Key takeaways
- Most small businesses are already somewhere in the middle of AI adoption without a plan — this roadmap gives the messy reality a clear sequence.
- Five phases, in order: assess where you stand, set the rules, train your people, automate a real task, then scale what works.
- Skipping a phase to move faster usually costs more time later — automating before you have rules just means automating a problem faster.
- 76% of small businesses already use AI, but only 14% say it's fully integrated into core operations — this roadmap is how you close that gap deliberately.
- You don't have to do this alone or all at once; a Kickstart engagement runs exactly this sequence with your team, starting from wherever you are today.
Most small businesses don't lack AI tools — they lack a sequence. Someone's using a chatbot, someone else has an AI feature turned on in their invoicing software, and there's a vague sense that "we should be doing more with this," but no clear idea of what comes next or in what order. This roadmap gives that messy, half-started reality a clear path: five phases, done roughly in order, that take a business from scattered use to something that's actually working and measurable.
None of the five phases below require a large budget or a dedicated technical hire. They require, mostly, a willingness to slow down at the start — assessing and setting rules before jumping to the more exciting parts — so that the training and automation that follow actually hold up instead of needing to be redone six months later.
Why the order matters more than the speed
Goldman Sachs' 10,000 Small Businesses survey found 76% of small businesses report currently using AI, but only 14% say it's fully integrated into their core operations. That 62-point gap isn't a speed problem — it's a sequencing problem. Businesses in that gap usually skipped straight to using tools without ever settling the rules or training that make the use safe and repeatable, so it stays scattered no matter how long they keep using AI.
The order below isn't arbitrary. Each phase depends on the one before it: rules without an assessment address problems you haven't identified yet; training without rules teaches people to move fast with no guardrails; automation without training builds on a foundation that isn't solid; and scaling before you have a working example just multiplies whatever wasn't working in the first place.
Jumping straight to phase 4 — automation — because it's the most exciting phase, while skipping phases 2 and 3. This is how a business ends up with an impressive-looking AI tool and no rules governing what data feeds it.
Phase 1: Assess
Before changing anything, find out where you actually stand. This means an honest look at which AI tools are already in use across the business (including the ones nobody officially approved), what rules — if any — already exist, and whether anyone has been trained. Most businesses are surprised by what this phase reveals, usually in the direction of "more scattered use than we realized." This phase is also the fastest of the five — most businesses can complete it in a single sitting, since it's about gathering an honest picture rather than building anything new.
Our free AI readiness scorecard is built for exactly this phase — twenty questions across rules, tool use, training, and measurement, scored into a band that tells you honestly where you're starting from.
Phase 2: Set rules
With a clear picture of where you stand, the next phase is writing down what's allowed and what isn't. This doesn't need to be a long document — a written data rule, an approved tools list, and a named point of contact cover the essentials for most small businesses. The goal is a foundation solid enough that training and automation can build on it safely. Most small businesses can draft this in an afternoon; the harder part is usually deciding who owns keeping it current, which is worth settling explicitly rather than leaving to whoever happens to write the first draft.
Our AI acceptable use policy guide covers the eleven sections a working policy should include, with a copy-ready starter template.
Phase 3: Train
Rules only work if people know them and know how to apply them to their actual job. This phase is where a written policy turns into behavior — a training session covering the rules, hands-on practice with approved tools, and role-specific guidance for each function on the team. Thryv's 2026 survey found 70% of small business owners say they need more, or significantly more, training to use AI productively, so this phase deserves real time, not a rushed afternoon.
See our guides on training employees on AI and what each role should learn for a full plan.
Phase 4: Automate
With rules and training in place, it's time to point AI at a real, recurring task and actually remove work from someone's plate — not just draft the occasional email faster. Pick one well-defined task (the weekly report someone rebuilds every Monday is a common, low-risk starting point), automate it, and measure the result before moving to the next one. Resist the urge to pick the most ambitious use case first — a smaller, well-defined task that clearly works builds the confidence and the internal case for tackling a bigger one next.
Our guides on automating the weekly report and measuring AI ROI cover this phase in detail — including the hours-saved worksheet that makes the result provable rather than anecdotal.
Phase 5: Scale
Once one use case is automated and measured, the final phase is deliberately repeating that success rather than letting AI use stay confined to a single lucky win. This means picking the next task using the same rigor — baseline, automate, measure — and revisiting your rules and training as new tools and use cases get added. Scaling well is what separates the 14% of businesses that report AI fully integrated into their operations from the much larger group still stuck with scattered, unmeasured use. It's also the phase where it pays to loop back to phase 1 periodically — a fresh assessment every six months or so catches new shadow AI use and outdated rules before they become the next gap.
The roadmap at a glance
| Phase | Main output | Guide |
|---|---|---|
| 1. Assess | An honest picture of current AI use, rules, and training gaps | AI readiness assessment |
| 2. Set rules | A written acceptable use policy and approved tools list | Acceptable use policy |
| 3. Train | A team that knows the rules and has practiced with approved tools | Training employees on AI |
| 4. Automate | One real, measured use case saving actual time | Automating the weekly report |
| 5. Scale | A repeatable process for adding the next use case | Measuring AI ROI |
Most small businesses move through all five phases over a few months, not overnight — and that's fine. The point of the roadmap isn't speed, it's making sure each phase has something solid underneath it before the next one starts.
How to tell you're ready for the next phase
Rather than picking an arbitrary timeline for each phase, look for a concrete signal that the current one is actually done — not just started.
- Ready to leave Assess: You can name, specifically, which AI tools are in use, what rules (if any) already exist, and who has and hasn't been trained. Vague answers mean the assessment isn't finished yet.
- Ready to leave Set rules: The policy exists in writing, is easy to find, and at least a few employees have actually read it — not just received it as an email attachment.
- Ready to leave Train: Employees across different roles can describe the data rule in their own words and know who to ask with a question, a month after training, not just on the day of the session.
- Ready to leave Automate: You have a real before-and-after number for at least one use case, not just a general sense that "it's helping."
If a phase doesn't show its signal yet, it's usually better to stay there a little longer than to move on and build the next phase on a foundation that isn't actually solid. Moving on too early is the single most common way these roadmaps stall — not because the plan was wrong, but because a phase was marked done before it actually was.
Where to get help with any phase
You can run every phase of this roadmap yourself using the guides linked above — none of it requires specialized software or a large budget. If you'd rather not build it alone, this exact sequence — assess, set rules, train, automate, scale — is the structure of a Kickstart engagement: yforest AI Labs comes to your company, runs the assessment, helps set the rules, trains your team, and ships your first automated use case, starting from wherever your business is today.
◆ Small Business AI Kickstart
Get AI ready today.
Before it's too late.
yforest AI Labs comes to your company, trains your team, and ships your first tools.
FAQ
Do we have to do all five phases in order?
Yes, roughly. Each phase depends on the one before it — automating before you have rules and training in place just means moving fast without the guardrails that keep it safe.
How long does this roadmap usually take?
Most small businesses move through all five phases over a few months rather than all at once. The pace matters less than making sure each phase has something solid before the next begins.
What if we've already done some of these phases informally?
Start with the assessment phase regardless — it's the fastest way to confirm what's actually solid versus what only feels solid, before you build the next phase on top of it.
Can we automate first and go back for rules and training later?
You can, but it's the most common mistake in this roadmap. Rules and training address the risk that automation actually creates, so doing it out of order tends to cost more time fixing problems later.
Is this roadmap only for businesses starting from zero?
No. A business that's already using AI heavily but without rules or measurement can start at phase 2 — the roadmap works as a check on any point in the process, not just a beginning.
Sources
This guide is general information, not legal advice. Have a qualified attorney review any policy before you adopt it.