This will happen to you in 2027 (if not already): Someone senior has been reading about AI all year — news headlines, LinkedIn posts, something they heard over dinner — and now they have tons of questions about how their marketing team is using AI.
It could be a budget review, a planning session, or one of those end-of-quarter conversations that was supposed to be about pipeline, but suddenly pivots to AI and marketing.
The questions are fair, of course, and I’d be asking them too if I were paying for something I can’t really see. So it’s time to start prepping your responses:
1. What is your AI spend supposed to move?
Most teams can tell you what they’re spending on, but not everyone can tell you what it was supposed to do.
Your AI tool seats got added because someone needed them, and the usage crept. A tool got renewed because switching looked like more hassle than paying.
So pick your spend number before, while you’re still deciding whether to spend, written somewhere you can’t edit later when the number turns out to be inconvenient.
If you’re already past that point — and most of us are — there’s a shortcut that forces the same clarity. If your budget were halved tomorrow, what goes first? If you can’t rank your own stack in about thirty seconds, that’s your answer.
Here’s the version of this I’ve started using: take your monthly AI spend and convert it into people. Not metaphorically — divide it by what a person actually costs you. A small team spending a few hundred a month has bought some fraction of a colleague. Are you getting that fraction of a colleague’s work out of it? It’s a crude test and it clarifies things fast.
What I say now: "Here's the number each tool was bought to move, and here's what I'd cut first."
2. What can you do now that you couldn't do eighteen months ago?
This is the one I’d least like to be asked cold, because for a long stretch my honest answer was “the same things, faster.”
Which sounds fine. It’s what everyone says. It’s also what you say when you haven’t checked, and I hadn’t. Faster is table stakes in 2027, and nobody’s impressed, and more to the point, nobody can see it.
What counts is a thing that exists that didn’t before. A new channel you finally tested. A segment that wasn’t worth the effort and now is. The research you’d been meaning to do for two years.
There’s research doing the rounds suggesting a lot of people report saving real time each week with AI, while most of them get no guidance at all about what the extra time is used for. And time you free up but never assign doesn’t become capacity — it becomes slightly less pressure, but absorbed into the week and gone.
What helped me was naming the destination in advance. Not just “more time for strategic work”. Something specific enough that in six months I could point at it.
What I say now: "Two things exist now that didn't before. Here they are."
3. Which of your team’s skills are you deliberately letting go, and what did you do with the time?
A skill gets used less because there’s a faster route now. Then rarely. Then the person who was good at it moves on and isn’t replaced, because on paper the tool covers it. Nobody sat down and chose to lose the capability, but it quietly went, and you find out it’s gone at the exact moment you need it back.
Doing it on purpose means saying out loud what you’re handing over — and, more importantly, what you’re keeping in human hands deliberately, knowing it’s slower.
This also ties back to the second question: when you do let something go, where did the freed time actually land? If you can’t trace it to something specific, you didn’t reallocate anything. You just got a bit less busy, which is lovely but awkward to defend in a budget meeting.
On a team of five, one person’s skill is the team’s capability, with no adjacent department holding the same knowledge. And the hiring market here doesn’t refill a lost skill on demand: try replacing someone who genuinely understood paid social across three markets and two languages and see how long that takes.
What I say now: "Here's what we've handed over, here's what we're keeping on purpose, and here's where the freed time went."
4. It’s been over a year, where’s the ROI?
The one everybody’s dreading. Pro tip: think of the shape rather than just numbers.
“Where’s the ROI” assumes a timeline that doesn’t match how AI use actually shows up. Early on you’re buying capability — people working out what the tools are for, workflows getting rebuilt, things tried and dropped. Wanting financial evidence from that phase means either you’re misreading the signal or you’re spending on the wrong thing.
What shows up first is qualitative: someone pointing at a specific workflow that changed and what it did to their numbers. Financial evidence comes later, usually as subtraction — a contractor you stopped needing, a tool you cancelled, something you stopped outsourcing.
What I say now: “Here’s what we expected by this point, here’s what we’re actually seeing, here’s what should show up next and roughly when.”
None of these four questions are difficult once you’ve thought about them for an afternoon. You don’t need perfect answers — AI use cases and governance are changing too fast for that. You just need answers at all, for when your boss asks.
The AI;DR
Elsewhere in the AIverse
Google ships vertical AI for legal and financial services. Google has launched customised workflow products for legal and financial services professionals — ready-to-deploy agents, pre-built skills for common tasks, and connections into third-party ecosystems. It’s a direct move onto Anthropic and OpenAI’s enterprise turf. If you sell into either sector, your buyers are about to have a much better-informed default assistant sitting between them and you.
Alibaba opens up Wan 3.0. Alibaba widened access to Wan 3.0, which generates clips up to 30 seconds from almost any input — text, video, documents, spreadsheets, slides, webpages. Slides-to-video is the interesting input here — most of us are sitting on a lot of decks. Worth thirty minutes of testing before you brief anyone for social video.
SpaceXAI is putting AI compute in orbit. SpaceXAI is adopting Nvidia’s Vera CPU cluster, including a space-optimised version riding aboard Starmind satellites into orbit in Q4 2027.