Technology — Kevin Ziegler (Data, Permissions & Governance)

Republished with permission from the FrUn SMB AI Guide. Canonical chapter: Superhuman Docs → · All chapters

Kevin Ziegler · Fractional CTO, KMZ Consulting · LinkedIn

1. Mindset — What an SMB Should Believe Before Spending a Dollar on AI

The belief I would want every SMB owner to get right is this: AI is not a strategy. It is leverage on a strategy, a process, or a decision you already understand.

That sounds simple, but it is where I see people waste the most time and money. They buy an AI tool hoping it will create clarity. It usually does the opposite. If your sales process is unclear, AI will help you create unclear emails faster. If your customer data is a mess, AI will help you summarize the mess. If nobody owns technology decisions, AI will give five different employees five different ways to solve the same problem.

In the CTO seat, AI helps most when it is pointed at repeatable work: documenting processes, summarizing meetings, drafting requirements, comparing vendors, writing first-pass policies, organizing support issues, analyzing spreadsheets, creating training material, and turning messy notes into usable operating procedures. It is very good at moving work from blank page to strong first draft. It is also useful as a thought partner when an owner needs to pressure-test a decision: Should we replace this system? What questions should we ask this vendor? What data do we need before making this call?

Where it only feels productive is the demo-heavy stuff: building a chatbot before you have clean answers, asking AI to make dashboards from unreliable data, connecting every app because it sounds modern, or spending weeks comparing models when nobody has agreed on the workflow. That is motion, not progress.

Before AI pays off, three things need to be true inside the business. First, someone has to own the workflow. AI projects fail when everyone is excited and nobody is accountable. Second, the business needs at least a rough version of the process written down. It does not have to be perfect, but you need to know what good looks like. Third, the source material has to be findable: customer lists, SOPs, pricing rules, vendor contracts, support history, project notes, and policies should live somewhere more reliable than one person’s memory.

For most SMBs, the first AI project should not be ambitious. It should be useful. Pick the recurring task that makes the owner or operator say, “I cannot believe I am still doing this manually.” Start there.

2. The SMB Tool Stack — Start Cheap, Stay Simple

For a small business, the right AI stack is usually three things: one general AI assistant, the AI already inside your office suite, and one simple automation tool. You do not need an enterprise AI platform to start. You need a controlled place where the team can work, a way to keep AI close to documents and email, and a way to move information between systems without custom software.

Start with one general AI assistant. My short list would be ChatGPT Business or Claude Team. As of June 2026, Claude Team is currently listed at $20 per seat per month if billed annually, or $25 monthly, with a team range of 5 to 150 users. ChatGPT Business pricing changes by plan and market, but historically lands in the same rough SMB range. What this replaces or augments: blank-page writing, process documentation, meeting synthesis, spreadsheet explanation, requirements drafting, vendor comparison, and first-pass policy work. Do not buy seats for everyone on day one. Buy seats for the people who actually write, decide, analyze, or train others.

Second, use the AI already connected to your productivity suite. If you live in Microsoft 365, look at Microsoft 365 Copilot Business. Current public pricing (June 2026) shows promotional annual pricing around $18 per user per month and monthly pricing around $25.20 per user per month, with a qualifying Microsoft 365 plan required. If you live in Google Workspace, Gemini is now built into the Workspace tiers; Business Standard is listed at $14 per user per month after the current promotional period and includes Gemini across Gmail, Docs, Meet, and more. What this replaces or augments: email drafting, meeting notes, document summaries, file search, spreadsheet help, and internal knowledge retrieval.

Third, add a lightweight automation layer only after you have one workflow worth automating. Zapier is the usual SMB starting point because it connects a large number of apps and does not require a developer for basic workflows. Its Professional plan is listed from $19.99 per month. What it replaces or augments: copying lead forms into a CRM, sending follow-up reminders, creating tasks from emails, routing support requests, posting notifications to Slack or Teams, and keeping a simple spreadsheet or database updated.

Tools to postpone: enterprise data platforms, custom internal chatbots, agent platforms that require a dedicated admin, and anything sold primarily through a long enterprise contract. Also postpone model-by-model optimization. For most SMBs, the bottleneck is not whether one model is 8% better. The bottleneck is that nobody has written down the process, cleaned up the permissions, or decided who reviews the output.

The right time to bring in a fractional CTO is usually not the first week you hear about AI. It is when the questions become cross-functional: Which tools should connect to customer data? What should employees be allowed to upload? Should we automate this workflow or fix the process first? How do we keep five departments from buying five overlapping tools? That is when a few hours a month of technical leadership can save a lot of rework.

A free or near-free starting point: use the free tier of ChatGPT, Claude, Gemini, or Copilot Chat, but keep sensitive customer, employee, legal, and financial data out of personal accounts. Pick one public, non-sensitive workflow such as rewriting a job description, drafting an SOP from generic notes, summarizing a public vendor proposal, or creating a checklist from a meeting agenda. Learn the habit before you connect the business.

3. The First 30 Days — Where to Start

The highest-ROI first workflow for most SMBs is an AI-assisted operating knowledge base: take the questions the owner, office manager, or team leads answer repeatedly and turn them into reviewed, searchable procedures.

This is not glamorous, which is why it works. Every small business has tribal knowledge: how to onboard a customer, how to quote a job, how to handle a refund, how to set up a new employee, how to escalate a tech issue, how to use the CRM, how to close out a project. AI is excellent at turning scattered notes, call transcripts, emails, and screen recordings into a first draft. The human still decides what is true.

Week 1: Pick the workflow and collect the raw material. Choose one area where repeated questions are costing real time. I would start with either customer onboarding, employee onboarding, or support/request intake. Create a simple folder called “AI Working Drafts.” Drop in 5 to 10 examples: old emails, checklists, screenshots, call notes, forms, templates, or Loom transcripts. Make a list of the top 20 questions people ask about that workflow.

Week 2: Generate first drafts, then review them like an operator. Use your AI assistant to turn the material into SOPs, checklists, templates, and decision rules. Ask for plain language. Ask it to identify missing steps, unclear owners, risks, and handoffs. Then have the person who actually owns the work review each draft. The owner/operator should not be polishing commas. They should be answering: Is this true? Is this how we want the business to run? Who owns each step? Where does judgment enter?

Week 3: Put the content where people already work. Do not create a beautiful knowledge base nobody opens. If the team lives in Google Drive, put it there. If they live in SharePoint, put it there. If they use Notion, ClickUp, Asana, or a shared CRM, put it there. Add a simple naming convention: “SOP - Customer Onboarding,” “Checklist - New Hire Setup,” “Template - Vendor Evaluation.” Then pilot it with two or three employees and collect the questions they still ask.

Week 4: Add one automation and one review cadence. The automation should be boring: a form submission creates a task, a customer onboarding email creates a checklist, a support request gets summarized into a ticket, or a weekly digest lists unresolved items. Then schedule a 30-minute monthly review. AI documentation decays if nobody owns it. Your review question is simple: What changed in the business that the procedure has not caught up with?

Good at the end of 30 days looks like this: one workflow is documented, reviewed, and used by real employees; at least 10 recurring questions now have standard answers; one simple automation removes manual copying or chasing; and the owner can point to specific time saved. If you cannot name the workflow, owner, documents created, and hours saved, you did an AI experiment, not an implementation.

If the 30-day pilot works, the next decision is whether AI stays a useful team habit or turns into scattered experimentation. That is the natural checkpoint to review the pilot, choose the next two workflows, clean up permissions, define tool standards, and decide what should remain human-owned before the business scales usage.

4. A Real SMB Use Case

A relatable example: a 20-person B2B services company with a working owner, an operations lead, a handful of field or delivery staff, and a small admin team. The company had grown past the point where the owner could keep being the answer desk, but they had not built much internal infrastructure. Their systems were typical SMB systems: Google Workspace, a CRM, QuickBooks, shared spreadsheets, and a few vendor portals.

What was broken was not one dramatic technology failure. It was operational drag. New employees asked the same setup questions. Customer onboarding varied depending on who handled it. The ops lead was copying information from forms into tasks. The owner was still reviewing too many routine emails, vendor decisions, and customer exceptions because the rules lived in his head.

AI did the first-draft and synthesis work. We fed it non-sensitive examples: anonymized onboarding emails, old checklists, sales-to-operations handoff notes, and a few meeting transcripts. It produced draft SOPs, customer onboarding templates, a new-hire technology checklist, a vendor comparison worksheet, and a support/request intake form. It also summarized recurring issues from notes and grouped them into categories: access problems, missing customer data, unclear handoffs, billing questions, and exceptions needing owner approval.

The owner and ops lead did the judgment work. They corrected the SOPs, made decisions where the old process was inconsistent, named who owned each step, and marked which exceptions required approval. That human review mattered. AI could infer a process from examples, but it could not decide whether the inferred process was the process they wanted to keep.

The tools were simple: Google Workspace, one paid AI assistant seat during the buildout, and Zapier for the first automation. No custom software. No data warehouse. No enterprise platform. The leadership role was not to run every prompt or turn the project into a technology program. It was to frame the workflow, keep the tool choices simple, identify the risk points, and make sure the owner and ops lead had a structure they could keep using after the first month. The cost was roughly 12 to 15 hours of internal time over a month, a paid AI subscription in the $20 to $30 per month range, and an automation plan around $20 per month once they had something worth automating.

The result was not a press-release AI transformation. It was better than that: the owner stopped answering a meaningful chunk of repetitive questions, onboarding became more consistent, and the ops lead had a single place to send people instead of rewriting the same guidance. The business recovered several hours a week and reduced avoidable errors in customer handoff. That is the kind of AI ROI SMBs should be looking for first.

5. Guardrails — What an SMB Should NOT Do

Do not connect AI to your whole email, drive, CRM, or accounting system on day one. Start with narrow access and known documents. The convenience of broad access is real, but so is the risk of exposing payroll files, customer contracts, health information, legal disputes, or acquisition documents to employees who should not see them.

Do not let employees use personal AI accounts for business-sensitive work. This is one of the most common SMB mistakes. A salesperson pastes a customer contract into a personal chatbot. A manager uploads a compensation spreadsheet. A coordinator summarizes a medical or legal document. Nobody meant to create risk, but the business now has no central control, no offboarding, and no policy trail.

Do not launch a customer-facing chatbot unless the source content is accurate, reviewed, and maintained. A chatbot on top of stale policies is just a fast way to give customers the wrong answer. If you want a customer-facing bot, start with limited topics: hours, location, service area, appointment scheduling, order status, or routing to a human. Keep pricing exceptions, legal claims, financial advice, medical advice, and warranty commitments under human review.

Do not automate approvals that carry financial, legal, HR, safety, or customer relationship risk. AI can draft the refund response, summarize the dispute, or compare the contract terms. A person should still approve refunds above a threshold, terminations, salary decisions, legal language, collection escalations, security exceptions, and anything that materially changes a customer commitment.

Do not use AI-generated code or formulas in production without review. This applies even if the tool sounds confident. AI can write useful scripts, spreadsheet formulas, and website snippets, but small errors can become billing mistakes, broken integrations, or security holes. Back up the file, test on sample data, and have someone competent review it before relying on it.

The privacy issue SMBs underestimate is permissions. Most breaches of judgment are not hackers in hoodies. They are normal employees getting access to the wrong folder, the wrong transcript, or the wrong connected app. Before you connect AI to company knowledge, clean up shared drives, remove ex-employees, require MFA, and decide what categories of data are off limits.

6. Lessons Learned + One Thing You’d Tell Every SMB Owner

• AI rewards businesses that are willing to be boring first. Clean folders, named owners, reviewed procedures, and basic permissions are not exciting, but they are what make AI useful.
• One implemented workflow beats ten clever demos. If AI saves the owner three hours every week, that is a win. Do that before chasing bigger ideas.
• The best AI use cases usually start as frustration, not innovation. Listen for the sentence, “Why are we still doing this manually?” That is often your starting point.
• Human review is not a temporary crutch. In an SMB, the owner and managers hold context the tools do not. Keep humans on judgment, exceptions, relationships, and risk.
• Do not buy complexity before you have adoption. A simple tool used weekly will outperform a sophisticated platform nobody trusts or understands.
• Bring in fractional CTO help when the AI questions stop being about prompts and start being about architecture, permissions, vendors, integrations, governance, or cross-functional trade-offs.

The one thing I would tell every SMB owner is this: start with the work you already understand, use AI to remove the repetitive drag around it, and only scale once you can point to a real workflow, a real owner, and real time saved.

Part of the FrUn SMB AI Guide. Read on canonical guide →