👋 Greetings earthlings,

Emma here. This time last year, we shared a guide to AI use cases across the creative workflow - from research and strategy to production and client services. It identified the key integration points and recommended which tools to try. It went down a storm. 

I started working on updating this for you for 2026, but quickly realised that the conversation has shifted and its now too simplistic. Most agencies aren't asking "where could we use AI?" anymore like last year. Instead, they're asking "why isn't it sticking?" The tools are there and the licenses are paid for. But what most agencies are experiencing now that that too often experimentation never becomes operation.

So this year's guide is different. Rather than focusing on what AI could do, I’m highlighting what is actually working - what we are seeing across agencies we talk to and work with. I’ll share 8 proven workflows and strategies people are implementing right now. And for each each I’ll give you a two step roadmap: we see agencies start here, then scale there. Scroll down for a free cheatsheet download too.

Intentionally I’ve kept it simple, I love stuff you could just implement right after reading this. So none of the examples require custom tool building or developer resource. You can execute everything here using your existing AI subscriptions, using Custom GPTs, Gems, Projects, NotebookLM etc.

Before I finish this overly-long introduction (which goes against all my own rules of good newsletter writing!) I wanted to point out that there is one common hallmark we see about the agencies successfully operationalising these types of workflows: they have all invested in skills for their people. They don’t want a few enthusiasts tinkering in corners; AI success is about teams who all understand AI well enough to spot the opportunities, deploy solutions themselves and turn pilots into new workflows. To really innovate in their roles and for the business. That is what turns a subscription into a capability and when you get the ROI.

Right, let's get into it.

• How are agencies actually using AI?

  • 1. The Growth Director

  • 2. The Virtual Stakeholder

  • 3. The Brand Brain

  • 4. Hyper-speed Admin

  • 5. The Brief Interrogator

  • 6. The Knowledge Base

  • 7. Asset Intelligence

  • 8. Cultural Nudges

  • The Bottom Line

• Webinar: AI, Agencies and Value

How are agencies actually using AI?

We've moved past the "wow" phase of AI and into the "how" phase. The most progressive agencies are now operationalising AI and creating a new moat from their competitors. They're building virtual stakeholders, simulating difficult clients, and automating the administrative grind are part of their day-to-day workflows.

Here are eight concrete strategies we see agencies adopt right now. For each, there's a starting point you can try tomorrow, and a more sophisticated version for when you're ready to scale.

Download the cheatsheet:

1. The Growth Director: punching above your weight

Business development is resource-intensive. It can be a struggle to do the deep market analysis required to impress prospects. The solution? Hire AI agents to handle the heavy lifting.

We have helped agencies build AI assistants they're calling "Growth Directors". These are Deep Research agents designed to analyse an ICP, identify prospects and source the right C-suite contacts for outreach. You can equally use it to analyse your existing data rich accounts, combine that with research into industry trends, and give your CS team hooks to grow their accounts.

At Spark, we apply this to our sales cycle. We use all the questions that come up in our sales calls to power a Gem that assists with web copy, brochures, proposal emails, and it even supports the team live during a pitch. Post call, we link our Gem to the transcript to generate first draft emails in our specific brand voice. It is simple, but it guarantees instant follow ups and prevents any detail from being overlooked. 

Why it matters for your business model: The Growth Director assistant example is a revenue play and the sales call support standardises team wide consistency. By automating the capture of critical meeting data, we accelerate follow up cycles and eliminate the reliance on manual memory. It ensures every proposal is more thorough and highly tailored, informed by what was actually said, not what we managed to recall at the end of the day. 

Start here: Build a custom GPT (in ChatGPT) or Gem (in Gemini) or a Project (in Copilot or Claude). The name varies by platform, but the function is the same. Feed it your ICP data, past winning proposals and brand voice guidelines. In the instructions, consider the job you want the assistant to do and how you want the outputs to look. Use it to draft personalised outreach and pressure-test your messaging before it goes out.

Scale to this: Use Copilot Studio or Google Workspace Studio to set up an automation. Have it look in your calendar for sales meetings, and 24 hours before, have it identify the domain name of the external people attending the meeting. Have an agent or Gem look up their website and summarise their company and market positioning, gather trends in their industry, and create a briefing for the meeting, and post it to your email or to your Teams or Google Chat. Save your team fifteen minutes per call - over the week you'll find that's a few hours.

2. The Virtual Stakeholder: stress-testing before it leaves the building

You spend weeks developing a creative route, only for the client to kill it because it conflicts with a priority buried in a quarterly report, or because it misses the procurement manager's specific KPIs. You can mitigate the risk by building synthetic personas to stress test work before it ever goes out. 

Try configuring specialised agents to serve as internal critics. For example, some agencies utilise an “Executive Creative Director” persona, informed by strict brand constraints and previous call transcripts, to identify conceptual weaknesses or likely client objections early in the process. Others have developed "Client Sounding Boards" by integrating a clients website, core values and stated purpose to simulate their perspective. In one case, a synthetic persona successfully identified a shift in client strategy from a previous workshop that the human team had overlooked, helping a presentation land where otherwise it would have missed. Similarly, a business development lead used a persona to refine a cold outreach. The AI told him his note was "too focused on culture" rather than sales metrics. He rewrote it, and it landed.

I personally use a version of this method: a proprietary ICP Gem used to vet new service offerings before committing internal resources. (I also run this newsletter past it once it’s written and it always suggests improvements!).

Why it matters for your business model: Fewer revisions, better pitching, better client confidence.

Start here: Build a persona simulator for your most critical stakeholders. they can be individual clients, or archetypes like a CFO or CMO. Ground the persona by uploading relevant documentations: correspondence, project briefs, annual reports and public statements. Ask it to critique your proposals and identify what resonates and what will frustrate. you don't have to implement its suggestions, but it'll definitely make you think.

Scale to this: Build a panel of synthetic stakeholders that work together. A procurement persona, a sustainability lead, a CMO, and an end-consumer all reviewing the same brief, or same proposal, in sequence, each flagging different concerns. Link them together using a simple workflow in Google Workspace Studio or Copilot Studio, and post the feedback as an email or chat in Teams. Make it a routine step before every deliverable.

3. The Brand Brain: institutional memory on demand

Onboarding new creatives to an account takes weeks. Agencies are solving this by moving from dormant files to interactive “Brand Brains” (incidentally WPP got their first and uses the name WPP Brains inside their proprietary AI platform WPP Open).

Creating dedicated client workspaces, loaded with the brief, the SOW, client documents, any research you’ve done, meeting notesand call transcripts, is a very simple but helpful tool. Teams use these ‘brand brains’ to onboard members instantly, verify facts, and generate proactive ideas that are strictly on - brand.

Others are building TOV coaches where copywriters paste their draft and the AI critiques it against brand guidelines. Crucially, it acts as a coach, explaining why something is off-brand and suggesting improvements rather than just rewriting it, making the process transparent, leaving room for human taste and judgment, and upskilling the writer in the process.

At Spark, we set up a NotebookLM for every client project to keep the team aligned asynchronously. We also use shared team Gems loaded with our strategy, messaging and positioning. 

Why it matters for your business model: This reduces dependency on senior memory holders, speeds up onboarding, and makes brand stewardship tangible. It's a scaling and margin lever.

Start here: Create a project workspace for one key client. Google NotebookLM or Microsoft Notebook are perfect for this as they can handle hundreds of documents, and ground their answers only in the data you've given them. use it to test your work, but also to onboard new staff: "Read these documents and explain the brand's core values to me as if I were a designer joining the accounts today."

Scale to this: Build your brand brain into your content production workflow. When a creative submits a draft, it's automatically checked against brand guidelines, with feedback returned before a human reviewer ever sees it. Integrate it with your DAM so it can pull approved assets and reference past work when generating new concepts.

4. Hyper-speed admin

Fee-earning talent spends too much time on non-billable work, from proposals and pitches to case studies and awards entries. We see agencies turning these burdens into structured AI workflows. For awards, feed an AI (a ChatGPT Project or custom GPT, a Claude Project, a Gemini Gem) the judging criteria for specific awards, winning examples from previous years, and summary of the work you've done, then ask the AI to draft the submission.

Why it matters for your business model: More output with the same team, higher win rates, margin protection.

Start here: assemble templates for your common outputs: pitches, proposals and case studies. Next, pullout five examples of each that you think represent your best work. build a GPT or a Gem trained on those examples that helps you turn your notes into a draft or critique the draft you've made yourself. you'll cut the time it takes to make these things in half.

Try this: try Wispr Flow for dictation - it’s the best, and works across all your apps on your computer and phone (I’m a huge fan!). Brilliant for dropping lots of context into a large language model, or dictating detailed instructions for a prompt.

Scale to this: Build an end-to-end awards workflow. When a project is marked complete in your PM tool, trigger an automation that pulls the brief, results data, and team credits, then drafts submissions against the specific criteria of different awards. Brilliant for the heavy lifting of a first draft.

5. The Brief Interrogator: perfecting the input  

Garbage in, garbage out. Clients send vague briefs via email, leading to wasted creative time and rounds of revisions. We advise agencies using AI as a gatekeeper to ensure clarity before work begins.

In our AI for Client Services programme we show agencies how to build a workflow where, upon receiving a brief via email, AI immediately critiques it and drafts a reply asking for clarification on missing points: budget, timeline, target audience - all the elements of your briefing template. You can then review it then fire it off. You can also use AI to read client briefing documents and instantly generate a list of ten targeted, insightful questions to ask on the kickoff call, and have the call guide sent to your Teams or Google Chat automatically. This ensures the team are incredibly well-prepared and uncovers missing information immediately.

Why it matters for your business model: Better inputs mean fewer revisions, clearer scope, and protected margins. And a happier client services team!

Start here: Create a "Brief Interrogator" GPT loaded with your standard briefing template. Give it a prompt to identify any missing information, sure that you're thinking one step ahead, and draft a polite email asking for these specific details.

Scale to this: Connect your email to an automation. When a brief arrives, AI parses it against your template, scores its completeness, and either routes it to the account team (if complete) or automatically sends a clarification request back to the client. Write a row into your new biz opportunities sheet and log everything to your PM tool so nothing falls through the cracks. Yep, that's right - it's easy to achieve with Google Workspace Studio or Microsoft Copilot Studio without writing a line of code.

6. HR and culture: the safe space knowledge base

People are often afraid to ask "stupid" questions or sensitive HR queries (maternity leave policies, for example), creating friction and anxiety. Why not build private, conversational interfaces based on your internal policies knowledge.

We’ve worked with several agencies creating HR GPTs loaded with the employee handbook and benefits packages. This allows people to ask sensitive questions privately first. Others are using AI to scan CVs against job descriptions to generate personalised interview questions that probe specific skills, or to write job descriptions that reveal blind spots in their hiring process. One of our clients successfully used Deep Research to find candidates for a senior role, saving thousands on recruitment fees.

Another agency leader used a custom GPT to streamline Personal Development Reviews. Essential for developing your people, but time consuming to do well. They fed the AI the behaviours and vision they wanted for the business, then dictated his thoughts on each employee via voice notes. The AI formatted these into consistent, high-quality PDR documents, turning a painstaking task into minutes of work. 

Why it matters for your business model: Reduces management overhead and improves team experience without adding headcount.

Start here: Upload your Employee Handbook and policy documents to a secure, enterprise-grade AI instance. Instruct it: "You are a helpful HR assistant. Answer queries based ONLY on the uploaded documents."

Scale to this: Deploy an internal chatbot via Slack or Teams that staff can query anonymously. Connect it to your HRIS so it can answer personalised questions like "How many holiday days do I have left?" or "What's my pension contribution?" Log anonymised query patterns to identify where your policies need clarification.

7. Asset Intelligence: finding the needle in the haystack

Agencies accumulate massive archives of work, but finding "that logo we did for a bank three years ago" is manually laborious. The solution is natural language search.

We see agencies implementing Digital Asset Management systems with AI natural language search. This removes the reliance on rigid file naming conventions. Instead teams can query the archive for visual concepts, such as "logo with sans serif text" or "pictures of beer bottles" and AI identifies relevant assets instantly.

Moreover, agencies are applying AI to their operational history. By analysing historical project data from platforms like Synergist or Float, they are identifying the specific variables that cause projects to exceed budget or schedule. Having predictive scoping allows for improved accuracy on quoting and better protected profit margins. 

Why it matters for your business model: Faster retrieval, better quoting, reduced waste.

Start here: Export your project history to a CSV and upload it to an LLM with the prompt: "Analyse this data. Which client accounts have the lowest profit margin? Which project types run over budget most often?" If you're using ChatGPT, don't forget to turn on Agent mode for best results.

Scale to this: Implement an AI-enabled DAM with visual search and auto-tagging. Connect it to your PM and finance tools so you can query across systems: "Show me all pharma projects from the last two years that came in under budget and won awards." Use the patterns to inform pricing and resourcing decisions.

8. Cultural nudges: gamifying adoption

The primary barrier to AI adoption is rarely the technology itself, but the "frozen middle." While leadership recognises the strategic necessity, the broader team is often either too consumed by daily deliverables or too skeptical to overall established habits. We work with agencies to solve this with cultural nudges and gamification.

Some have introduced "Gold Star" awards for the best use of AI (who didn’t love a gold star), encouraging team members to share wins publicly. Others have created "Prompt Lab" tools where the team puts rough ideas in and it spits out a highly detailed, technical prompt (including camera angles and lighting) to use in image generation tools. Make sure the leadership team are equally included in these initiatives as everyone else and truly lead by example, get them to share how they are using AI. This builds confidence for junior staff. One agency protects human creativity by instituting a "Golden Hour" of brainstorming before AI is allowed to be switched on, so the original spark is always human. But do what’s right for your unique culture and positioning.

Why it matters for your business model: Culture is the multiplier. It drives capability faster than any AI tool subscription. Building the flywheel in your business (see last week's newsletter) is key to getting this working, and forms a big part of our Leadership programme at Spark.

Start here: Give people an hour a week to build AI assistants and automations to help them with their work. Then add a standing "AI Win of the Week" agenda item to your Monday morning meeting. Ask one person each week to demo a specific task they improved or automated that week. the leadership team are not exempt here. You have to do it too. Give AI proficiency some social currency in your agency, and watch adoption and innovation grow.

Scale to this: Create a backlog of AI assistants that you want to build for the business and delegate them to people in your team to build. this becomes your AI programme plan. Tie AI capability development to career progression so it becomes part of how people grow, not just something they squeeze in.

The bottom line ⬇️

The agencies winning with AI aren't trying to automate their creativity. They're automating the friction that surrounds it. They're building second brains for their brands, simulators for their clients, and safety rails for their juniors.

Start where you are. Pick one strategy, try the entry point tomorrow, then scale when you're ready. Move from playing with AI to running your agency with it.

Building these workflows takes more than knowing what's possible. Through our AI Accelerator, we give agencies the skills to implement systems exactly like these and much more across their teams - client services, strategy, creative and the leadership team. And if you'd rather skip to the chase, we also can build them for you.

👉 Book a chat

Webinar: AI, Agencies and Value

One more thing before I go: my co-founder Jules will be in conversation with Rory Spence, Commercial Director of The Wow Company, this Thursday at 11am. Think AI and agency value. Don’t forget to register. I’ve seen (or might have written…) the discussion topics - it’s going to be a gold mine of information.

👉 Register for your place

See you next time, 

Emma 

Co-founder and CEO of Spark AI

About Spark AI

Spark AI empowers creative and brand leaders turn AI curiosity into confidence through structured training and business transformation.

We have worked with 60+ agencies running AI Fundamentals workshops and AI Accelerator programmes based on our # 1 bestselling book Shift – AI for Agencies.

Trusted by Oxford University Saïd Business School and backed by Innovate UK.

👉 Find out more 

Not a subscriber yet?