How to implement AI in your company (without being Google)

Most conversations about AI start in the wrong place: with the technology. Companies hear about large language models, see a demo, and immediately ask "Where can we use this?" That's backwards.

The right question is: "What are the most painful, high-frequency problems in our business?" — and then asking whether AI can solve them better than anything else.

Step 1: Identify the right use cases

Not all problems are a good fit for AI. The best candidates share three characteristics:

• High frequency — happens dozens or hundreds of times per day
• Repetitive structure — follows a predictable pattern, even if the content varies
• Costly to get wrong — errors have real consequences (lost time, money, or trust)

Classic examples that work well for SMBs:

• Drafting first versions of customer responses or proposals
• Classifying support tickets and routing them to the right team
• Extracting structured data from unstructured documents (invoices, contracts, emails)
• Summarizing long documents for executive review

Step 2: Pick a model and understand its limits

We primarily work with the Claude API (by Anthropic), for three reasons:

1. Long context window — Claude can process entire documents in a single call
2. Instruction following — it reliably follows structured prompts and output formats
3. Safety by default — conservative defaults make it easier to trust in production

Here's a minimal example of calling Claude to classify support tickets:

import Anthropic from "@anthropic-ai/sdk";
 
const client = new Anthropic();
 
async function classifyTicket(ticketText: string): Promise<string> {
  const prompt =
    "Classify this support ticket into one of: billing, technical, general.\n" +
    "Ticket: " + ticketText + "\n" +
    "Return only the category name.";
 
  const message = await client.messages.create({
    model: "claude-opus-4-6",
    max_tokens: 64,
    messages: [{ role: "user", content: prompt }],
  });
  const block = message.content[0];
  return block.type === "text" ? block.text.trim() : "general";
}

Step 3: Build a feedback loop from day one

The biggest mistake we see is deploying an AI feature and treating it as "done." AI systems degrade silently — the world changes, your data changes, and the model's responses drift.

Build logging from the start:

Every AI interaction should be logged with: input, output, timestamp, user, and a flag for whether the result was used or corrected. This is your future training data and your quality dashboard.

A simple structure to capture this in a database:

CREATE TABLE ai_interactions (
  id          UUID PRIMARY KEY DEFAULT gen_random_uuid(),
  created_at  TIMESTAMPTZ NOT NULL DEFAULT NOW(),
  model       TEXT NOT NULL,
  prompt      TEXT NOT NULL,
  response    TEXT NOT NULL,
  used        BOOLEAN,
  corrected   TEXT
);

Reviewing the logs weekly

Set a recurring 30-minute block every week to review a sample of AI interactions. Ask:

• Is the model making errors it wasn't making last month?
• Are users correcting the output frequently?
• Has the underlying prompt drifted from what actually works?

This discipline — more than any technical sophistication — is what separates AI implementations that compound in value from those that quietly break.

Step 4: Expand intentionally

Once your first use case is stable and delivering value, you have organizational credibility to expand. But don't rush into the next shiny use case. Instead:

1. Harden the first one — add monitoring, handle edge cases, document the system
2. Extract the patterns — what did you learn about building AI features in your specific stack?
3. Identify the next high-value problem — apply the same evaluation framework

The bottom line

AI adoption for SMBs is less about technology and more about discipline. The companies that get real value from AI are not the ones with the biggest models — they're the ones with the clearest problems, the tightest feedback loops, and the patience to iterate.

If you'd like help identifying where AI can have the most impact in your business, get in touch — that's exactly what we do.