👋 Greetings {{first_name|earthlings}},

August is usually the month for taking stock, which is why I thought it would be the perfect time to share the answers we give to the five questions we get asked most. Quick versions below, each with a longer read behind it if you want to go further.

It's been less of a quiet month here. We've been working with teams in the US, the UK and Scandinavia, and building six (!) new services in response to what clients have been asking for. They land in September and there's a preview at the end of this newsletter. Let me know if you'd like to be first in line.

Someone works out a brilliant way of doing something, but it stays with them only. When they move on, it leaves with them too. Sound familiar? The agencies where AI is making a difference to the bottom line have moved from treating AI as a skill for individual productivity to changing team and business ways of working. They have named an owner(s), created a rhythm for sharing what works, and a process for updating the way work gets done. The blocker to overcome, nine times out of ten, is that nobody has a spare hour.

→ How to move beyond individual productivity

You need a plan for the ninety days after the workshops end. Ours runs in four three-week blocks - set the goals, build the rhythm, scale what works into new SOPs, then commercialise where appropriate. An AI lead with protected time can focus on one agent for each core workflow, built by the team who'll use it - so it’s sustainable, improves over time, and builds capability in your team. We are talking about small visible wins, shared openly and progress one step at a time.

→ Read the 90-day plan

Some people come to us with a vision for transforming the business, and some simply want their team trained up to use AI well. We should stop treating those as two separate things, because the competitive and therefore commercial gains from AI all sit in redesigning how the work gets done. One agency leader at Cannes gave us a good example – 1.2x productivity from bolting AI onto an existing process, against 6x from redesigning the process around it. If you want a bit of inspiration on what's possible, check out this case study of a B2B agency that rebuilt its whole operating system around AI.

→ Read the redesign case study and exercise

Everyone asks about juniors. Entry-level roles are narrowing, and our research found that strategy feels most exposed – the research and analysis juniors used to cut their teeth on is exactly what AI does well. What's changed more is what agencies look for when they hire. Only 6% now recruit for specific AI expertise. They're hiring for aptitude instead – curiosity, adaptability and a willingness to work things out. Which means the old format of the junior apprenticeship needs a major redesign.

→ Hiring for aptitude and how to rebuild the junior apprenticeship

Pricing by the hour means every efficiency you find becomes a discount you hand over. There are four routes out. Output-based pricing, bundled offers, pricing the client's upside, and outcome-based fees. Around a quarter of McKinsey's fees are now tied to outcomes, which tells you something. Test new terms on a new client, learn, adjust and let the proof of it working build.

→ Read why AI breaks the billable hour

We'll publish the split in the next issue.

Everyone’s talking about agents these days. It’s easy to feel that if you’ve not got 12 agents running your whole company for you then you’ve fallen hopelessly behind. So want a clear-eyed view into what they really are, and how you can build them?

On Thursday 20 August, 2pm. Jules Love, my brilliant co-founder, will explain what AI agents really are (hype-free I promise!), where they fit into day-to-day agency work, and what makes one useful rather than novel and then redundant the next week. He'll walk through practical examples and help you scope an agent you could start testing the same week.

More precisely this is Article 50 of the EU AI Act, which has been on the books since 2024 but became fully enforceable from August 2026.

If you're in the UK there's no equivalent rule here yet, so the question is whether your work touches the EU. EU clients, or work published into an EU market, puts you in scope for the Act. Fines run to €15m or 3% of worldwide turnover, so it's not one to file under "later" unfortunately.

The part that's your responsibility is disclosure, and it works differently for images and for text:

Deepfakes – any AI-generated or manipulated image, audio or video – always has to be disclosed with a machine readable watermark embedded in the asset. No public interest test, no editing exemption. Evidently artistic or satirical work gets a lighter form of disclosure, but it still needs one. Adobe Firefly, ChatGPT Image, and Nano Banana all embed content credentials. Midjourney still doesn't. If it's in your production pipeline, that's your first gap.

Text only comes into scope when it's published to inform the public on a matter of public interest (think public sector, health or news-adjacent accounts), and real editorial review takes it back out again. Real means someone with relevant expertise examining the substance, with the authority to reject it, and a named person carrying legal responsibility. A spellcheck doesn't count.

→ Read the Commission's guidance

On 30 July, OpenAI cut Luna – its lightweight, high-volume model – by 80%, to $0.20 per million input tokens. Mid-tier Terra came down 20%. OpenAI says this comes from running things more efficiently.

These are API prices - not ChatGPT subscriptions - so the saving shows up if you're building on the models rather than paying per seat. If you've been putting off automating the high-volume, low-glamour work – tagging, first reviews, transcript summaries – the maths has changed.

→ Read OpenAI's announcement

We've been listening to what you need, and building it. Here’s what’s coming up:

We all love to know where we stand with our AI use against our peers. So we have taken our years of agency assessment and built them into one benchmark. A team-wide survey, leadership interviews, and a score against the Spark AI Maturity Model® and 80+ agencies and brand teams. You get a clear read on where you sit against your peers, and a tailored 90-day plan for what to do about it.

If question one is the one you're sitting with, this is the quickest way to find out where you stand.

And three others landing at the same time:

This one is much anticipated and much requested! It’s for ops teams who keep finding out about problems too late. Your team learns to turn messy inputs into clean records, interrogate the numbers leadership relies on, and build agent around your own workflows.

For planners who want plans that stand up to client scrutiny. Work with AI to interrogate briefs, clarify audiences, test channel roles and model budget scenarios. Connect every decision to the business outcome, stress-test the logic and deliver recommendations your clients trust.

Learn how to commission, scope, budget, plan and deliver top tier AI first productions. Everything from IP and image rights to how to handle nervous clients.

Two days with your team scoping, building and testing agents on real work, then we come back a week later to refine them, and set up the governance to keep them running.

For those of you already well down the road with AI, where the job now is staying ahead and keeping that momentum. Coaching, building and training on a quarterly cycle to keep you at the front.

Not sure which one fits? Reply and tell me what's on your plate this autumn, and I'll tell you which of these would help – or whether none of them would.

If someone in your leadership team is asking one of the five questions we see coming up time and time again, feel free to send this over.

See you next time,

Emma

Co-founder, Spark AI

Spark AI is the AI performance partner that builds organisation-wide AI capability for agencies and brand teams. How? Through role-specific AI training, leadership coaching, AI consultancy and agent building.

We’ve worked with 80+ agencies, published the #1 bestselling book on AI for agencies and publish the AI in agencies benchmark, and teach AI Strategy at Oxford University.