Guide
How to choose an AI agency.
A practical guide for your first meeting with any AI agency, including us. What to ask, which answers to listen for and which warning signs to avoid.
Updated
Sebastián Delsalto, Founder of Daleki Lab and Daleki Global
Read more about Sebastián at onesebdel.com
In short
Choose an AI agency that starts with your problem rather than a tool, shows you something working before asking for a large investment, keeps accounts, data and code in your company's name, defines human oversight and acceptance criteria in writing and can operate what it builds.
Ten questions for your first meeting.
Use them with any provider. Under each one you will find how we answer it and where you can check.
- 01
Do you start with my problem or with a tool?
Why it matters. An agency that brings a solution before understanding your process usually sells what it already has.
- A good answer
- Questions about your operation, your data and the result you want to change, before any talk of models or platforms.
- Warning sign
- A generic demo in the first meeting and a proposal written without seeing your process.
- How we do it
- We start by understanding the challenge and tell you when AI is not the right fit. See our method
- 02
Can I see something working before committing to a large investment?
Why it matters. Deciding with a prototype reduces the risk of building the wrong thing.
- A good answer
- A first version with an agreed scope and clear limits on what it is not yet.
- Warning sign
- A large contract with no validation stage.
- How we do it
- Before a proposal, we show you a first version of your idea with an agreed scope. How the first version works
- 03
Who will own the code, the accounts and the data?
Why it matters. If the agency controls everything, switching providers becomes expensive or impossible.
- A good answer
- Accounts and data in your company's name and code ownership defined in writing.
- Warning sign
- Agency-owned platforms with no way to export your information.
- How we do it
- Accounts, data and infrastructure in your name. Code developed for your project becomes yours once the solution fee is paid. Read the terms
- 04
What will AI decide on its own, and what will a person approve?
Why it matters. An agent without limits can send, record or publish something it should not.
- A good answer
- Permissions, human approval points and a log of what the system does, defined from the design stage.
- Warning sign
- Promises of full automation with no mention of errors or exceptions.
- How we do it
- Every important decision keeps an owner on your team, and every system action leaves a trail. How we design human oversight
- 05
How will we measure whether it worked?
Why it matters. Without a baseline, any result is just an opinion.
- A good answer
- A metric agreed before building, such as time, errors, quality or cost, and estimates presented as estimates.
- Warning sign
- Guaranteed savings percentages before anyone understands your operation.
- How we do it
- We agree on the baseline at the start and present results as measurements, not promises. What a baseline is
- 06
What costs will there be beyond your work?
Why it matters. AI models, messaging and hosting are billed by usage and grow with volume.
- A good answer
- A proposal that separates the initial investment, monthly support and third-party costs, with caps and monitoring.
- Warning sign
- A fixed price that never mentions usage costs or who pays them.
- How we do it
- We work with a solution fee and a monthly fee. Third-party costs are listed separately and, unless agreed otherwise, contracted in your company's name. See how we quote
- 07
Who operates the system after launch?
Why it matters. An AI system needs ongoing attention: data, models and your business all change.
- A good answer
- Owners, response times and a monthly capacity agreed in writing.
- Warning sign
- Delivering and disappearing, or unlimited support with no clear terms.
- How we do it
- The same team that builds the system supports its operation, with agreed capacity and priorities. AI operations and evolution
- 08
How do you protect my company's and my customers' information?
Why it matters. In Ecuador, the Organic Law on Personal Data Protection requires a legal basis, security measures and identified providers.
- A good answer
- Least-privilege access, declared providers, a confidentiality agreement and anonymized test data whenever possible.
- Warning sign
- Asking for passwords or full customer databases over WhatsApp in the first conversation.
- How we do it
- We sign your confidentiality agreement, declare our providers and, whenever possible, start with anonymized samples. Read the privacy policy
- 09
What have you built that I can see and use?
Why it matters. A portfolio of live platforms says more than a wall of logos.
- A good answer
- Real projects you can open, with an honest description of what was done.
- Warning sign
- Anonymous testimonials, unsourced figures or awards you cannot verify.
- How we do it
- Every project in our portfolio links to its live platform and describes the scope, not invented results. See the projects
- 10
What happens if I want to switch providers?
Why it matters. A good relationship also defines how it ends.
- A good answer
- Exit terms in the proposal, delivered documentation and full access to your accounts.
- Warning sign
- Long lock-in periods or hidden dependencies.
- How we do it
- Exit terms are defined in the proposal and, because the accounts and data are yours, you can continue with your team or another provider. Read the operations FAQ
Warning signs.
- Guaranteed savings or sales before anyone understands your operation.
- Agency-owned platforms with no way to export your data.
- No mention of errors, exceptions or human approval.
- Fixed prices that ignore model usage costs.
- Requests for passwords or customer databases in the first conversation.
- A portfolio with no projects you can open and verify.
Agency, in-house team or off-the-shelf tool.
You do not always need an agency. This comparison helps you choose the right path for your need.
Specialized agency
When it makes sense
You need strategy, product and engineering together without building a full team, or you want to validate an idea before investing more.
What to watch
Require ownership of what is built, documentation and exit terms in writing.
In-house team
When it makes sense
AI is central to your product and you can sustain a specialized team over the long term.
What to watch
Hiring and retaining AI talent takes time, and knowledge can end up concentrated in a few people.
Off-the-shelf tool
When it makes sense
Your need is common, such as transcribing meetings, and an existing tool solves it without customization.
What to watch
Your data lives on the vendor's platform and adapting it to your process is limited.
Many companies combine all three: a tool for common needs, an agency for specific ones and an internal owner who governs both.
Questions about hiring an AI agency
How much does it cost to hire an AI agency?
It depends on the scope, the integrations and the ongoing support. Be wary of a generic price for different needs. Ask for a proposal that separates the build, monthly support and third-party costs such as AI models and messaging.
Is a local or an international agency better?
What matters is that it understands your market, your language and your rules, such as your country's data protection law, and can support you in your time zone. A local agency adds business context and in-person meetings. Expect the same ownership, transparency and control from any of them.
What should I prepare before talking to an agency?
You do not need a technical document. Be clear about which process you want to improve, who uses it, which tools and data exist and what result would tell you it worked.
How long does a first AI project take?
It depends on the scope. A healthy approach splits it into stages with deliverables you can review: a first version to validate the approach, the build and a launch with measurement.
Ask us these ten questions.
Book a conversation and ask us exactly what you would ask any agency. We answer with documents you can review.