Choose creative advertising software with a task specimen, not a sales demo

Choose against a real production job, not a feature count. The useful question is whether the whole team can move an approved concept into required placements without losing evidence, rights information or control. A polished demonstration can show a function while hiding the work needed around it.

What to take away

Write a task specimen

Take one representative campaign and remove client identifiers. Keep the awkward parts: the longest approved headline, a late price change, several aspect ratios, one restricted image licence, two approval roles and a withdrawal request. State which data must never enter a trial account.

List the outputs and decisions the tool must support. Separate them into essential, desirable and excluded. An essential item has a test and an owner. "Easy collaboration" is too vague; "a reviewer can reject one variant without changing its approved siblings" can be observed.

The Government Service Standard asks public teams to understand total ownership cost, future change, AI and effects on users when they choose tools and technology.

Collect evidence before the sales call

Build one sheet with these fields: requirement, vendor claim, source URL, date, proposed test, result, limitation and decision. Record whether the evidence is a public product page, contract term, security document, live demonstration or hands-on observation. These sources do not carry equal weight.

One worked row shows the shape. The campaign in it is a sample, not a client file.

Exclude candidates that fail a non-negotiable condition on paper. Examples: an unacceptable licence for uploaded work, no usable export route, an unsupported placement or a data path the buyer cannot approve.

Candidates usually fall into three categories. A digital asset management (DAM) system holds the masters, the rights information and the storage. A creative automation tool resizes approved art into named placements. An ad-serving platform covers delivery, pacing and reporting. Give each category its own essential test before you compare products.

The current GOV.UK guidance on commercial off-the-shelf products tells government teams to understand the problem and try options before committing.

Run a bounded trial

Use the same specimen, participants and clock for each surviving option. Observe setup, template construction, variant changes, review, permissions, export and recovery from a deliberate mistake. Do not import personal data, confidential client files or live credentials merely to make the exercise realistic.

Keep the trial short. Two to three weeks with five or six participants is enough. The group needs two people who build assets, one reviewer, one administrator and the person who owns rights decisions.

Capture time and failure points without declaring a universal productivity result. A trial with one team answers whether that team could complete those tasks under those conditions. It does not establish performance for every English advertiser.

Price the operating model

Ask for a comparable written basis covering seats, usage, storage, serving, integrations, services, support and exit. Add internal administration, training, migration and review effort. If suppliers meter different units, present the mismatch instead of manufacturing one total.

The bill usually tracks three things. Seat count comes first: a team of ten pays for ten seats even when four people make assets. Storage comes second, because every master, resize and review copy is retained.

Migration is the variable. It is often priced per file, per day or per user profile rather than per seat. Ask for each unit price before you ask for a total.

Make judgement visible

Score the observed task before discussing preference. Use a simple scale: completed without help, completed with an acceptable intervention, unresolved, or failed an essential condition. Keep notes beside the score so another reviewer can see why a polished output still failed, perhaps because the source licence vanished or the approval trail could not be exported.

Ask makers and reviewers separately about friction. A fast authoring route may push extra work onto legal review, media operations or administrators. Record that transfer rather than averaging everyone into a satisfaction number. Where participants disagree, retain both accounts and decide which risk owner has authority.

Revisit the original exclusions at the end of the exercise. If the demonstration depended on a function, integration or asset source outside the proposed agreement, the observed route is not yet the purchased route. Request written clarification or repeat the task inside the actual boundary.

End with four outcomes: reject, clarify, pilot or negotiate. A pilot needs a named decision date, success conditions and a safe exit. Selection is complete only when the evidence sheet explains why the chosen service beat the alternatives on the buyer's own work.

Before you act

Common questions

Why is a sales demonstration not enough to judge advertising software?

A demo runs in the vendor's conditions, not in yours. The team's own production job is the only test that shows the surrounding work.

What should a task specimen include?

One representative campaign, with client identifiers removed and the awkward parts left in as listed above. State which data must never enter a trial account.

How should a buyer price the operating model?

Start from the written basis described above. Then add internal administration, training, migration and review effort, and compare like for like.