(01)Horizon Research & Consulting · Publication No. 002
WhatActually Works
2026 Edition
Almost every agency now uses AI somewhere in production. Very few can say what it has actually done to their output, their delivery time, or their margin. This report is about the line that decides it.
(02)Why it exists
The wrong question.
The conversation about AI in agencies has been stuck for two years, because it has been asking the wrong question. Whether you use AI is settled and uninteresting — adoption is close to universal, with or without a policy. The useful question is narrower and much harder: where does it actually pay, and where is it costing you without ever appearing on an invoice?
That question is hard because the evidence is genuinely uneven. A great deal of what circulates as industry data about AI in marketing traces back to content-marketing blogs republishing one another with no stated methodology. We excluded all of it. This report rests on a small number of controlled studies and one methodologically transparent industry report — each named, on a Sources page, so every figure can be checked.
(03)The evidence
Every figure traced.
No adoption percentages recycled from content-marketing blogs. Four named primary sources, and a Sources page so you can check each one.
758
Consultants in a randomised controlled trial
Organization Science, 2026
25.1%
Faster, inside AI's capability — at 40%+ higher quality
Dell'Acqua et al.
19pt
More likely to reach the wrong answer, just outside it
Same tool, same people
95%
Of enterprise generative-AI pilots produced no measurable return
MIT Project NANDA, 2025
(04)What's inside
Inside the report.
The jagged frontier, and why the line is invisible
Where AI reliably earns its place
Where it quietly costs you
The homogenisation risk specific to agencies
A four-question test for any task
What disciplined adoption looks like
What we still don't know
(05)Who it's for
Who it's for.
Written for owners and senior leaders of independent marketing and creative agencies — the people who decide where AI is allowed to operate and who carry the cost when that decision is made badly. It assumes no technical background. The findings apply to any professional services firm; one section, on homogenisation, applies to creative businesses and almost nobody else.
Stop asking whether your agency uses AI, and start asking which side of the line each task sits on — because the tasks that sit just outside it are the ones costing you money you cannot see.
(06)Research Report
Read it this week.
The report is free. We ask for an email because we send it as an attachment — and because knowing who's reading tells us what to research next.
Get the report
Enter your email and we'll send the full PDF straight over. No sequence, no sales follow-up — the report, and nothing you didn't ask for.
Where does your agency stand?
This report describes the terrain. It can’t tell you your position on it. The AI Readiness Assessment returns the stage your agency has actually reached across five areas, and your weakest link — in about three minutes.
Weighing a specific decision?
If you’re facing a concrete AI question and want the same structured, evidence-based thinking applied to it directly, we’d be glad to talk it through.