👋 Greetings earthlings,

Emma here. A question that comes up constantly in our conversations with agency leaders: "We're doing loads of AI stuff, so why hasn't anything actually changed?"

New research from Harvard Business Review puts data behind that feeling: there's a widening gap between people using AI and actually embedding it in your organisation.

The researchers call it the "productivity J-curve": an initial dip as organisations adapt, followed by sustained gains once the right investments pay off. Honestly, I think "J-curve" undersells it. For most agencies, it feels more like a plateau that never ends. More AI activity, but nothing fundamentally different.

What the research confirms is what we see every week, and what 149 agency leaders told us in our research with The Industry Club: lots of experimentation, pockets of enthusiasm, but no systematic way to turn individual wins into organisational capability.

Helping agencies climb out of that dip is exactly what we do. This edition is about how to do it.

What's inside:

• Why AI use gets stuck

• Your 90 day AI plan: from experimenting to adoption

• What we are watching and you should too

Why AI use gets stuck

Ask yourself: is your team systematically learning what works when using AI, or just accumulating lots of anecdotes? (If you're not sure, it's probably just anecdotes.)

Most AI programmes fail not because the technology doesn't work, but because the learning never gets captured. Someone figures out a brilliant workflow, keeps it to themselves, then leaves or moves on. The insight disappears with them.

The HBR research points to three factors that separate organisations making real progress from those stuck in the dip:

1. Clear accountability: Someone owns AI progress, not as a side project

2. Regular sharing rhythms: Teams actually talk about what's working

3. Captured learnings: Insights get documented, not just discussed

None of this requires big budgets. All it requires is clear leadership intention.

Try this week: 90-day plan from experimentation to structure

Moving from ambition to real progress requires three things: someone accountable, a rhythm for sharing what works, and a way to capture learnings before they disappear. Here's the agency-specific 90-day framework we use with our clients.

Weeks 1-3: Set the rails

Set a clear goal. 
Get clear on what AI is for in your agency. Efficiency? New services? Better work? Scale? Write it down in a single sentence. Share it with your team. If you're not sure where to start, our AI Archetypes framework from Spark Intelligence #36 can help.

Level the playing field. 
Ensure everyone has foundational AI fluency. Not a "lunch and learn." Proper, agency-specific training to Give everyone the skills to use AI in a sophisticated way. Not redrafting emails, but creating AI Assistants to support real work.

Assign ownership. 
Appoint an AI Lead - one person with protected time and actual authority to make this work. Not a committee. A name on the whiteboard. When everyone is responsible, no one is responsible.

Choose three initiatives.
Pick three familiar workflows where AI could make a measurable difference - perhaps one each for client services, strategy, and creative. Map those workflows step by step. Identify where you can build AI Assistants to support specific tasks. Don’t forget to define what success looks like: higher quality output, faster turnaround, or greater consistency.

Deliverable by end of Week 3: One-page AI charter with your goal, your AI Lead's name, and three initiative briefs with success metrics.

Weeks 4-6: Build your rhythm

Create a sharing ritual. 
Establish a recurring slot for learning - a 10-minute segment in the Monday meeting, Friday lunch-and-learns, or a dedicated Slack channel. Keep it brief and specific: a clever prompt technique, a new feature someone's discovered, an unexpected application that saved hours. A client win. This simple ritual signals that continuous learning is part of your culture.

Build and test. 
Develop one AI Assistant for each of your three workflows. Give it to someone in the relevant team to build it. If they build it, they own it and they'll use it. Then get them to share it at the next team meeting. That's how you get the flywheel going.

Measure what matters.
Track the metrics you defined in Week 1. Time saved. Quality improvements. Client feedback. If you said you'd reduce briefing time by 30%, start counting.

Deliverable by end of Week 6: Three working AI Assistants in active use, with initial metrics captured and at least one documented "what we learned" from each workflow.

Weeks 7-9: Scale what works

Turn experiments into SOPs.
Once you've proven success in one area, you've built something far more valuable than an efficient process - you've created a template for change. Write up your successful workflows so a new joiner could follow them on day one.

Package the client story. 
Start thinking about how you'll talk to clients about AI. Not as a cost-cutting measure - as a capability that makes your work better. Prepare your team to answer the inevitable question: "How are you using AI?" What's your manifesto? Make sure your client services team are confident talking about it.

Review your governance. 
Check your AI policies. Review your contracts with clients, freelancers, and partners. Can you confidently answer compliance questions? How are you using client data? Are your AI tools on enterprise-grade plans? Consider a one-page "Responsible AI" statement ready for your website and proposals, and internal FAQs.

Deliverable by end of Week 9: At least one documented SOP ready for wider rollout, draft client-facing AI positioning, and an updated AI policy reviewed by leadership.

Weeks 10-12: Commercialise

Update proposals. 
Reframe how you present value. Lead with how you solve your client’s business problem, not “we build websites”, or “we're a communications agency.” Show your "with AI" edge - the additional creative routes explored, the deeper research, the faster iteration.

Tier your value.
Consider how AI changes what you can offer. Will it let you deliver projects more quickly at lower cost? Or explore more routes and conduct more detailed research? Does it allow you to sell systems that create outputs rather than the outputs themselves? Think about outcome-based or subscription pricing for appropriate work.

Publish your approach.
Publish a "Responsible AI" page on your website. A clear articulation of how you use AI, what guardrails you have, and why it makes your work better. In our experience, agencies that openly discuss their AI approach build significantly stronger client trust.

Deliverable by end of Week 12: Updated proposal templates with AI positioning, pricing options reviewed, and public-facing AI statement published.

The difference between busy and better

I'd put bets on the agencies that really capitalise on AI over the next couple of years not being the ones with the biggest budgets. They'll be the teams that deliberately turn experimentation into structure. We've seen this with production partners like Tuncarp and POD LDN who are already reshaping pricing models and delivery workflows but by systematically upskilling teams. Mindset over money.

There's a people benefit too. The HBR research found that initial sceptics often changed their minds within weeks of using a well-designed system for embedding AI in the business. We've seen this in our own clients too. Some of the people who are least interested in AI at the start of our program end up becoming the power users, exploring everything that they can do with it and completely changing the way that they work.

But that only happens when there's a real system in place. Not just enthusiasm. It’s time to build your system.

Spark AI Accelerator

This is exactly what our AI Accelerator programme is designed to do: take you from scattered experimentation to structured capability across your teams in 90 days. If you want to be in better shape by summer, now's the time to start.

👉 Book a call with me

What we are watching

WPP launches Agent Hub

WPP has released Agent Hub on their WPP Open platform, used by 75,000 employees and clients including Coca-Cola and Nestlé. This follows their Open Pro launch last October, which let brands bypass agencies entirely for certain work. Agent Hub goes further - codifying institutional knowledge into deployable AI agents.

The interesting question isn't whether this is just marketing. It's whether codified thinking can ever capture what actually makes an agency good: the judgment, the taste, the pattern recognition that comes from experience. I'm sceptical. But it does certainly diversity their business model. What repeatable thinking could you encode? What methodology do you do better than anyone else?

Read more

Anthropic launches Cowork

Claude Code has been making waves over the last couple of months as early adopters have found it transforming the way they work, but putting Code in the title makes it a bit intimidating for anyone who's not a developer. So Anthropic released Cowork last week, bringing Claude Code's capabilities to non-technical users. It’s available for Claude Max subscribers (£90/month) in the desktop app.

Grant Claude access to a folder on your computer, describe what you need, and it works through tasks autonomously. Tasks like: 

• Batch-renaming client files

• Converting invoices into a summary spreadsheet

• Assembling a brief from notes spread across multiple documents 

• Synthesising research across web, articles, papers and notes, voice notes

• Generating Excel files with working formulas and conditional formatting across multiple tabs

• Data visualisation and transformation 

• Slide decks straight from notes or transcripts

I’m very excited about this as we’d love to see more of our clients building tools to solve their problems - and Claude Code can be a bit daunting. Just upgraded our Claude account to Max so that we can explore what we can do. Why don't you try it too?

Claude Cowork

Runway integrates with Adobe Creative Cloud

Adobe and Runway have partnered to integrate Runway's Gen-4 video model directly into Premiere Pro and After Effects. This means editors can generate, extend and remix video clips without leaving their timeline.

Runway also announced GWM-1, their first physics-aware world model for simulating reality. In practice: more believable movement and interactions. Gen-4.5 now generates video and audio together rather than bolting sound on in post, catching up with Sora, VEO and Seedance. And prompts now support camera movement, timing and multiple elements in a single instruction. So reframe it as directing, not just prompting. Something most of us are far more accustomed to.

The Unlimited plan is $95/month (unlimited generations in a slow queue, plus 2,250 credits/month for fast generations on 5–10s clips).

If you're already in Adobe, test whether Runway-in-Adobe fits your existing edit workflow before adding another standalone tool to the pile.

If you haven’t already noticed, Adobe has been slowly turning Firefly into a model marketplace. Here's what else is now available:

• Image: Firefly Image Model 5, plus Black Forest Labs' FLUX.1 and FLUX.2

• Video: Runway Gen-4, Luma AI, Pika, Moonvalley

• Audio: Generate Soundtrack (licensed instrumental tracks from mood descriptions) and Generate Speech (voiceovers in 20+ languages via Firefly or ElevenLabs)

• Other: Topaz Labs for upscaling, Google Gemini 2.5 Flash and OpenAI for text

Coming soon: Project Graph, a node-based workflow editor for chaining models into reusable tools. We are looking forward to seeing how it compares with our platform of choice Weavy, which was recently acquired by Adobe's arch-rival, Figma

Google's image prompting guide is worth your time

Google released a guide on getting better outputs from its incredible image model, Nano Banana Pro. The real value isn't the tool, it's the prompting habits you can reuse anywhere.

The core insight: better prompts aren't longer prompts. They're clearer prompts.

Google's framework focuses on four areas:

1. Goal (what "good" looks like)

2. Audience (who it's for)

3. Constraints (format, length, tone, must-include details)

4. Examples (one good reference beats ten vague instructions).

Google's image prompting guide

Google and Microsoft: agents now complete purchases

The gap between researching a product and buying it is quickly disappearing. Google announced the Universal Commerce Protocol (UCP), an open standard letting AI agents handle the full shopping journey from discovery to checkout. This means shoppers can buy products directly within Google's AI Mode and Gemini without visiting a website. Microsoft is moving the same direction with Copilot Checkout and Brand Agents.

For any e-commerce brands you work with, product feeds, Q&A content, policies and structured data become as important as creative work. An AI agent will only recommend brands whose information is accurate and well-organised.

Google announcement | Microsoft announcement

Lastly, a quick reminder - we’re hosting a live conversation on what AI is starting to mean for agency value. We’ll cover:

• Margins: where AI creates efficiency - and where it creates pressure

• Operating models: what’s actually changing inside agencies

• Client dynamics: how buyer expectations are shifting

• Valuation: how AI readiness is starting to influence PE interest and exit multiples

It’s for agency founders, MDs and senior leaders who want to think commercially about AI.

What makes this session different is Rory’s BenchPress data (UK's largest benchmarking survey for independent agencies) + insights from The Spark Report on AI agency AI adoption + patterns we’re seeing across clients. Great minds backed up with great data, I can’t wait.

👉 Reserve 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?