Should You Still Work With Marketing Agencies?

Here's the deal: I've worked both client-side and agency-side. Both have their pros and cons, but we're not here to talk about that. This is what I want to talk about...

Software has never been cheaper or faster. A blog post, a set of social graphics or a basic ad campaign now takes minutes rather than days. Yet for many growing businesses, predictable customer acquisition feels harder than it did three years ago.

The traditional agency pitch was simple: "we have the people, tools and execution capability you don't." That premise has weakened. The tools are now a monthly subscription, and most leadership teams have tried them. So at renewal time, many Managing Directors ask a reasonable question: why are we still paying a retainer set before AI existed?

It is the wrong question to stop at. The more useful one is whether your agency is selling commoditised execution or strategic growth orchestration. This article sets out what the evidence says about bringing marketing in-house with AI, what is happening to the agency model, and how to tell which kind of agency you are paying for.

The in-house illusion: why DIY AI execution stalls

AI adoption among UK businesses is no longer niche. The British Chambers of Commerce and Atos found that 54% of UK firms are now actively using AI, up from 25% in 2024, and around 94% of the firms surveyed were SMEs. Marketing is one of the most common uses. In the Department for Science, Innovation and Technology's AI Adoption Research, 72% of businesses using or planning to use AI named marketing as an area of application.

The same research shows where the value is landing. DSIT surveyed 3,500 UK businesses between February and May 2025, including 700 that were using AI:

75% of AI-using businesses reported improved workforce productivity. 12% reported an increase in revenue. 77% had not yet seen any change in revenue.

Faster output is not the same as more customers. Productivity gains mean a task takes less time. Revenue gains mean a buyer chose you. The DSIT figures suggest most businesses have the first without the second. The "not yet" matters, and some of those firms adopted recently. But it is the gap that in-house AI strategies need to close.

The pattern appears in larger companies too. MIT's NANDA initiative, in its report The GenAI Divide, found that 95% of enterprise generative AI pilots delivered little to no measurable effect on profit and loss. The study drew on 150 interviews, a survey of 350 employees and 300 public deployments. It covers enterprise pilots rather than SME marketing teams, and its sample is modest, so read it as directional rather than definitive.

Outcome	Expected	Experienced
Speed and nimbleness	93%	40%
Cost effectiveness	93%	60%

That study predates the current wave of generative AI, and AI lowers the cost of building an internal team. But the gap between expectation and result is a useful warning. Capability does not appear because a team has access to tools.

The practical problem is differentiation. When every competitor can produce competent copy and polished graphics in seconds, competent stops being an advantage. In-house teams using AI tend to produce more content. Whether more content produces more pipeline depends on positioning, distribution, measurement and sales alignment. Those are the parts software does not supply.

---

What is dying: the execution-only agency

If the in-house route is harder than it looks, that does not mean every agency deserves its retainer. The model under the most pressure is the one built on selling hours, volume and activity.

Signs your agency is in the old model

• Reports list activity: posts published, hours logged, keywords ranked.

• The scope and price have not changed since before generative AI.

• Production tasks that once took a day are billed at the same rate.

• Nobody can show how their work connects to revenue.

  
  

Hourly billing shows the problem most clearly. Suppose a technical build once took eight hours and AI-assisted review now takes two. Under time-based billing, the agency's revenue drops by 75% for doing the job better, while a client who knows the work has become faster feels overcharged. Efficiency becomes a penalty, and the incentive shifts toward bloat. The market is moving away from the model. World Federation of Advertisers research, reported by Marketing-Interactive, shows labour-based remuneration falling from 54% of agency compensation in 2011 to 17% today. That data covers large multinational advertisers, but the direction applies to smaller buyers too.

The industry numbers are consistent with that shift. Moore Kingston Smith's Marketing Services Outlook Survey of UK agency leaders found average expected revenue growth of 5.8% for 2026, down from the 8.1% predicted a year earlier, with staff numbers expected to fall by 0.9%. Some 27% of agencies expected revenue to decline. eMarketer's analysis shows that since 2023 media ad spend growth has exceeded agency holding company revenue growth. In other words, money is flowing into advertising, but not into the traditional agency structure.

The large networks show the divergence. WPP reported a 6.7% revenue decline in Q1 2026 and, according to the Financial Times, has lost around 11,000 jobs since the start of 2025 with up to 1,000 more to go. Publicis, by contrast, reported 5.6% organic growth for 2025 and guided to 4-5% for 2026, after pivoting toward technology, data and AI services. Havas grew organically by 3.1% while its headcount fell 1.2%. The pattern is the same: growth without adding people, for firms that changed what they sell.

Production roles are being automated first. Sunup research, cited by eMarketer, found that 57% of agencies have slowed or paused entry-level hiring. US Census Bureau data cited by Adweek shows workers aged 20 to 24 made up 6.5% of US advertising and PR jobs in the latest year reported, down from 10.5% in 2019.

Search reporting is changing as well. Monthly "blue link" ranking reports assume that position on the results page translates into clicks. That assumption is weakening. Ahrefs found that AI Overviews correlate with a 58% lower click-through rate for the top-ranking page, based on 300,000 keywords. Seer Interactive's study of 3,119 informational queries across 42 organisations found organic click-through rate fell 61%, from 1.76% to 0.61%, where an AI Overview appeared. A report that tells you where you rank, without showing whether anyone clicked, is measuring the wrong thing.

What is thriving: the strategic orchestrator

The same forces that devalue production raise the value of work that software cannot do alone. Three capabilities stand out.

1. Systems integration and martech architecture. Most growing businesses own a CRM, an email platform, an analytics suite, an ad account and a handful of AI tools, often connected badly or not at all. The commercial return comes from making those systems share data reliably. That is an engineering and process problem, and it sits across departments, which is where internal teams often struggle to see the whole picture.

2. Answer Engine Optimisation (AEO). Buyers increasingly get answers from ChatGPT, Perplexity and Google's AI Overviews before they ever reach a website. Optimising for those environments means being the source that gets cited, which draws on clear entity information, authoritative content, structured data and third-party mentions. The Seer Interactive study found that brands cited within AI Overviews earned about 35% more organic clicks than uncited brands, although the researchers noted they cannot prove citation causes the difference.

The traffic that does arrive from AI platforms may be unusually valuable. In a Seer Interactive analysis of one client's data, ChatGPT referrals converted at 15.9% against 1.76% for Google organic. The authors were explicit about the limits: it was a single client, ChatGPT accounted for roughly 11,000 sessions against almost 14 million from Google organic, and the explanation offered was a hypothesis. Treat it as a signal worth testing, not a benchmark.

No agency can guarantee AI citations. A credible AEO practice tests how your brand appears across the major answer engines, identifies gaps, changes content and structure, and re-tests. If an agency cannot explain its method, it is probably repackaging old SEO.

3. Closed-loop attribution. Impressions and clicks are easy to report and hard to take to a board. Closed-loop attribution connects campaign touchpoints to deals and revenue inside your CRM, so you can see which activity produced customers. It takes technical setup, clean data and agreement between marketing and sales on definitions. Few in-house teams have the capacity to build it alongside day-to-day delivery, and it is the clearest line between an agency selling activity and one selling outcomes.

These capabilities matter more as business purchases involve several stakeholders and long consideration periods. Reaching a group of people over weeks, across channels, needs coordinated orchestration rather than a steady supply of posts.

The decision: change how you buy

Most businesses should not abandon agencies. They should change what they buy.

A hybrid model fits the evidence. Lean internal teams hold what only the business can supply, and a specialist partner takes the high-complexity work:

Keep in-house	Bring in a specialist
Brand voice and culture	Positioning diagnosis and growth strategy
Domain and customer knowledge	Martech integration and data architecture
Day-to-day content production, with AI support	AEO methodology and testing
Sales and customer relationships	Closed-loop attribution and reporting

Under this model the agency is judged on outcomes, and pricing follows. Where AI has reduced the effort behind a task, the saving should reach you through faster turnaround, wider scope or lower cost.

Disclosure: Snap Media Digital is a marketing and branding consultancy. We sell strategy and execution, so weigh our view accordingly. The sources above are linked so you can check the figures yourself.