Why Your Company Isn’t Ready for AI? (Yet)
How companies 3× their EBITDA with AI and 3 Blockers for implementing it in your company.
Hey 👋 Nikolay Roll from Tallinn Product Group here.
In this article, we talk about using AI to automate your operations and grow your margins from 10% to 30% (not everyone will get there…), and what it takes.
📖 Word count: ~1700
✨AI Word count: ~35
⌛ Time to read: ~7 min
“Before asking for more Headcount and resources, teams must demonstrate why they cannot get what they want done using AI.”
- Tobi Lutke, Shopify CEO, on X
If that’s the case, then teams at Shopify just got a way to request any headcount or resources they need. 👍
Many CEOs, C-level managers, mid-managers, etc. talk tirelessly how AI will change our lives. Everyone says something along the lines: “AI will change everything”. No one knows what to do with AI exactly, but everyone wants to do it really bad.
Why does everyone want AI?
It’s a way to compete. Launch a cool AI feature, and you get the first-mover advantage: “We do what incumbents do—just better, and with ✨AI✨.”
But the more interesting angle, in my opinion, is “productivity” from AI.
AI can now take on tasks that were mostly handled by people, like understanding messy input (think lawyers parsing laws), interpreting intent (think customer service chat), and even basic reasoning (think accountants analyzing reports).
Now, you can do what people do, but faster, cheaper, and sometimes even better(?).
This isn’t just about faster spreadsheets—it’s about real operational leverage.
Tech-Enabled Roll-Ups: AI to Scale
That’s exactly the premise behind tech-enabled vertical roll-ups.
You acquire operationally heavy service businesses—think hotels, accounting, parking - and apply software and AI to streamline workflows, boost margins, and scale faster than traditional operators.
Take Metropolis (raised $1.9B) which bought legacy parking operators like Premier Parking and SP Plus. They automated license plate recognition and payments, cut operating costs, removed gate hardware, and improved throughput.
Why it matters: in a low-margin industry (~10–20% EBITDA), a 10–15% uplift per location compounds fast.
If you boost margins to 30%:
Your cash flow triples
Your valuation jumps (2x becomes 5x)
You can take on more debt to fund even more growth
In the ZIRP era, the game was: spend all your revenue, lose money, and raise more.
(OGs remember Silicon Valley: “We’re pre-revenue—like Amazon!”) But now the new name of the game is “make profit”. Can you sustain operations even if you can’t raise a huge round?
That’s where AI and automation fit the moment perfectly.
But still “Why Your Company Isn’t Ready for AI (Yet)”
Let’s be honest for second. Most of what we call “AI transformation” today is just automation with a cooler interface. And companies have struggled with automation for decades.
Why are there still so many inefficient, manual processes?
There are three main reasons:
1. System fragmentation
Your tools don’t talk to each other.
One lives in an internal admin with customer PII, another in Salesforce, and a third is an Excel sheet that gets emailed around every Tuesday.
APIs are missing, security says “no,” and integrations break every other week.
Some processes can’t be automated simply because Jira can’t talk to internal tools, because security.
You want to automate something in Salesforce? It depends on:
a) the Salesforce team’s roadmap, and
b) whether your subscription tier even allows it.
If you do not have integrations between your tools and there is no free flow of data between them - it is difficult to automate things.
Maybe you go rogue: CSV exports, a Chrome plugin, and Python scripts.
Not great, but kind of works… Until the security team steps in faster than your script runs once.
Now you’re stuck proving your hack is safe, even before anyone agrees it was useful.
2. Engineering resources
To build solid automations to solve system fragmentation you need:
a) developer resource and
b) budget for better Salesforce subscription and on-prem.
You think, “Let’s do it right—on-prem, secure APIs, clean data layers.” Sounds great... right?
But engineers are already fully booked shipping “solid features,” fixing bugs, and cleaning up tech debt from 2019. No one’s going to write a bulk download endpoint for accounting, unless accounting somehow becomes a “strategic initiative”.
Automation sounds like a smart investment - until you realize it’s competing for the same roadmap space with other new features which have impact in $millions against your “saving couple of clicks every hour”.
3. Process ambiguity
But honestly, the biggest problem is probably process management—or rather, the total lack of it.
If a process isn’t clearly defined, you can’t automate it. Period.
You get the usual signs:
“Our knowledge base is Slack.”
“Code is the best documentation.”
“Ask Martin.”
All of these are just polite ways of saying: we actually don’t know how our processes work.
Many internal processes aren’t really “designed” at all, they’re “improvised” through chaos, handovers, and legacy decisions no one remembers. Different teams follow different rules, often without realizing it.
Take something as basic as refunds. If the process isn’t documented, your agents are thinking every time:
“It’s been 5 days since the order—do we refund or not?”
“What’s our refund policy again?”
“Wait, do we even do refunds on Fridays?”
You can’t automate tribal knowledge.
Automation needs a clear, rule-based process. In code, that usually means writing a lot of “if/else” statements.
If you can’t figure out your “ifs”, “elses”, and “thens”, no automation will work.
But AI can solve it, right?
All the same problems still apply. But now we’ve added a new layer of complexity:
Where does the model run? (Cloud or on-prem?)
What data can it access, Jira tickets, Zendesk tickets, DB, PII? (And what shouldn't it?)
What happens when a prompt update accidentally refunds €10,000 to the wrong customer?
Oh—and who’s responsible when it goes wrong?
AI doesn’t magically solve process ambiguity or system fragmentation.
It just guesses better than your average rule-based bot—until it doesn’t.
And when it fails, it fails confidently.
AI is powerful when it comes to converting unstructured data into structured, understanding intent, reasoning, and more. That’s why CS use case is so compelling. But it’s not plug-and-play.
It still needs clean processes, solid data flows, and real operational thinking.
The only difference is that now, if you get it right, the leverage is 10x.
But if you get it wrong? You just built the world’s most expensive random number generator…
So, “Why Your Company Isn’t Ready for AI? (Yet)”
Before you implement AI, ask yourself: What problem are you actually trying to solve? (very PM vibe)
Start by identifying the processes you want to improve.
This means mapping how things work today, estimating the potential impact of change, and figuring out the cost and complexity of making it happen.
Once you have clarity on the process and the numbers to back it up, you’re ready for the next step: getting buy-in.
That means selling your idea to PMs, EMs, your M, other teams’ Ms, and whoever else controls budget and engineering resources. You need allies before you ask for action.
Then? Don’t start with AI.
Start with something simple and rule-based.
Most workflows can be improved with basic automation - no LLMs required.
Think of customer support: long before AI chatbots, just showing the FAQ in the chat window solved most issues.
Once that works, and you’ve proven the value, then it makes sense to explore how AI might take it further.
It’s like learning to walk before you run.
Or better yet - as Maksim Butsenko shared in our episode on Bolt’s ML journey.
Start with a rule-based system. Once that’s working, then level up to machine learning. Then, maybe, go to LLMs.
If your company is already at the stage where every process is automated with AI, then congrats!
Make sure to share your story with us at tpg.ee, or drop it in the Tallinn Product WhatsApp group (if you know, you know). 🙂
Thanks for reading—and subscribe for weekly Product Content!
As an extra note. In general, I’m genuinely happy to see more C-levels, managers, etc. starting to care about automation and improving general health of their business. My only concern is that all these words need to turn into real action - and in the right direction.