A month ago, I had a great time with Marcos González—and learned a lot from him—. We facilitated a discussion about AI with some SMEs managers and technicians.
They main idea we put on the table to spark the discussion was the need of an AI strategy.
Is strategy just for big companies?
No.
The term is scary: it seems like we’re going to spend more time defining it than actually doing things. Also that we’re going to prune flexibility… But strategy is just a very inflated word to say that instead of everyone doing whatever they want or what they think is best, if we want to avoid wasting efforts and resources, it’s better to know beforehand what we all as a group want to achieve (side note: What we want to achieve is not ‘Use AI’. AI is a tool.) And once we know what we want to do, we thisk about how we’re going to get there.
We don’t need a 20-pages document. On one sheet. We can say what we do first and how, and what things we’re not going to do. Then we can make it look nice and stick it on the wall so that anyone is aware and the decisions we make follow the plan we’ve agreed on. We don’t want silos doing things others are not sure what they’re worth, or small kingdoms coexisting with AI deniers. We want to be sure we all pull in the same direction.
Of course plans can change. They evolve. I don’t believe in love at first sight or doing things perfectly the first time. If I see something isn’t working, I iterate and improve it. And I continue. But without a plan, I don’t know where I’m going or how I want to get there. And if I don’t know—as a manager—I don’t expect anyone else to know.
Cool… do you have some examples of how a strategy looks like? I have my ideas… In my opinion, the difference between succeeding with AI and throwing away money or time while someone else eats your lunch comes down to these 4 fundamental points.
1.- Invest in Robust tools (and don’t be stingy).
In some countries like Spain, we struggle with spending, but we need to remember that (1) buying an AI tool is an investment and (2) free versions of tools are for testing or small things, not for serious work. Sometimes we know of a tool that works for us but instead of investing in paying for it, we have the temptation to invest time (which is money) trying to find another tool that does the same thing for free. And why would there be someone offering the same thing for free? Maybe it’s because it’s not the same.
When you pay for an AI tool, you might be paying for many things like what I list below. But I reckon it’s a very long paragraph! How about this: For a person earning €2,000 (gross) who costs the company €2,900 per month (+€650 social security +€200 office & tools), I can recover the €50 of a paid app if that given person saves just 2.8 hours a month with the tool. That’s less than it takes to find a free alternative to one that works (plus testing it and discover it’s not the same 😛 ).
If savings don’t convince you… this is what you pay for in a paid app:
- Professional usage limits. Free levels have fewer messages, less priority, and more restrictions.
- Advanced features. Obviously, when you pay you get more things that make a difference. For example:
- Perplexity (Pro / Enterprise Pro). More powerful searches (Pro Search / Deep Research) and much higher or unlimited usage. Advanced file usage: upload many documents, analyze them, and cross‑reference information in a single research. Choice of advanced models (more powerful and up‑to‑date models). Spaces / shared projects for teams and collaboration. Search over internal knowledge (your own connected docs and sources). Enterprise controls: SSO, data governance, compliance (SOC2/GDPR), and guarantees that company data is not used to train public models <- This last point seems crucial to me.
- ChatGPT (Plus / Team / Enterprise). More powerful and faster models (for example GPT‑4) that are not available, or are very limited, on the free tier. Much larger context windows to handle long documents and extensive conversations (not having to explain the same thing twice). Advanced Data Analysis (Code Interpreter): upload files, analyze data, generate charts, run code safely. Fewer usage limits (more messages, more throughput) and priority access to resources. SSO, user administration, shared workspaces, and centralized policies. Guarantees that your data is not used to train public models, plus enterprise‑grade security and compliance (SOC2, etc.).
- Cursor (paid plans). Much higher AI usage limits (more completions and more context per day). Better repo context: it can use more files and larger project sizes at once. More “rules” and configuration options per project (coding style, guidelines). Better performance for heavy tasks: large refactors, full‑file generation, tests, etc. Team features: share AI configuration, common policies, centralized management.
- Sometimes it is not about paying for an AI tool itself but for the premium tier of something you already use that lets you integrate/use AI. For example, on WordPress.com, paid plans give you much more usage of the AI assistant without early limits, better integration in the editor to draft and rewrite texts, access to more design options and custom domains so what you create looks professional, and the ability to install advanced AI plugins (chatbots, bulk content generation, automations) that are not available or are very limited on the free plan.
- Integrations with your other systems so that AI stops being an isolated tab and you start doing magic without constantly copying and pasting, or so that AI is directly designed around your workflow and does exactly what you need. For this, you can look at https://theresanaiforthat.com/, where as of today there are 44,632 AI tools listed for 11,353 different tasks or 5,167 jobs. For example, today I see 37 AI tools for restaurants and some interesting ones to prepare for ISO audits…
- Security and compliance: Certifications, data protection with encryption, access control, data governance with contractual options that your prompts and documents aren’t used to train general models, regional data residency, audit logs, and incident response plans.
- Real technical support. When AI is in a critical element, you can’t afford the system to go down at dawn. Examples: Chatbots with 24/7 support with different SLAs and channels for plans, precisely to guarantee service continuity.
2.-When in doubt: buy and test.
You think there’s a tool that could help you improve a workflow, save time on something, bill more, gain customers… Is there a possibility that some AI tool could generate revenue or savings worth more than the 30 euros it costs per month? Even if that possibility is remote, don’t get paralyzed analyzing. Most tools have trial periods or money-back guarantees, but even if they don’t, what you’ll learn about the tools themselves, or what they’ll make you reflect on about how you do things, is worth those 30 euros.
Put it another way… according to the Economist, the average loss in gambling by adult in the US is well over $400/yr and in the UK just a bit under that figure. There’s much more probability you’ll find an app that makes your life easier for $40 than winning something significant when gambling 10x money.
3.- Learn by doing.
In our Automattic Capacitation Framework, we’ve chosen a model to represent how we Competent/Expert/Native in AI where each person goes through several stages like exploring and gaining familiarity with new skills, practicing these new skills, being able to guide others… and finally, innovating by applying AI to create real solutions and improvements. And to transition from one stage to the next ones, you have to learn.
We don’t believe you learn by attending to courses alone. We firmly believe we learn by doing. (Constructivism. Again). I suggest you consider learning AI by doing for your strategy:
- I’d expect two-thirds of learning to happen by doing things. Touching, testing, breaking, making mistakes, and iteratively improving—little by little—our solutions to real problems. This is what the first two points in this post are really about: if there’s an AI tool or AI capability in a tool you already use… Go for it! Give it a chance and learn.
- Of the remaining third, two-thirds of learning will likely happen by sharing with other colleagues (Social Learning). This can be within the company, pooling what’s learned, or going to events, conferences, meetings where you can chat and share with people who share our same concerns, problems…
- The third of the third would correspond to formal learning: face-to-face courses or MOOCs, certifications, books, manuals.
In other words, first use AI to solve real problems. Then learn with and from your colleagues. After that, take the course.
4.- Create a culture where what’s learned is documented and shared.
For that social learning to happen, we need to enable a “whiteboard” where the team shares:
- What tools we’re testing / have tested
- What use cases it works for and which ones it doesn’t. What we recommend it for
- What workflows give good results
- What mistakes we’ve made
Yes, sure, a whiteboard, like in startup movies. Well, it could be, but if I aim to test a good number of tools, it might get too small too soon :). That whiteboard can actually be a shared Word or Google Docs, an internal blog, a Slack channel, Trello cards, or whatever you like best. What’s important is that it’s accessible and that you create a culture where everyone’s contribution is rewarded.
What if I work alone in the company? Well, the whiteboard is still useful for self-documentation. And especially value chatting with professional colleagues or peers and sharing what you’ve learned and learning from them.
Epilogue: How to comply with everything in one stroke.
The first version of OpenAI’s ChatGPT that triggered the fever appeared on November 30, 2022. Our CEO, Matt, gave us the starting signal for AI use, encouraging us to try any AI tool we thought could mean a productivity improvement, right after Christmas that year. Automattic would cover the cost during the trial period and later if we considered it useful, and in exchange we just had to share the experience explicitly and visibly for colleagues, suggesting or not its use for certain use cases.
Today, we have a generous monthly limit for AI tools without validation requirements (reviewed just over a month ago). For larger expenses, there’s a very light process: explain what you need and what for. In exchange, we commit to share what we learn in a way that’s easy to find and link to. Some exaples are:
- Documentation on whiteboards – accessible repositories where any colleague can search and link knowledge
- Live demos – Zoom sessions where we show how we use AI in interesting contexts, with recordings for those who can’t attend
- Shadowing – Observing how a colleague works with AI, to learn with them, even noticing elements that might go unnoticed when you explain what you do
- AI workshops, sharing sessions, or group projects at meetups
- …
The deal is simple: freedom to experiment in exchange for generating collective knowledge. Invest, Test, Learn by Doing, and Share.
See how it wasn’t so hard to define and apply a strategy?
Leave a Reply