What challenges should agencies expect when adopting AI tools?
The main challenge with AI adoption in agencies is rarely access to tools. Most teams can already use ChatGPT, Claude, Gemini, Perplexity, Midjourney, or the AI features inside the software they use every day.
The harder work is trying to turn scattered staff use (a symptom of shadow AI) into a managed agency AI workflow.
Agencies run on delivery systems. Work moves from briefs into research, strategy, production, review, approval, reporting, and account management. AI tools can touch each stage, which means they can also create risk at each stage.
McKinsey’s 2025 global survey found that 78% of respondents said their organisations used AI in at least one business function, while 71% said their organisations regularly used generative AI in at least one function.
At Column, AI now touches research, article planning, source review, draft support, quality assurance, and workflow automation across parts of our work. The lesson has been that AI can increase production speed before it improves delivery control.
Individual AI use often comes before shared process
Most agency AI adoption starts with individual staff use. A strategist uses AI to build an outline, the content lead uses it to compare search results, the account manager uses it to summarise calls, and the designer uses it to produce moodboards. This all happens before everyone agrees on what counts as acceptable use.
That bottom-up pattern helps everyone learn, but it also creates uneven work. One person may check every claim, while another may paste sensitive client data into a tool without checking the settings, and a third accepts polished output because it sounds credible.
IAB’s 2025 State of Data report found that only 30% of agencies, brands, and publishers had fully integrated AI across the media campaign lifecycle. The same report said 50% of the industry lacked a strategic roadmap for AI.
That creates a basic management problem. Agency leaders need to decide who can use which tool, for which task, with which data, and under which review process.
More AI output creates more review work
AI tools increase draft volume. That can help your agency move faster, but it also increases the amount of work that needs human review.
Muck Rack’s 2026 State of AI in PR report found that 76% of PR professionals use generative AI in their workflow. Among users, 86% use it for editing and refinement, 74% use it for writing and content creation, and 68% use it for strategy and planning.
Those numbers sound about right. We know AI makes it easier to produce outlines, first drafts, social posts, media lists, variations, reports, and campaign ideas.
But higher volume can also expose weak review habits, with unsupported claims, off-brand wording, weak examples, and false confidence in the mix.
We’re already seeing this risk in advertising. IAB reported that 70% of marketers had experienced at least one AI incident, including hallucinated outputs, biased or inappropriate content, off-brand material, or regulatory failures. It also found that 40% had paused or pulled ads after AI-related issues.
Agencies need review systems that match the speed of production. That means source checks, named reviewers, brand rules, claim validation, client approval points, and a documented human review stage.
Client data becomes harder to control
Your agency may work with sensitive client material, including sales data, customer insights, product roadmaps, campaign results, user stories, contracts, budgets, and internal strategy documents.
AI tools make it easier to process all that material, but also make it easier to miss leakage.
IAB found that nearly two-thirds of industry respondents cited data quality, data protection, and fragmented tools as top barriers to AI integration.
That’s agency life in a nutshell, because many teams already work across project management software, analytics tools, creative tools, shared drives, chat platforms, dashboards, and client portals. AI adds another layer.
This means you need rules for approved tools, banned use cases, client consent, data retention, account settings, and audit trails—especially if you’re operating at scale.
The OECD AI Principles give agencies a useful baseline through transparency, privacy, human oversight, robustness, safety, and accountability.
In practice, those ideas become operating questions. An agency needs to know what data entered the system, what output came back, who checked it, and what client decision followed.
Pricing pressure from AI arrives early
Once clients know your agency uses AI tools, some will expect lower fees, faster delivery, or more output for the same budget. A fair expectation for some tasks, but not all.
For example, AI can reduce time spent on summaries, first drafts, keyword clustering, transcript analysis, reporting templates, and content variants.
If you were previously billing for manual podcast episode transcriptions at £1,000/mo, that number should drop significantly.
But increased efficiency from AI may not reduce time spent on positioning, argument, expert interviews, source verification, legal review, client management, or senior judgment.
The show host still needs to talk to guests, and using AI to generate questions doesn’t negate that part of the job.
If your agency sells hours, AI weakens your fee narrative. If you sell outcomes, AI becomes part of the delivery method.
You’ll need to decide which efficiencies benefit the client, which ones protect margin, and which ones should improve the work.
At Column, AI supports research handling, outline development, quality checks, and process design. The client value still comes from judgment and taste.
That judgment covers what to include, what to remove, which source to trust, which claim needs more work, and how the output supports a commercial goal.
Junior staff need a different training path
AI changes how junior agency staff learn. Traditional agency training often starts with lower-risk tasks such as research notes, first drafts, meeting summaries, social copy, reporting, and desk research. These days, AI can do those things quickly.
If juniors skip that work too early, they miss the repetition that builds judgment. But if you block AI use, you might slow down people who need to learn how modern delivery now works.
The better path is supervised use. Junior staff should learn how to prompt, compare outputs, check sources, protect client data, spot weak reasoning, and explain why they changed an AI draft.
Muck Rack’s report shows that companies are beginning to formalise this work. 51% of PR professionals now say their workplace has an AI use case policy, and 43% say their workplace offers AI training.
AI tool sprawl can hide weak process
Many agencies now have overlapping AI tools inside search, writing, design, automation, analytics, CRM, reporting, and meeting software. The danger is tool sprawl, where teams add AI features without deciding how they fit into delivery.
This creates duplicated work. It also makes measurement harder. You might know your staff uses AI every day, but not know whether that improves margin, cycle time, client satisfaction, retention, or quality.
The measurement questions should be operational: did the work move faster? Did review time rise or fall? Did source errors decrease? Did client revisions decrease? Did the team protect margin? Did junior staff improve?
What agencies should expect when adopting AI
Agencies adopting AI tools should expect speed gains first and management problems second.
The strongest early use cases will usually include research support, summarisation, ideation, reporting, workflow automation, and draft support.
The harder problems will come from review, data protection, pricing, client trust, staff training, and measurement.
Agency AI adoption works best when leaders treat it as a delivery change rather than a software rollout.
The practical test is whether the agency can explain who used which tool, on which task, with which data, under which review process, and with what effect on margin, speed, and quality.
If your agency is already using AI tools but still relies on scattered prompts, uneven review, and unclear data rules, my free AI audit will help you see where the risks sit and where the strongest workflow gains are likely to come from. You can start here.