A procurement process for AI marketing tools

What to take away

A marketing manager at a Leeds professional services firm has £60,000 pencilled in for AI marketing tools in the 2027 budget, three shortlisted suppliers and no agreed way to tell them apart. That position is normal. The fix is process, not more demos.

Why the buyer list keeps growing

Marketing teams now pick from three families of supplier. Large software vendors sell general assistants with a marketing wrapper. Existing marketing suites sell generation features bolted onto email, a CMS or analytics. Specialist tools do one job, such as brief generation or asset tagging.

The categories overlap, which is why longlists run to twenty names. A customer data platform adds a copy assistant. An email platform adds send-time optimisation. Buyers then compare products that were never built to compete with each other.

Starting from a search for AI marketing tools produces a list shaped by vendor marketing spend. Starting from your own task list produces a list shaped by your work. Write down every recurring marketing task, then mark which ones a model could draft, rank or check.

Group those tasks by the data they touch. Tasks using published content and brand assets carry low risk. Tasks using customer records, purchase history or special category data carry more. That grouping tells you which suppliers need a full review and which can be trialled in a sandbox.

Watch for AI washing. A product that adds a template library and calls itself an AI platform is not the same purchase as one that runs a model over your own data. Ask what the model does that a rule could not.

Headcount matters too. A tool needing a prompt engineer, a data analyst and a part-time administrator is a different purchase from one a campaign manager can run alone. Count people hours before you count licences. Choosing between AI marketing tools and providers is a procurement task before it is a technology task.

Choosing a buying route

The route you take sets the price shape and the support burden. Most buyers end up in one of four positions.

The right starting point depends on your data, channel mix and engineering resource. A team with clean data and one clear problem does well with a point tool. A team running five channels usually wants a suite. A team with engineering resource may prefer to extend a platform it already runs.

Before you build a longlist, it helps to know what each category does well. Our round-up of the best AI marketing tools in England explains where each type of product tends to fit.

Prices differ in shape as much as in size. Point tools price per seat, while suites price per contact record or per send volume. Platform extensions price on consumption, and managed services price on retainer plus a share of media spend. Comparing a seat price with a consumption price without a volume assumption tells you nothing.

Run a paid pilot before a full contract. A 60-day pilot with ten users costs a fraction of an annual licence and exposes integration problems early. Set the success measure before the pilot starts, and write down what happens if it fails.

What the procurement brief covers

A brief of two pages beats a fifty-slide request for information. It should state the task, the current cost of doing it by hand, the data involved and the success measure. Suppliers who cannot answer the brief do not reach the demo stage.

Set out the commercial shape you want. Say whether you need monthly terms, a pilot licence or a break clause at 90 days. Say who owns the configuration and what happens to your prompts and datasets when the contract ends.

Include a like-for-like table so responses can be read side by side. Our guide to AI marketing tools supplier comparison in England shows the columns that reveal real differences, including volume tiers and overage rates.

Add an exit test. Ask each supplier to export your data and configuration into a readable format and to document the process. A supplier that cannot describe an exit rarely describes an implementation well either.

Ask for two references from organisations of a similar size in a similar sector. A reference from a global brand tells you little about support at your scale. Ask the referee what broke and how quickly it was fixed.

Keep the brief short enough that a busy marketing lead will read it. Three pages of requirements beat thirty pages nobody opens.

Sorting out data roles

Under the UK GDPR, a marketing team using an AI tool is normally the controller and the vendor is normally the processor. Some vendors argue for joint controllership, which changes your liability and your response duties. The ICO's guidance on controllers and processors explains how that division works and what a written contract needs to cover.

Ask one question in every sales call: does our input train your model? The answer belongs in the contract, not in a follow-up email. If the answer is yes by default, ask for the opt-out in writing.

Our checklist for AI marketing tools vendor due diligence in England covers the security questionnaire items that matter most, including sub-processors, retention periods and training data use.

Sub-processors are where most surprises live. A vendor may rely on a hosting provider, a model provider and an annotation supplier. Each one needs to appear on the list you approve, and you need notice when that list changes.

Special category data needs more care still. Health, ethnicity and political opinions rarely belong in a marketing model, and profiling with them needs a lawful basis and an extra assessment. Most teams are better off excluding those fields at source.

Public sector buying rules

Councils, NHS bodies, universities and arm's length organisations buy marketing tools under rules private firms do not face. They need a documented risk assessment, a lawful basis for processing and a route for people affected by automated decisions to ask questions.

The guide to using artificial intelligence in the public sector sets out the governance and procurement questions to work through before a tool touches citizen data.

Private buyers can borrow the same discipline. If a supplier cannot explain how a decision was reached about a customer, that gap will surface in a complaint or a subject access request.

Public sector teams also face transparency duties. Publishing a short record of which tools you use, what data they process and who owns the decision is cheap to produce and useful when a resident asks.

Framework agreements can shorten the work. Many public bodies buy through an existing framework rather than running a full tender, which fixes the terms in advance. Check whether the framework covers AI features or only the underlying software.

Media buying and programmatic

For paid media, the AI question is older and better documented. Bid management, audience building and creative rotation have been automated for years. The Back to Basics Guide to Programmatic from IAB UK explains how inventory is bought and where automation sits in the chain.

Use that knowledge in supplier meetings. If a media tool claims a performance uplift, ask which part of the chain it changes: data, bidding, creative or measurement. A claim that names no mechanism is marketing copy.

Separate platform native tools from third party tools. Features inside an ad platform are usually covered by your media spend. A third party tool adds a fee, so any uplift has to clear that fee before it earns anything.

Trading desks and agencies add a second layer. Ask who owns the optimisation settings, and whether you keep them when the contract ends.

Measurement is the hard part. Incrementality testing, holdout groups and clean rooms all claim to show what the automation added. Agree the measurement method before the campaign starts rather than after the invoice arrives.

Scorecards and templates

Build a scorecard before you meet suppliers, and weight it to your own priorities. A simple version scores data handling, integration effort, reporting, support and total cost across five columns. Weights might put data handling at 30 per cent and price at 15.

The DMA publishes toolkits and templates for marketing teams, which can be adapted for briefs, scorecards and records of processing.

Do not score the same capability twice. If two suppliers both generate on-brand copy, the tie-break is integration, support and cost, not another feature demo.

Score evidence, not enthusiasm. A recorded demo in your own tenant, with your own data, tells you more than a rehearsed walkthrough on a vendor laptop.

Weight the scorecard before you see any supplier. Weights set afterwards drift towards whichever vendor gave the best demo. A spreadsheet signed off by the budget holder is enough.

Build, buy or extend

Extending a platform is the route buyers forget. If your organisation already runs on a cloud provider, you may be able to build a narrow workflow with a managed model service and keep data inside your own tenant. Microsoft's Foundry documentation describes the tools for building and operating AI applications on that basis.

This route suits teams with an engineer or analyst who can own the workflow. It rarely suits a team buying its first AI tool, because the hidden cost is maintenance rather than the first build.

Total cost of ownership favours buying for most marketing teams. Building means you own the model choice, the prompts and the monitoring, plus the on-call rota. Buying means a vendor owns those, and you accept their roadmap.

Whichever route you choose, the delivery work looks similar. Our guide to AI marketing tool implementation in England covers the migration, testing and adoption steps that decide whether a licence turns into daily use.

Worked example with pound figures

A 40-person marketing team at a Leeds professional services firm sets a first-year budget of £60,000. Every figure below is illustrative.

The licence is £24,000, so 40 per cent of year one. Year two falls to £28,000 if seat numbers stay flat and support costs £4,000. Across three years the illustrative total is £116,000.

A narrower point tool changes the arithmetic. Two seats at £150 a seat per month is £3,600 a year, plus £5,000 of setup, so an illustrative first year of £8,600. The saving is real, but so is the limit on how many workflows it covers.

The comparison that matters is against the manual cost, not against another licence. If the team spends 300 hours a month on work the tool now drafts, the case rests on redeployed time rather than headcount cuts.

Year two is where programmes slip. The pilot owner moves roles and nobody owns the workflow. Our operations and delivery guide for 2027 sets out the ownership, review and reporting rhythm that keeps a tool in use.

Common questions

How many AI marketing tools should a mid-sized team run?

Two or three at most in the first year. Each tool adds a contract, a data flow and a training burden. Add a fourth only when an existing tool clearly cannot cover the task.

Do we need a written contract with every AI supplier?

Yes, where personal data is involved. A processor agreement is required under the UK GDPR and should name sub-processors and state retention and deletion terms.

What separates a suite from a point tool?

A suite covers several workflows under one contract and shares data between them. A point tool solves one task, usually deploys faster and is simpler to exit.

How do we measure whether a tool worked?

Choose one measure before the pilot and hold it for a quarter. Time saved per campaign, brief turnaround and asset revision count are easier to defend than a general productivity claim.