Technology — Ben Shichman (The Starter Stack & Quick Wins)

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

Ben Shichman · Fractional CTO | AI & Technology Advisor · LinkedIn

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

Every owner I talk to eventually asks the same question: “What AI should I buy?”

I usually answer with another question. What are your employees doing every day that they shouldn’t have to do anymore?

That’s where AI belongs.

I’ve spent more than 30 years helping companies use technology to solve business problems. AI is the latest tool, but the fundamentals haven’t changed. Technology doesn’t fix broken processes. If your customer data is scattered across five systems, or nobody follows the same sales process, AI won’t solve that. It will simply help you do the wrong things faster.

The companies getting the most value out of AI aren’t replacing people. They’re taking repetitive work off their plates. Drafting proposals. Summarizing meetings. Pulling information out of spreadsheets. Writing documentation. Creating reports. Even helping developers write code. That’s where I consistently see businesses save time.

Then there’s the work that feels productive but isn’t. Open-ended “research” with no decision attached. Tuning a chatbot for two weeks. Building an automation for a problem you don’t actually have. The test is simple. If the session ends with something a person uses, it was real work. If it ends with a feeling of progress, it wasn’t.

Before you spend a dollar, three things have to be true. Someone owns the process you’re improving. The information that process needs is reasonably organized. And you can say in one sentence how you’ll measure success. You don’t need perfect data or written procedures. You need to be able to tell when the output is wrong.

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

Don’t overcomplicate this.

Start with one general AI assistant. For a single owner or a small team testing the water, that’s Claude Pro or ChatGPT Plus at $20 a month, one seat. That one tool can help with research, first drafts, spreadsheet analysis, proposals, documentation, and dozens of small jobs that normally eat up part of someone’s day. Learn it well before you chase the next model. When it’s clearly working and you want the whole team on it, move up to a team plan. Claude Team and ChatGPT Business run roughly $25 to $30 a seat and add shared workspaces, admin controls, and a promise not to train on your data. That’s the upgrade. It is not the starting point.

If your company already lives in Microsoft 365 or Google Workspace, look at the AI built into those products first. On Google Workspace, Gemini is now bundled into the paid plans. If you’re on Business Standard at about $14 a seat, you’ve probably already paid for it. On Microsoft 365, Copilot is a separate add-on at roughly $30 a seat, so turn it on for the people who’ll actually use it, not everyone. Either way, AI that lives inside the tools your team already opens every day gets adopted. AI in a separate browser tab gets forgotten.

Once you’ve found one repetitive workflow worth automating, add a connector. Zapier is the easiest at about $20 a month. Make is cheaper at around $10 and more flexible, with a steeper learning curve. Buy this the week you catch yourself doing the same copy-paste for the third time. Not before.

AI has also changed the economics of custom software. Internal tools that used to cost tens of thousands of dollars can often be built for much less today. That doesn’t mean every business should build software. Every application has to be maintained, updated, secured, and supported. If an off-the-shelf product solves your problem, buy it. Build custom software when it gives you a real competitive advantage or when nothing else fits.

Skip the rest for now. Enterprise AI platforms, custom chatbots, agent frameworks, expensive consulting engagements. If a tool needs a dedicated person to set it up and babysit it, it isn’t an SMB tool yet, no matter how good the demo looked. The same goes for the open-source automation tools people will tell you to self-host to save money. The software is free, but software that needs someone to run a server is not free. It’s a part-time job you haven’t hired for. Prove AI is saving real time first, with the simple paid tools. Then decide whether you’ve outgrown them. Most businesses never do.

No budget at all? Start free. The free tiers of Claude and ChatGPT are genuinely capable in 2026, not crippled demos. Use one every day for a week. When you hit the usage limit, that wall is your signal the $20 is worth it.

3. The First 30 Days — Where to Start

The best place to start is almost always the same. It’s the work your one or two most experienced people repeat every day. The proposals they rebuild from scratch. The same customer questions they answer over and over. The procedures that live only in their heads. Capture that and you get two things at once: immediate value today, and a backup plan for the day those people are out.

Week 1: Follow your people around. Write down every repetitive task, and every question they answer more than once.

Week 2: Use your assistant to turn those answers into procedures, templates, and checklists. AI writes the first draft from the notes you took. This part is fast. A week of documenting by hand becomes an afternoon.

Week 3: Test everything with the people who do the work. AI creates the draft. Your employees decide whether it’s correct. Fix what they don’t trust.

Week 4: Automate the one process that wastes the most time today using Zapier or Make.

At the end of thirty days you should be able to point to one workflow that is measurably faster. A task that took an hour now takes ten minutes, and the person who owns it still uses it without you asking. One real success beats ten experiments.

4. A Real SMB Use Case

I worked with a field service business with about twenty-five employees. Customer information lived in Outlook and Excel. Quotes were built in spreadsheets. Scheduling was manual. Most of how the business actually ran lived in the heads of two experienced employees.

The problem wasn’t that they needed AI. The problem was that the business depended on two people remembering how everything worked. New employees asked the same questions every day. Quotes were recreated from old jobs. Customer information was entered multiple times because Outlook, Excel, and scheduling didn’t talk to each other.

We didn’t start by asking an LLM to solve the business. We started by capturing what those two employees knew. I interviewed them about how they quoted jobs, scheduled work, handled customer questions, and dealt with exceptions. We gathered old proposals, customer emails, spreadsheets, and other documents they used every day.

We used an LLM through a conversational interface. In this case, we used the owner’s existing ChatGPT account, but the same approach would work with Claude, Gemini, or another comparable LLM. We used it to turn interview notes, old proposals, customer emails, and existing spreadsheets into first drafts of operating procedures, proposal templates, customer email templates, call summaries, and internal documentation. Along the way, we built a small prompt library for recurring tasks so employees weren’t reinventing the wheel every time. One prompt generated proposals from customer notes, another summarized phone calls into CRM-ready notes, another rewrote customer emails in a consistent tone, and another turned interview notes into standard operating procedures. Every draft was reviewed by the owner before anyone used it. AI accelerated the writing. It never made business decisions. The technology was intentionally simple. Outlook remained the inbox. Excel remained the quoting system. Zapier handled the plumbing between them. New customer emails automatically created rows in the quote tracker so the office staff stopped entering the same information multiple times. The LLM never had direct access to any business systems or customer data. People used it when they needed to create or summarize information. Zapier moved information between systems.

The software costs were minimal. The bigger investment was time. Over the first month, the owner and two key employees spent a few hours each week documenting processes, reviewing drafts, and refining the prompt library until it reflected how the business actually wanted to operate.

The result wasn’t fewer employees. It was that the same people spent less time copying information, rewriting emails, and answering the same questions over and over. Proposals that used to take half a day went out in under an hour. Customer questions had consistent answers. Most importantly, the knowledge that used to exist only in two people’s heads became documentation the entire business could use.

5. Guardrails — What an SMB Should NOT Do

• Don’t put confidential business or customer information into personal, free AI accounts.
• Don’t let AI make financial, legal, HR, or customer commitments. It drafts. A human decides.
• Don’t let every department buy a different AI product. Pick one and standardize on it.
• Don’t put AI-generated code into production without someone who understands it reviewing it first.

Some jobs stay human no matter how good the tools get. The final send on anything a customer reads. Anything a CPA should check before it goes out, or a lawyer should check before you sign it. Any code change that touches payments, customer data, or your live systems. And the judgment call on what’s actually true. The model is confidently wrong on a regular basis, and it never warns you when it’s guessing.

The privacy piece is the one SMBs underestimate most, and it’s the one I spend the most time on. If you handle customer data, you have obligations you may not have thought about. PCI if you touch card data. GDPR or CCPA if you have customers in Europe or California. Free and personal AI tiers can use what you type to improve their models. The business and team tiers, and the API, exclude your data from training by default. That difference stops being academic the moment real customer information is involved. So keep payment data out of these tools entirely, use a paid business tier for anything with customer data in it, and read the data-retention terms once before you commit. And get ahead of shadow AI. Your team is already pasting company information into tools you’ve never approved. One sanctioned tool with one clear rule beats ten secret ones.

Most importantly, don’t mistake activity for progress. The goal isn’t to use AI everywhere. The goal is to improve the business.

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

• Don’t automate a bad process. You’ll just make the mistakes happen faster.
• Your data matters more than your prompts. A clear question against organized information beats a clever prompt against a mess.
• Learn one AI platform well before chasing the next one.
• Treat the model like a fast, overconfident junior employee. Great at first drafts. Never the final word.
• AI should eliminate repetitive work so your people can focus on higher-value work. That’s the whole point.

If I could leave every SMB owner with one thought, it’s this:

Don’t try to transform your business all at once. Find one process that’s wasting time every day. Fix it. Measure it. Then move on to the next one.

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