Martech that earns its licence fee: stack design, automation and AI-led operations
Most martech underperformance is a design problem, not a product problem. Start from the ten use cases that would change commercial outcomes, work back to the data and capabilities they require, then buy the smallest stack that delivers them. Adoption is governed by process ownership and data quality, not by feature count.
What is a martech stack?
A martech stack is the set of connected technologies a marketing organisation uses to collect data, manage content, execute and automate campaigns, and measure outcomes — typically spanning a CRM or customer data platform, automation and messaging tools, content and asset management, advertising integrations, and analytics.
The use-case-first method
Stack conversations usually begin with categories — do we need a CDP, should we move automation platforms — and that framing guarantees an expensive answer. The productive starting point is a list of the specific things you want to be able to do, written concretely enough to test.
Recover abandoned high-intent journeys within an hour. Suppress prospecting spend against existing customers across all channels. Route a qualified enquiry to the right salesperson within five minutes with full context. Personalise the top three pages by segment. Trigger a retention play when usage drops below a threshold. Report channel contribution by cohort without a manual spreadsheet.
Ten of those, prioritised by commercial value and feasibility, will tell you exactly which capabilities and data you need — and usually reveal that half are achievable with tools already licensed.
Data foundations before automation
Automation multiplies whatever is underneath it. Built on inconsistent identity, undefined events and unmanaged consent, it multiplies error at speed and volume — duplicate messaging, wrong-language sends, campaigns to churned customers.
Four foundations are non-negotiable: identity resolution so one person is one record across systems; a documented event and field dictionary so 'active user' and 'qualified lead' mean one thing; consent and preference management captured at source and honoured everywhere; and data quality monitoring with alerting, because silent decay is the norm.
This work is unglamorous and always under-scoped. It is also the difference between a stack that produces compounding advantage and one that produces incidents.
Choosing between platform consolidation and best-of-breed
Suite consolidation buys integration and reduces vendor management; it costs flexibility and often capability depth. Best-of-breed buys capability; it costs integration effort and a permanent data engineering burden.
The decision should follow team capacity. Organisations without dedicated marketing operations and engineering support are almost always better served by consolidation, because the theoretical advantage of best-of-breed is only realised by people with time to maintain the connections. Teams with genuine operations capability can extract more from specialised tools.
Where AI actually pays now
The reliable returns from AI in marketing operations are currently in production and monitoring rather than judgement.
What still requires human ownership: the strategic choice, the claim being made, the offer, and anything with legal, safety or reputational consequence. The governance question to answer before deployment is simple — who is accountable for what the system publishes.
Adoption is a governance problem
Shelfware is created by unassigned ownership. Every tool in the stack should have a named owner, a documented process it supports, a small set of trained users, and one metric proving use. Anything that fails those tests at annual review is either fixed or removed.
A quarterly stack review — licences, actual usage, overlaps, integration health, cost per active use case — routinely finds fifteen to thirty percent of spend recoverable. That recovered budget usually funds the data foundation work the stack needed in the first place.
Vendor-led stack vs. use-case-led stack
Frequently asked questions
Do we need a customer data platform?
Only if you have multiple meaningful data sources, a real need for unified profiles across channels, and the operational capacity to maintain them. Many organisations get most of the benefit from a well-configured CRM plus a warehouse and disciplined event tracking.
How long does a martech implementation take?
Plan in phases: four to eight weeks for use-case definition and data foundation design, then six to twelve weeks per capability release. Programmes attempting full-stack transformation in a single release routinely overrun and lose sponsor confidence.
How do we stop buying tools we do not use?
Require a written use case, a named owner and a success metric before purchase, and run a quarterly utilisation review with removal authority. Procurement discipline solves more martech problems than platform selection.
Should AI-generated content be published without review?
Not where accuracy, legal exposure or brand voice matter. A practical standard is machine drafting with human editorial accountability, and a published record of who approved what.
A short diagnostic conversation is usually enough to tell you whether there is a real opportunity here — and what it would take.