In my last role, I was a one-person marketing team. I ran paid media, SEO, events and everything in between on my own. I learned pretty quickly how to work smarter. I became a little obsessed with finding tools that could take repetitive work off my plate, freeing me up to focus on what really mattered.
So I’ve been the buyer of marketing automation. These days I’m on the other side of the table, and I talk to marketing leaders about AI every week. The question on their mind is: “We know we need help with AI automation, but who can we actually trust to build it?”
So here’s what I’d ask any marketing automation agency or consultant before signing a contract with them. Some of this is general advice for picking a partner. But AI has added a whole new layer of things to watch out for.
Marketing automation means something different now
For years, marketing automation meant a platform. You bought e.g. HubSpot or Pardot, and an agency set up your email flows, lead scoring and forms. If you search for a marketing automation agency today, that’s still mostly what you’ll find: platform specialists selling setup.
There’s still plenty of that work. But AI has added a new layer on top of it. A workflow can now do steps that used to need someone’s judgement. It can read an inbound lead and decide whether it fits your ideal customer profile, find the missing data, draft the first reply and write its reasoning into your CRM.
Singa, a karaoke software company we work with, is a good example. Their inbound leads used to be sorted by hand every morning, which took 45 minutes. Now every lead is enriched, scored against their ideal customer profile and routed to the right rep in HubSpot in under 2 minutes, with a note that explains the decision. You can read the full Singa case.
That’s a very different project from setting up nurture emails. You’ll need a partner who understands your process, your data, the AI models, and who’s still around when the first version gets something wrong. Which it will.
Before you talk to any agency
First sort out 3 things on your side.
Know where your process actually breaks down. You should know where your team spends too much time, gets stuck, or makes mistakes. The more specific you are, the easier it is for an agency to find the right solution. For example, “our sales team gets 200 inbound leads a week, someone manually checks each company’s website, and good leads sometimes wait two days for a reply” gives them something concrete to work with.
Have a clear owner for automation. Someone on your team needs to own the finished automations and the relationship with the agency. They need time in their calendar to give input, make decisions, and involve the right people.
Be realistic about what you’re willing to change. Automation doesn’t always mean taking your existing process and making it faster. Sometimes the process itself needs to change. For example, if your reps update the CRM differently each day, an agency can’t just automate that inconsistency away.
If you’re still deciding whether you need outside help at all, start with what an AI automation agency does.
8 questions to ask before you sign
1. What would you automate first, and what would you leave alone?
A good partner should be able to tell you what not to automate. The best tasks to automate first are repetitive, follow clear rules, and produce results you can easily verify. Tasks that depend on taste or judgement stay with people, at least for now.
A good agency should give you a ranked list of automation opportunities based on business impact, technical feasibility, maintenance and your priorities.
With Singa, the first 2 weeks went into mapping the inbound processes, which turned up 30 to 40 automation opportunities. We ranked them together, agreed on what to tackle first, and had the first automation live in week 3.
If the answer is “we can automate everything”, I’d be careful.
2. Where does a human approve?
Every AI workflow needs clear human approval points. Without them, errors can go unnoticed, bad decisions can multiply, and customer-facing automation can turn into the AI slop and spam we all hate.
For example, we built a reply automation for a client whose outbound campaigns brought in more positive replies than the sales team could answer. Claude drafts each reply, and the draft goes to Slack, where a person edits it, sends it or rejects it. If a conversation reaches a third round, the AI steps out and a person takes over the thread entirely.
A partner who can’t show you where humans review, approve, or take over hasn’t thought seriously enough about the potential risks.
3. Can we build automations inside the tools we already have?
Your existing tools may be enough, but it depends on the use case. A good agency should help you weigh the options, not just sell you another platform.
For example, S-Group’s marketing team already had access to Microsoft Copilot Studio, so it made sense to build there rather than introduce another platform. Now they have a Copilot Studio agent that drafts annual marketing plans from their concept portfolio. It lives inside their existing Microsoft environment, reads from SharePoint, and people reach it through Teams. Check the S-Group case for more info.
Beware: a partner who only works with one platform will usually recommend that platform. Always push to get a clear explanation of why it’s the best fit for your needs and what the alternatives are.
4. What happens if we stop working together?
You should be able to run your automations without depending on the agency that built them. Ask who owns the workflows, prompts, and accounts, and whether everything is documented well enough for your new hire or another agency to take over.
The same goes for the technology. AI models and tools change quickly, and what’s best today might not be best in six months. You want a system where you can replace parts without rebuilding everything.
A good partner builds with that independence in mind. If they can’t ensure easy portability and maintenance, consider it a warning sign.
5. How will we know it worked?
Before the project starts, you and the agency should agree on what success looks like for each automation. Is it time saved, better quality, higher conversion rates, or something else? You should also establish a baseline so you can measure what actually improves.
For example, Olo-apteekki, a Finnish online pharmacy, used to spend 20 to 30 minutes producing each campaign’s set of banners and animations. The app we built for them now generates every format with one click (check the Olo-apteekki case for more). Singa’s morning lead sorting went from 45 minutes to single-digit minutes.
Knowing where you started makes it easier to judge whether the investment was worth it. A typical AI automation project costs EUR 15,000 to 50,000, depending on scope, so you should have some idea of the expected return before committing. Our agency guide shows how to calculate the payback.
6. Do we need a custom solution or an off-the-shelf tool?
Not every problem needs a custom solution. Sometimes an existing tool can do the job just as well. A good agency should help you compare the options.
Olo-apteekki tried several approaches before landing on a custom web app. It uses their real product images and lets AI handle only the layout decisions.
On the other hand, a custom system needs someone to fix it whenever an API changes. If nobody on your team can do that, a ready-made tool might be the smarter choice, even if it costs more per month. If you’re comparing platforms yourself, our workflow automation tools comparison is a good place to start.
7. What training will our team get?
A good automation project shouldn’t leave your team dependent on the agency for every small change. People need to understand how to use and tweak the automations. Always ask what training is included and who it’s for.
When we built Olo-apteekki’s visual production app, we also ran hands-on Claude training for their commercial team. Within days, people had set up their own Claude Projects, and they hit their usage limits in the first week. (I’d count that as a good problem.)
8. What will you need from our IT team?
IT involvement can have a bigger impact on your project timeline than you might expect. Data access, software licences, integrations, and security reviews can all delay the work if they’re not addressed.
“IT has banned everything except Copilot” is something I hear often in discovery calls. But in my experience, IT is rarely the real obstacle. Their job is to protect the company’s systems and data, and they need good reasons to approve new tools.
When marketing brings IT a clear business case that covers costs, data handling, security, portability, and adoption, the conversation often changes. A good agency should help you build that case and involve IT early.
Red flags
- They recommend a platform before understanding your process. You risk paying for a solution that doesn’t solve your actual problem.
- They call every automation an “AI agent”. It suggests they’re more interested in selling AI than choosing the simplest solution that works.
- They can’t show measurable results from past projects. Without concrete numbers, you have little evidence that their work delivers value.
- They can’t explain where human oversight is needed. Without proper guardrails, mistakes can go unchecked and put your brand’s reputation at risk.
- The contract doesn’t make ownership clear. You could end up dependent on the agency to maintain, change, or even access your own automations.
TL;DR
Choosing a marketing automation partner means trusting someone to build systems that make decisions on your behalf. You want a partner who understands your processes, knows where human oversight matters, can show measurable results, and builds systems your team can own and maintain.
Our own AI automation projects follow all these principles.
And yes, we’d expect you to ask us all 8 questions.
