The 10-20-70 rule for AI implementation is a planning idea that places roughly 10 percent of the effort on technology, 20 percent on data and process foundations, and 70 percent on people, workflow change, and adoption. It is not a fixed formula or a guarantee, but it is a useful reminder that buying an AI tool is usually the smallest part of making it work.

For a small business, the rule helps shift attention from a product demo to the daily work around it. The important questions are who will use the tool, what process will change, how outputs will be reviewed, and how the team will know whether the new method is helping.

Put the question in the context of daily work

For a local or service business, an AI decision is useful only when it makes a real part of the work clearer, faster, or more consistent. Start with the work your team repeats: answering inquiries, preparing estimates, documenting jobs, scheduling appointments, following up after service, or keeping internal information organized. A tool should support those activities without creating a second process that staff have to remember during a busy day.

Before changing anything, write down how the current task moves from beginning to end. Note the trigger, the information needed, the person responsible, the customer-facing step, and the usual exception. This makes it easier to see whether AI is a reasonable support for the task or whether the underlying process needs attention first.

What the 10-20-70 rule for AI implementation looks like in a small business

The technology portion includes selecting a tool, setting up access, and connecting it to the systems your team already uses. The data and process portion includes cleaning the inputs, defining an approved workflow, and deciding what information is safe to use. The largest portion is people: explaining the purpose, training staff, collecting feedback, updating habits, and keeping an owner accountable after launch.

Keep the first version narrow. A good initial workflow has a predictable input, an output your team can recognize as useful, and a person who can review it. For example, an office manager might use approved information to prepare a follow-up draft, while a service manager reviews it before it reaches a customer. This is more manageable than trying to automate every message or decision at once.

  • Choose one recurring task rather than a whole department.
  • Describe what a good result includes and excludes.
  • Use a small set of real, low-risk examples for testing.
  • Assign one person to collect feedback and update the process.
  • Keep a clear fallback when the output is incomplete or wrong.

Start with the business problem, not the software

A feature list is not a plan. Begin by identifying the cost of the current problem: delayed lead response, time spent searching for information, repeated data entry, inconsistent documentation, or missed follow-up. Then decide what would improve it. The outcome may be a shorter turnaround time, fewer handoffs, a more complete record, or more time for the work that requires experience and judgment.

Get input from the people who do the task. They usually know which requests are routine, which details are often missing, and where a shortcut would create risk. This step prevents an owner from selecting a tool based on a demonstration that does not reflect the realities of dispatch, client service, scheduling, or delivery work.

Run a controlled pilot

Use a short pilot before making a process permanent. Test one workflow for two to four weeks with a limited group of records, customers, or team members. Keep the previous process available while you learn. A pilot is not a promise that the tool will stay; it is a way to determine whether the workflow helps enough to justify the time, cost, and training involved.

  1. Record a baseline for the current task before changing it.
  2. Set one or two specific outcomes for the pilot.
  3. Give participants approved instructions, examples, and escalation rules.
  4. Review results frequently during the first week.
  5. Decide whether to refine, pause, stop, or expand using the evidence.

A practical pilot includes imperfect cases, not only easy examples. Test what happens when a client leaves out key information, a request is urgent, a calendar changes, or a staff member needs to hand work to someone else. Those are the moments when a workflow either supports operations or adds another point of failure.

Keep people responsible for important decisions

AI can prepare, summarize, organize, and suggest. It should not quietly take responsibility for commitments your business makes to customers. Keep a person accountable for final pricing, service scope, legal terms, safety issues, refunds, complaints, and unusual requests. The same approach applies to internal decisions affecting employees or sensitive customer relationships.

Make escalation rules simple enough to use under pressure. For instance, route a request to a person when it involves a complaint, a deadline, an exception to policy, a financial commitment, or unclear information. A visible rule is more dependable than expecting staff to remember a general warning about “using judgment.”

Protect customer information and trust

Treat the percentages as a conversation starter, not a budget allocation. Do not upload customer records simply because a tool can accept them, and do not assume an integration is safe without reviewing permissions. Keep human review for customer commitments, sensitive information, and exceptions until the process has earned trust.

Review vendor settings before connecting an AI tool to email, a CRM, shared documents, or scheduling software. Limit access to the information needed for the defined task and use role-based permissions where available. Do not paste payment data, credentials, private health information, employment records, or confidential contract details into a tool unless your business has specifically verified that use is appropriate.

Customer trust also depends on clarity. If a message is AI-assisted, it should still be accurate, understandable, and consistent with the service your business can actually deliver. Review templates for local terminology, service-area limits, pricing language, and promises about timing. A polished draft is not useful if it creates the wrong expectation.

Train the team around one repeatable method

Training does not need to be a lengthy technical course. Give staff a short operating guide: when to use the workflow, what information to provide, examples of acceptable output, what to check before acting, and when to escalate. Pair that guide with a few examples from your own business so people can see the standard in context.

Ask team members to flag recurring corrections. If the same problem appears repeatedly, change the template, source information, or process rather than asking each person to fix it individually. This turns training into gradual improvement instead of a one-time launch event that fades when work gets busy.

Measure a useful result

Track whether the team actually uses the approved workflow, how much correction it requires, and whether the original problem improves. A useful baseline might include response time, time spent preparing documents, follow-up completion, or the number of handoffs. Pair those measures with staff feedback about whether the change fits the work.

Review both efficiency and quality. A task completed more quickly is not a win if it produces unclear estimates, duplicate appointments, incorrect records, or more customer callbacks. Compare the pilot with the baseline at a regular time each week, and include feedback from the staff who use the workflow and the people who receive its output.

Also include the full cost of the change: subscription fees, setup time, training, review time, and any integration work. A modest improvement may still be worthwhile, but the decision should be based on the actual operating picture rather than a tool’s activity dashboard.

Build the process into normal operations

If the pilot works, document the process before expanding it. A simple standard operating procedure should name the trigger, owner, approved inputs, review step, escalation path, and metric. Store it where the team already looks for guidance. This is especially helpful for small teams, where one employee’s absence can otherwise interrupt a workflow that only they understand.

Expand to an adjacent task only after the first one is stable. A business might move from internal meeting summaries to proposal preparation, or from inbox sorting to follow-up reminders. Sequencing work this way reduces tool sprawl and gives the team time to develop confidence before the next change.

Common mistakes to avoid

  • Buying several overlapping tools before proving one use case.
  • Launching customer-facing automation without a review and escalation path.
  • Assuming a tool can fix inconsistent data or an undefined process.
  • Leaving ownership unclear after the initial setup.
  • Measuring usage while ignoring quality, corrections, and customer impact.

The best result is not the most automated operation. It is a dependable process that gives your team more capacity for skilled work, helps customers receive a consistent experience, and remains understandable when your business gets busy.

Practical next steps

For a phased rollout, read How to Implement AI Without Disrupting Your Business. Use How to Measure Success with AI Implementation to set evidence-based checkpoints, and review how to train a team to use AI tools before expanding.

Choose a modest next step: map one process, collect a baseline, identify an owner, and test an AI-assisted version with clear review rules. That approach gives a small business useful evidence without committing its team or budget to a broad change before it is ready.

Bottom line

The 10-20-70 rule for AI implementation is a planning idea that places roughly 10 percent of the effort on technology, 20 percent on data and process foundations, and 70 percent on people, workflow change, and adoption. It is not a fixed formula or a guarantee, but it is a useful reminder that buying an AI tool is usually the smallest part of making it work.

Keep the scope focused, use people to make consequential decisions, and let real operating results determine what comes next. That is a practical way to adopt AI while protecting the service and relationships your business depends on.