record content, approval point, suppression source, expected output and stopping event.

A useful requirement: within one working day of an eligible enquiry, create a task for the assigned sales owner. Send one approved email only where the channel rule permits it. A vague engagement goal is not testable.

Mark hard exclusions beside those journeys. These may include unsupported identity management, an inability to prevent unapproved publishing, no usable audit trail, inaccessible authoring, an undocumented export route or a contract that leaves deletion responsibilities unclear. A supplier that misses an exclusion should not reach a weighted score, however polished its demonstration.

England is the operating context, though much relevant law is UK-wide. Software permission is not legal permission. The ICO's electronic-mail guidance covers separate PECR questions for email, text and comparable stored messages.

The ICO's controller and processor contract guidance sets required processing terms, but does not settle the wider commercial bargain. Both pages need publication-day review, as relevant ICO material may change after the Data (Use and Access) Act 2025.

Build an evidence ladder

Do not give every sales statement equal weight. A practical evidence ladder has four levels:

Capture gaps rather than filling them with assumptions. If a web page says that a product connects to other applications, it does not establish that the required object, direction, field, latency or error handling works. Ask the supplier to demonstrate the exact exchange, then make the buyer reproduce it without hidden administrator help.

The same discipline applies to artificial intelligence. Record the model or service involved, whether customer content is used for training, available human controls, logging, regional availability, consumption units and the behaviour when the feature is disabled. An announced capability is not a reason to place it in a production journey.

A dated discovery view of five products

The following products illustrate different starting points. They are listed alphabetically, not ranked or recommended. Inclusion required a live first-party automation record and a live first-party price, packaging or sales-contact boundary on the research date. Agencies and implementation partners are outside this product view.

The table is a route into discovery, not a complete market review. Five public sites cannot establish product quality or the range of suppliers serving England. Features can also sit behind particular editions, add-ons, consumption credits or regional releases. The buyer's requirements and controlled tests decide whether any entry remains relevant.

Editions and units named in public records

Each candidate publishes a pricing unit, not a bill. Taxes and fees sit outside headline figures, and Mailchimp marks non-dollar amounts as exchange-rate estimates. A dated quote in pounds is therefore the workable comparison. No vendor paid for a place in this view.

Recheck these names, units and boundaries on the decision date, because packaging and regional availability move. Request UK billing terms before any trial and model the contact tier you expect to run.

Ask Microsoft which component you need, then test profile counting, qualifying licences and failed interactions. For Salesforce, test the journey, administrator roles, add-on boundaries and the extraction route, and agree who owns consent data in the CRM.

Compare the full cost boundary

Licence headline costs are rarely decision-ready, so normalise each proposal to one period and scenario. Record currency, tax treatment, billing frequency, minimum term, renewal mechanism.

Record users, marketable contacts, stored profiles, sends, interactions. Record environments, API calls, artificial intelligence or message credits, support tier, expected overage.

Keep one-off discovery, configuration, migration, integration, training visible, not buried in monthly totals. Keep assurance and exit work visible.

Calculate at least three volumes: the current clean dataset, a credible operating case and a stress case. Ask how suppression records, duplicates, inactive contacts and historical profiles count. The aim is not to predict a perfect 2027 bill. It is to expose which measurement can move the charge and which team will monitor it.

Treat discounts separately from the underlying model. A temporary offer can be recorded with its expiry and eligibility, but procurement should also show the undiscounted renewal case. Where a supplier provides only customised pricing, preserve the assumptions in the written quote so two bids can be compared on a common basis.

Keep six assurance tracks separate

One broad "compliance and security" score hides different decisions:

Assign an owner and pass condition to each track. A privacy specialist should not be expected to approve penetration-test evidence, and a security lead should not interpret the PECR status of a recipient. The purchasing decision is the point where their separate conclusions meet.

Consent and the processor boundary

Each platform acts as a processor for the personal data a buyer loads into it. Consent, lawful basis, retention and deletion stay with the buyer.

A consent record the business cannot produce on request does not evidence consent.

PECR's soft opt-in covers existing customers only, so a bought list needs another lawful basis. The ICO's direct marketing guidance under PECR sets out the consent and soft opt-in tests.

Ask each vendor where data is hosted, which subprocessors are used, how long records are kept and how deletion is proved. Agree the deletion rule before any import, and record where the platform hosts and transfers personal data.

Run a buyer-controlled proof

Give shortlisted suppliers the same script. Use a small non-production dataset or approved synthetic records, with keyboard-only authoring and rendered-message checks.

Ask the buyer's own operator to build a journey, apply a suppression, change a permission and recover from a connector failure.

Export the resulting records and find relevant audit evidence, and record every supplier intervention, as assisted success may not reflect routine operation.

Acceptance thresholds should be observable. Examples include no send after a central opt-out reaches the test system, rejection of a user without publishing authority, an alert when a required integration fails and an export that retains identifiers needed for reconciliation. These are test designs, not claims that any product will pass.

Stop the exercise if protected data appears outside the approved boundary, a required control cannot be demonstrated, material costs remain undefined or the supplier substitutes a different edition from the one quoted. A failed test can be more valuable than a persuasive presentation because it prevents an unsuitable commitment.

Decide with an exit already designed

Score only evidence available for the proposed configuration. Keep hard exclusions outside the arithmetic, cap scores where evidence is merely a supplier assertion and document who accepted each residual risk. The decision note should include rejected options and reasons, not just the chosen platform.

Before signature, rehearse departure on paper, then list notice dates, export formats, technical dependencies, supplier help and internal owners. Add parallel-running needs, deletion evidence and where suppression records live after exit.

The ICO's contract guidance requires processor terms to cover returning or deleting personal information at the controller's choice when services end, unless law requires storage. That regulatory requirement does not answer every practical migration question, so the commercial schedule must do more.

For 2027, the sensible next step is a requirements workshop, not a sales call. Select two journeys, document exclusions, assign the six assurance owners and issue one evidence request to every candidate.

Freeze the articles and quotations used in the decision. Recheck supplier pages, ICO guidance, price units and contractual terms immediately before approval.

Publication of this guide also remains on hold until the named specialists have completed their reviews.

Before you act

Common questions

Why should a buyer define the operating problem before seeing a demo?

The article says choosing tools should begin with a controlled business process, not a feature leaderboard. A buyer must know which journeys are worth automating, what data may enter them, who approves a send and how the organisation will leave the service. Only then can a demonstration answer useful questions.

What does the evidence ladder mean in practice?

It means not giving every sales statement equal weight. The four levels are a dated public product or pricing record, a written response tied to the buyer's requirement, a buyer-controlled scripted demonstration and a time-limited test with protected data, acceptance thresholds and a rollback route.

How should a buyer compare costs across suppliers?

Normalise each proposal over the same period and scenario. Record currency, tax, billing frequency, minimum term, renewal, users, contacts, sends, interactions, environments, API calls, credits, support and expected overage. Calculate current, credible and stress volumes, and keep one-off discovery, migration, training and exit work visible.

Why is there no ranking?

A ranking would need tested performance and comparable commercial terms. Neither appears on the public pages reviewed here, so the five entries run alphabetically.