Module 1. The AI transformation landscape

Before you build anything, you need a clear way to talk about what AI changes inside a small company, which Claude product fits which problem, and what evidence backs your claims.

1.1 What AI transformation means for an SMB

For a company of 10 to 200 people, AI transformation is not a moonshot. It is a sequence of small changes to how work gets done, each one measurable. The useful mental model has three levels.

1.2 Where the value is

Value hides in tasks, not job titles. A finance clerk's job contains dozens of tasks; a few of them (matching invoices, chasing late payers, writing the monthly variance comment) are text-heavy, repetitive and rule-guided. Those are your targets.

A quick screen you can run in any client conversation. A task is a good first candidate when it scores yes on at least four of these:

1. It happens weekly or more often
2. Its inputs and outputs are mostly text, documents or tables
3. A good result can be checked by a person in under a minute
4. Mistakes are recoverable (a draft, not a payment)
5. The data it needs is already digital and reachable
6. Someone senior is annoyed by how long it takes now

1.3 Choosing a Claude model

Anthropic ships models in tiers. In October 2026 the public models are Claude Haiku 5.5, Claude Sonnet 5.5 and Claude Opus 5.5, plus Claude Fable 5.1 in the top Mythos tier with extra safeguards. Version numbers change; the tiers stay the same, so think in tiers.

Exact prices, context windows and model IDs change. Always read the current models overview and pricing page before you put a number in a proposal.

1.4 The Claude product surfaces

Clients will hear brand names. You need to map each one to a job.

Skills, plugins, Projects and artifacts are capabilities inside these products, not separate products. "Claude for Financial Services" and similar names are packages of Claude for an industry.

1.5 Plans, data and the first privacy questions

The first three questions every owner asks: "Will it train on our data?", "Who can see what?", and "What does it cost per person?". Have a clear, sourced answer for each.

1.6 Evidence: the Anthropic Economic Index

The Anthropic Economic Index publishes how Claude is actually used, broken down by work task, occupation category and country (Israel included). It shows patterns of use, not job outcomes, and you should present it that way.

Three ways to use it in consulting:

• Prioritise: tasks that already see heavy Claude usage elsewhere are proven territory for a pilot.
• Benchmark: compare Israel's usage index with other countries to frame the opportunity.
• Frame: show the split between augmentation (people working with Claude) and automation (Claude completing the task) for the client's function.

Correct phrasing: "conversations matched to bookkeeping tasks", not "accountants use Claude".

mkdir claude-transformation && cd claude-transformation
python -m venv .venv && source .venv/bin/activate
pip install anthropic python-dotenv
echo "ANTHROPIC_API_KEY=sk-ant-..." > .env   # key from console.anthropic.com, set a monthly spend limit

import os, time, json
from dotenv import load_dotenv
import anthropic

load_dotenv()
client = anthropic.Anthropic()

# Check current model IDs at docs.claude.com before running
MODELS = ["claude-haiku-5-5", "claude-sonnet-5-5"]

TASK = """You are helping a 40-person Israeli logistics company.
Below is a customer email. Return JSON with keys:
intent (quote|complaint|tracking|other), urgency (1-5), language,
summary (one sentence, English), draft_reply (in the customer's language).

EMAIL:
שלום, המשלוח שלנו לחיפה מתעכב כבר שלושה ימים ואנחנו צריכים אותו עד יום חמישי.
מה קורה? יוסי, מחסני הצפון"""

def run(model: str) -> dict:
    start = time.perf_counter()
    msg = client.messages.create(
        model=model,
        max_tokens=600,
        messages=[{"role": "user", "content": TASK}],
    )
    elapsed = time.perf_counter() - start
    return {
        "model": model,
        "seconds": round(elapsed, 2),
        "input_tokens": msg.usage.input_tokens,
        "output_tokens": msg.usage.output_tokens,
        "text": msg.content[0].text,
    }

if __name__ == "__main__":
    results = [run(m) for m in MODELS]
    for r in results:
        print(f"\n=== {r['model']}  {r['seconds']}s  in={r['input_tokens']} out={r['output_tokens']}")
        print(r["text"])
    with open("results_lab1.json", "w", encoding="utf-8") as f:
        json.dump(results, f, ensure_ascii=False, indent=2)

Deliverable: "Claude for SMBs" one-page briefing

A single page you can send to a prospect after the first call. Write it with Claude, but every claim must be yours and sourced. Suggested structure:

1. Headline: one sentence on what AI can do for a company their size this quarter
2. Three levels: assist, augment, automate, each with an example from their industry
3. Evidence: two numbers from the Economic Index, with date and link
4. What it costs and what it protects: plan options and the data-use answer, with links
5. How we start: a two-week readiness audit, what they get, what you need from them
6. About Vastama: three lines and a contact

Save it as a Claude Docs document or a one-page PDF. Keep an English version and plan a Hebrew one.

Self-check

1. A client wants Claude to "handle all supplier emails automatically" from day one. What do you propose instead, and why?
   Model answer

   Start at augment: Claude classifies and drafts, a person approves. Measure accuracy for 2–4 weeks, then automate only the categories that hit the agreed accuracy. This limits risk and builds trust.
2. Which model would you start with for a nightly job that tags 50,000 product descriptions, and which API feature would you use?
   Model answer

   Haiku via the Batch API, after a small eval confirms tag quality. Batch processing is cheaper and suits non-urgent jobs.
3. An owner asks whether Claude trains on their data. What do you say?
   Model answer

   It depends on the plan and terms. Show the current commercial terms and Trust Center pages for the plan you recommend, with the date checked, rather than answering from memory.
4. Name the product surface for each: a team that lives in Slack; a web tool with no API; a custom invoice pipeline.
   Model answer

   Claude Tag in Slack; Claude in Chrome; the Claude API (possibly with the Batch API) in a Python service.
5. Rewrite this sentence correctly: "Economic Index shows lawyers are being replaced by Claude."
   Model answer

   "The Economic Index shows a share of Claude conversations matched to legal tasks, and how much of that is augmentation versus automation." It describes usage patterns, not jobs.

Further reading

• Claude models overview
• Claude API overview
• Anthropic Economic Index
• Claude Help Center (plans, admin, privacy)
• Anthropic Academy (free courses)