The 10/20/70 Rule of AI: What Service Business Owners Get Wrong
You bought an AI tool three months ago. Maybe a chatbot for your website. Maybe an AI workflow platform the demo made look effortless. You set it up in an afternoon. And now it sits there — the dashboard has the same data it had on day one, the chatbot answers two questions a week, and nobody on your team trusts the workflow enough to let it run unsupervised.
The tool wasn't the problem. The AI model powering it wasn't the problem either. The problem was everything else — the process nobody redesigned, the data that lived in three different places, the team that never learned to work with it. And that's exactly what the 10/20/70 rule for AI is about.
If you're still wondering where to start with AI for your service business, this framework is the missing piece most vendors skip.
What the 10/20/70 Rule for AI Actually Says
The 10/20/70 rule comes from observing what makes AI projects succeed and fail in real businesses. Here's the breakdown:
10% — The model. The AI algorithm, the language model, the prediction engine. This is what vendors demo. It's also the smallest piece of a successful deployment.
20% — The technology and data plumbing. The integrations, the APIs, the data cleanup, the infrastructure. Connecting the AI to your CRM. Making sure it has clean customer records to work with. Building the pipeline so data flows one way and results flow back.
70% — The process redesign and people change. Redesigning the workflow so the AI fits. Training your team when to trust the output and when to override it. Redefining who does what. Communicating to clients that they're getting faster responses — and that a human still reviews everything.
Most business owners spend 90% of their energy on the 10%. They obsess over which model to use — ChatGPT versus Claude versus Gemini — while ignoring that the model is the least important variable.
The 10%: The AI Model Is the Easy Part
Picking the model is not the hard part. They all work well enough. For 95% of what a service business needs — answering client questions, drafting estimates, summarizing job notes — any modern model delivers.
The reason most AI purchases fail has nothing to do with which model you chose. It has to do with what happened after you chose it — or more accurately, what didn't happen.
You see this pattern constantly: a business buys an AI tool, spends a weekend setting it up, shows it to the team, and waits. Nothing changes. The team keeps doing things the old way because the old way is what they know, and the AI doesn't fit into how the work actually flows. As we covered in Don't Buy Another Tool Until You've Mapped Your Workflow, the tool should be the last thing you choose, not the first.
The 20%: The Tech Plumbing Is Where Budgets Bleed
This is the silent budget killer. That AI chatbot you bought? It needs access to your pricing database, your scheduling system, your customer history. Those systems don't talk to each other. So you pay for a middleware tool. Then you pay someone to map the fields. Then you discover your pricing data has three different formats depending on who entered it.
Before you know it, you've spent more on the plumbing than the AI model itself.
The businesses that skip this trap are the ones that already run on one system. When your website, CRM, analytics, and client communication live in one place, the AI plugs in once and reaches everything. That's the whole point of replacing tool sprawl with a single system — you eliminate the integration work before the AI ever arrives.
If your data is spread across six tools, clean it up before you buy another AI subscription. The plumbing costs more than the model, every time.
The 70%: Process and People Are Where You Win
Here's what the 70% looks like in practice.
You want an AI agent to handle client booking requests. That's not a model problem — every model can do that. The real work is:
- Redesigning your intake process so the agent asks the right questions in the right order
- Deciding what happens when the agent can't answer — who gets the handoff, and how fast
- Training your receptionist to review the agent's bookings every morning and catch edge cases
- Creating a feedback loop so the agent improves — not by retraining the model, but by updating the instructions and the data it references
- Telling your best clients they'll get faster responses now, and reassuring them a human still backs everything up
That's five changes to process and people. Zero changes to the model. And that's why the 70% is where the ROI lives.
This is exactly what the Map, Architect, Deploy, Calibrate framework solves — it treats the process redesign as the primary work, not an afterthought. You map the current workflow first. You architect the new one with the AI in it. Then you deploy, then you calibrate. In that order.
Most companies do the opposite. They buy the tool, try to fit it into whatever process exists, and wonder why nothing changes.
Where to Start: Pick One Workflow
You don't need to overhaul your entire business. Pick one repetitive process that costs you time every week — client intake, booking, estimate follow-up, status updates. That single workflow is your candidate.
There are five clear signals that a process is ready for an AI agent: it's repetitive, rules-based, high-volume, low-judgment, and the data it needs is already digitized. If your process checks those boxes, it's a candidate. If it doesn't, fix those gaps first — that's the process redesign work, and it belongs in the 70%.
Here's your starting point. Pick one workflow this week. Map it on paper — every step, every handoff, every tool involved. Identify which steps a machine could do and which need a human. Redesign the flow assuming the AI handles its piece. Then, and only then, look at what model or tool fits.
The real cost of AI isn't the subscription. It's the investment in getting the process right. And that investment is what makes the subscription actually pay off.
If you'd like a hand mapping where your team is losing hours — the first step in how we work — book a working conversation at recursive-solutions.com. No pitch. Just a conversation about where your time is going.
FAQ
What is the 10/20-70 rule for AI?
The 10/20/70 rule says that successful AI deployments require 10% effort on the AI model, 20% on the technology and data infrastructure, and 70% on process redesign and people change. The model is the smallest piece of a working AI system.
What is AI sprawl?
AI sprawl is the accumulation of disconnected AI tools and subscriptions across a business — one chatbot for the website, a separate AI for internal queries, another for quoting — none of which talk to each other. It's the same problem as tool sprawl, now with AI tools.
What can a small business use AI for?
A small business can use AI for client communication (answering FAQs, booking appointments), drafting estimates and proposals, summarizing job notes and project updates, automating follow-up emails, and flagging anomalies in operations data. The key is to start with one specific repetitive process.
What AI automations can be done for businesses?
Common business AI automations include automated lead follow-up, appointment scheduling and confirmations, client status updates, invoice reminders, data entry and form processing, and quality assurance checks. The highest ROI automations replace tasks that are repetitive, rules-based, and high-volume.
How much does a business AI assistant cost?
A business AI assistant typically ranges from a few hundred to a few thousand dollars per month, depending on complexity and whether it's a self-managed tool or a done-for-you system. The real cost isn't the subscription — it's the process redesign work to make the AI effective.
Want to talk this through? Book a free 15-minute discovery call