6 min read · 1,210 words
Eighteen months ago I published a piece on this site predicting how AI would reshape marketing agility and social media. Re-reading it now is a useful exercise in humility — some of it held up, most of it undersold how fast the tooling layer itself would change, and none of it anticipated how much the actual bottleneck would shift from “can we produce content faster” to “can we tell which content is even worth producing.” This is the honest retrospective, not the prediction.
What the original piece got right
The core thesis — that AI would compress production timelines and force marketing teams to move faster — was directionally correct. What it underestimated was the scale. Content that took a team a week now takes days; localisation across markets that used to require separate agency briefs per market now runs through a single workflow with human review layered on top. That part of the prediction was, if anything, too conservative.
What changed that the original piece didn’t anticipate at all
The martech stack itself has consolidated hard. Tools that were separate line items eighteen months ago — a research tool, a drafting tool, an image tool, a translation tool — are increasingly bundled into single platforms, and the marketing operations question has shifted from “which point solution do we buy” to “which platform do we standardise the whole function on.” Gartner’s ongoing research into marketing technology has tracked this consolidation trend, and its martech research specifically flags stack simplification as a top priority for CMOs navigating budget pressure — fewer tools, deeper adoption, less shelfware.
Personalisation at scale is the second thing that moved further and faster than expected. Segment-of-one messaging that used to be an aspiration slide in a martech pitch deck is now operationally normal for teams with clean first-party data — the constraint has shifted from “can the tooling do this” to “do we actually have the customer data infrastructure to feed it,” which is a data engineering problem wearing a marketing costume. Salesforce’s own marketing research has tracked the same gap: AI tooling for personalisation is broadly available, but marketers consistently cite first-party data quality as the binding constraint on actually using it well.
The bottleneck moved from production to judgment
This is the finding that matters most and the one the original piece completely missed. When content production is cheap and fast, the scarce resource stops being output and becomes the judgment about what’s worth producing at all — what angle a competitor hasn’t taken, what a specific customer segment actually needs to hear, what’s going to read as generic versus genuinely useful. McKinsey’s research on generative AI’s economic impact has made a version of this point at the macro level: the productivity gains concentrate in execution tasks, while the strategic layer above them remains the differentiator between companies that benefit and companies that just produce more noise.
Content Marketing Institute’s ongoing research into AI adoption has tracked a related shift in what marketers report as their top concern — differentiation and originality, not production speed, now top the list of things marketers say they worry about with AI-assisted content. Eighteen months ago the conversation was almost entirely about speed. It has flipped.
This lines up with an argument I’ve made separately about data discipline in marketing more broadly, in data-driven marketing and unlocking insights: tooling was never really the constraint. The constraint has always been whether an organisation is disciplined enough to use the data and output it already has well, and AI just made that gap more visible by removing the excuse that production speed was the bottleneck.
Agility turned out to mean something different than expected
The original piece talked about “marketing agility” mostly in terms of campaign turnaround time — how fast can you get something live. The more useful definition that emerged over eighteen months is agility in testing and killing ideas fast, not just shipping them fast. Faster production means you can run more genuine experiments in the same budget window, which only pays off if the organisation is actually disciplined about reviewing results and killing what doesn’t work — most aren’t, and faster production without that discipline just means more mediocre content published faster.
| Prediction area | What actually happened |
|---|---|
| Content production speed | Faster than predicted — the underestimate, not the overestimate |
| Martech stack shape | Consolidated into fewer, broader platforms rather than more point solutions |
| Personalisation | Operationally normal where data infrastructure supports it; a data problem more than a marketing one |
| The real bottleneck | Shifted entirely — from production capacity to strategic judgment about what’s worth making |
Social media specifically — what changed there
The original piece leaned heavily on social media as the primary AI-marketing battleground. That’s held up less well than the broader thesis — social platforms’ own algorithm changes and the fragmentation of attention across more platforms have mattered more to social performance over the period than AI tooling has, a trend Hootsuite’s annual Social Trends research has tracked consistently as audiences spread thinner across a growing number of platforms. AI changed how content for social gets made; it changed much less about whether that content actually performs, which is still governed by platform mechanics largely outside a brand’s control.
What this means for you
If your team’s AI investment over the next 18 months is still primarily about production speed, that’s chasing a problem that’s mostly already solved. The higher-leverage investment now is in the judgment layer — training people to evaluate AI output critically, building genuine strategic differentiation into briefs before AI ever touches them, and building the review discipline to kill the content that speed alone produces but nobody actually needed. I’ve written more on where that judgment gap shows up specifically in brand work in what AI actually changed in brand work, and what it didn’t.
Frequently asked questions
Did AI’s impact on marketing match predictions from 18 months ago?
Partially. Production speed gains were underestimated, not overestimated. What wasn’t predicted at all was the shift in bottleneck from production capacity to strategic judgment about what content is actually worth making.
What’s the biggest change in the marketing technology stack over the past 18 months?
Consolidation. Point solutions for research, drafting, imagery and translation are increasingly bundled into single platforms, shifting the operational question from “which tool” to “which platform to standardise on.”
Has AI made marketing personalisation better?
The tooling capability is largely there for teams with clean first-party data. The constraint has shifted from technology to data infrastructure — personalisation at scale is now more of a data engineering problem than a marketing one.
Did AI change social media marketing performance, or just production?
Mostly production. Social platforms’ own algorithm shifts and audience fragmentation across more platforms have mattered more to actual performance than AI tooling has over the same period.
What should marketing teams invest in now instead of production speed?
The judgment layer — training people to evaluate AI output critically, building real differentiation into briefs before AI touches them, and the review discipline to kill content that speed produces but nobody needed.
What did you get wrong 18 months ago about how AI would change your own work? I’d genuinely like to compare notes.
