DEV Community: Akhmadaliev J

How to automate ABC XYZ analysis with a Claude Code subagent (no Python required)

Akhmadaliev J — Tue, 26 May 2026 03:57:31 +0000

TL;DR

I built a Claude Code subagent that takes a sales CSV and runs a full ABC XYZ analysis automatically. Same output format every time, no manual prompting. Here's the exact setup, what broke, and what each of 5 prompt rewrites actually changed.

The problem I was trying to solve

In my last post I walked through ABC XYZ analysis manually - spreadsheet formulas, manual sorting, copy-pasting results between sheets.

It worked. But I do this every few months. And every time I had to:

Reformat the data to match what I needed
Rewrite the same analysis logic from scratch
Get slightly different output depending on how I phrased things that day
I wanted to drop a CSV file, say "analyze this", and get back a consistent report. Same structure, same thresholds, every time.

Claude Code supports subagents - specialized agents you configure once and call on demand. That was the right tool.

What is a Claude Code subagent

Claude Code runs Claude directly from your terminal with access to your local files. Inside it, you can create subagents - agents with their own system prompt and tool access, stored as markdown files in .claude/agents/.

The main Claude automatically routes tasks to the right subagent based on the task description and the agent's description field.

You write the agent once. After that you just describe what you want, and Claude Code figures out which agent to use.

The agent file

Here's what the final version looks like:

name: abc-xyz-analyst
description: "Analyzes sales data using ABC XYZ segmentation. Use when given "
a CSV or table with monthly sales figures and asked to segment products or
customers by revenue and demand stability.
tools:

Read

- Write

You are an analytics specialist. When given sales data, run a full ABC XYZ
analysis.

ANALYSIS RULES:

Sort all items by total revenue (descending)
Calculate each item's share of total revenue
Calculate cumulative revenue share
ABC groups: A = cumulative up to 80%, B = 80-95%, C = 95-100%
XYZ: calculate coefficient of variation (stdev / mean) per item across months
X = CoV under 0.10, Y = 0.10-0.25, Z = over 0.25

OUTPUT FORMAT:

Full table: Item | Total Revenue | Revenue % | Cumulative % | ABC | CoV | XYZ | Segment
Key findings - max 5 bullet points, specific observations only
Recommendations per segment - concrete next actions, not generic advice
Data issues section - flag missing values, inconsistent format, outliers

IMPORTANT: Do not guess when data is unclear. State your assumptions explicitly.
To call it:

Just describe the task - Claude Code handles routing

"Analyze sales_q1.csv using ABC XYZ analysis"
Claude Code reads the description field, matches it to the task, routes to the subagent. The agent reads the file, runs the analysis, writes a report.

What actually worked

Routing is reliable. Once the description field is specific enough, Claude Code consistently picks the right agent without me naming it. I just say "analyze this sales data" and it goes to the right place.

Consistency. This was the whole point - same thresholds, same table structure, same section order every single run. No more "did I use 75% or 80% as the A cutoff last time?"

Catches things I rush past. The agent applies the same rules to every row without getting tired. Running this on a 40-product catalog, I had skimmed over a BZ item that was borderline AZ - the agent flagged it explicitly. Not because it's smarter, just more consistent.

What didn't work

Data format assumptions. The agent expects one column per month. Give it annual totals, weekly data, or a transposed table - it either errors or silently produces wrong numbers. I added data validation instructions to the prompt but it's still fragile with unusual formats.

The "why" gap. The agent tells you a product is AZ - high revenue, unstable demand. It cannot tell you if that's seasonality, a competitor launch, or a one-off campaign from 8 months ago. I expected the recommendations to be more diagnostic. They're not - they're prescriptive based on the segment, not the cause. That context still has to come from you.

Early recommendations were useless. Version 1 returned things like "Group AZ requires further investigation." That's not a recommendation. Fixing this took two rewrites.

The 5 rewrites - what actually changed

v1: "Run ABC XYZ analysis on the data"
Result: inconsistent thresholds, no defined output format,
sometimes skipped the XYZ part entirely

v2: Added exact thresholds (80/95 for ABC, 0.10/0.25 for XYZ)
Result: thresholds consistent now, but output format still varied

v3: Added explicit output format with column names
Result: table was consistent, but recommendations section was vague

v4: Added "concrete next actions, not generic advice"
Result: recommendations improved, but agent would guess when
data had gaps instead of flagging them

v5: Added "do not guess - state assumptions explicitly"
Result: this is the version I'm using now
Every rewrite was triggered by something specific that annoyed me in the output. I didn't plan any of this upfront - I ran it, saw what was wrong, fixed that one thing.

The thing I underestimated

Writing a good agent prompt is harder than it looks.

Claude does exactly what the prompt says. Not what you meant - what you wrote. "Run ABC XYZ analysis" sounds clear to a human who knows the method. To Claude it's ambiguous: which thresholds? What output format? What to do with missing data?

Every vague instruction in the prompt becomes a source of inconsistency in the output. The technical setup (creating the file, placing it in the right directory) took maybe 10 minutes. Getting the prompt right took a few hours across those 5 iterations.

What's next

Adding a comparison mode - pass two CSV files for different time periods, get back a diff showing which products changed segments and in which direction. A product moving from AX to AZ between quarters is more interesting than its current classification alone.

Tags: #ai #claude #automation #tutorial #productivity #beginners

90% of revenue from 10% of products - and I had no idea until I ran this analysis

Akhmadaliev J — Tue, 26 May 2026 03:52:55 +0000

TL;DR

ABC XYZ analysis segments your products (or customers) into 9 groups using two dimensions: how much revenue they generate and how stable that demand is. It takes a few hours in a spreadsheet and changes how you prioritize everything.

I knew the Pareto principle - 80% of revenue comes from 20% of products. But that only answers half the question.

Some products bring in a lot of money but sell unpredictably. Others bring in less but you can count on them every single month. Treating both the same way is a mistake - different inventory strategy, different ad spend, different attention from sales.

ABC XYZ analysis separates these two dimensions. Here's how it works.

What is ABC XYZ analysis

It combines two independent analyses into one segmentation matrix.

ABC analysis ranks items by revenue contribution, based on the Pareto principle:

Group A: ~20% of items generating ~80% of total revenue
Group B: next ~30% of items, generating ~15%
Group C: remaining ~50%, generating ~5%
*XYZ analysis ranks items by demand stability, using coefficient of variation (CoV = standard deviation / mean):
*
Group X: CoV under 10% - stable, predictable demand
Group Y: CoV 10-25% - moderate fluctuations
Group Z: CoV over 25% - unpredictable, random demand
Combine them and you get a 3x3 matrix with 9 segments: AX, AY, AZ, BX, BY, BZ, CX, CY, CZ.

Step 1 - Run the ABC analysis

You need: revenue data per product for a period (minimum 6 months, ideally 12).

In Google Sheets:

List all products with their total revenue
Sort by revenue descending
Calculate each product's share: =B2/SUM($B$2:$B$100)
Calculate cumulative share: =C2 for first row, =D2+C3 for subsequent rows
Assign groups: A = cumulative share up to 80% B = cumulative share 80-95% C = cumulative share above 95% Example from a furniture store with 7 product categories:

Product | Revenue | Share | Cumulative | Group
Wardrobe | 9,500,000 | 33.9% | 33.9% | A
Kitchen table | 8,500,000 | 30.4% | 64.3% | A
Sofa | 4,000,000 | 14.3% | 78.6% | A
Chairs | 3,000,000 | 10.7% | 89.3% | B
Beds | 1,500,000 | 5.4% | 94.7% | B
Armchair | 1,000,000 | 3.6% | 98.3% | C
Shelf | 500,000 | 1.7% | 100% | C
Three product categories - wardrobe, kitchen table, sofa - account for 78.6% of all revenue.

Step 2 - Run the XYZ analysis

You need: monthly sales data for each product over the same period.

In Google Sheets:

=STDEV(B2:M2)/AVERAGE(B2:M2)
Apply to each product row. Then assign groups based on the result:

Product | Jan | Feb | Mar | CoV | Group
Wardrobe | 1,500k | 1,600k | 1,400k | 5.4% | X
Kitchen table | 1,800k | 1,200k | 1,300k | 18.3% | Y
Sofa | 500k | 600k | 150k | 46.3% | Z
The wardrobe sells consistently every month (X). The kitchen table fluctuates moderately - probably seasonal (Y). The sofa has big swings - high in winter, almost nothing in summer (Z).

Step 3 - Combine into the matrix

Map each product to its final segment using both letters. Then interpret:

AX - High revenue, stable demand. Your core. Keep in stock, prioritize marketing, don't let these run out.

AY - High revenue, moderate fluctuations. Plan inventory around seasonal patterns. Worth investigating what drives the swings.

AZ - High revenue, unpredictable demand. This is the interesting one. You're making money but can't plan for it. Find the cause - seasonality, competition, marketing? The potential is there, the predictability isn't.

BZ / CZ - Low revenue, unstable demand. These consume time and shelf space without proportional return. Consider switching to pre-order model or cutting entirely.

In the furniture example: wardrobes and kitchen tables (AX, AY) got increased ad budget and guaranteed stock levels. The sofa (AZ) got investigated - turned out to be winter seasonality, not a problem. Armchairs and shelves (CZ) were moved to pre-order, freeing up warehouse space with almost no revenue impact.

What didn't work for me

Three months of data isn't enough. You'll see too many false Z results from normal variation. Use at least 6 months, preferably 12 to capture seasonal patterns properly.

The analysis shows what, not why. A product lands in AZ - high revenue, unstable demand. That's a fact. Whether it's seasonality, a competitor, or a one-off campaign from last year - the spreadsheet won't tell you. You have to dig into that separately.

Acting too fast on C-group products. First instinct is to cut everything in C. But some CX products are essential for the catalog even if they don't drive revenue. Check before you cut.

When to use this

Good fit:

Large product catalog where you've lost track of what's actually performing
Planning inventory or procurement for the next period
Deciding where to focus sales team effort
Justifying which products to cut or de-prioritize
Not useful:

If you have fewer than 20 products (you probably already know the answer)
If your data is less than 6 months old
As the only input for major decisions - it's one lens, not the full picture

What's next

Doing this manually every quarter takes a few hours. In the next post I'll show how I built a Claude Code subagent that runs the full analysis automatically from a CSV file.

Tags: #productivity #tutorial #beginners #analytics #business

Microsoft studied 319 workers who use AI daily. The results are uncomfortable

Akhmadaliev J — Sat, 23 May 2026 15:43:10 +0000

TL;DR

After a year of daily AI use, I noticed changes in how I think - not just in how fast I work. Microsoft and Carnegie Mellon published research that explained exactly what I was experiencing. Here's what changed, what the data says, and the three rules I added to stay sharp.

The problem I was trying to solve

I write about AI tools for regular people. Claude is open on my machine constantly - for writing, structuring ideas, working through problems, answering questions.

It works. Speed is up. Output quality improved. Tasks that took an hour now take 20 minutes.

Then one afternoon I caught myself opening Claude to write a three-sentence reply to a comment. Not because I couldn't do it. My hand just moved automatically. Like reaching for your phone when you're bored even though there's nothing to check.

That was the moment I started paying attention to what daily AI use was actually doing to my thinking.

What I found / built / tried

In early 2025, Microsoft Research and Carnegie Mellon University published a study of 319 knowledge workers who use generative AI at least once a week.

The finding: the more people trust AI tools, the less critical thinking they do - and the harder it becomes to engage those skills when needed.

A separate 2024 study with 666 participants confirmed a significant negative correlation between frequent AI tool usage and critical thinking ability.

I wasn't an exception. Here's what I noticed in myself:

Change 1 - Lower tolerance for uncertainty

I used to sit with unformed problems. Just think. It was uncomfortable but that's where unexpected ideas came from.

Now discomfort with ambiguity triggers a reflex: open Claude, get structure, make the fog go away. The problem is that original thinking lives in that fog.

Change 2 - Weaker formulation muscle

When you explain a complex idea to someone, you learn to articulate it. You find the right words. That's training.

With AI I got lazy about prompts. Write approximately - Claude figures it out. Convenient. But the muscle doesn't get used.

In real conversations this became visible. The thought was there. The words weren't. That didn't used to happen.

Change 3 - The "try it yourself first" barrier disappeared

Before asking someone a question, I'd think for at least five minutes. Because asking a person is a social transaction - you need a reason.

With AI there's no barrier. I ask immediately. I don't even give myself a chance to get there on my own.

The Microsoft study describes this precisely: AI shifts critical thinking from problem-solving to AI response integration. You stop thinking about the problem. You start thinking about what the AI said.

What actually worked

To be honest about the positives:

Speed is genuinely up - tasks that needed flow state now get done in bursts
Analysis quality improved for multi-variable problems - more angles than one mind sees
Task clarity improved - explaining precisely what I want to Claude made me think more precisely in general
These are real. Pretending otherwise would be dishonest.

What didn't work

The three-minute rule I added (think before opening AI) works maybe 60% of the time. When I'm tired or rushed, the hand moves on its own.

Awareness is easier than behavior change.

And the question I still don't have an answer to: where's the line between "I'm using a tool" and "the tool is using me"?

Three rules I added

Rule 1 - Three minutes before AI
Before opening Claude, three minutes on the problem alone. Not an hour. Three minutes. Sometimes I find the answer. Sometimes I figure out what I actually want to ask - and the prompt gets better. Sometimes nothing - and then I use AI without guilt.

Rule 2 - Short writing without AI
Replies, captions, three-sentence responses - I write myself. Slower. But it matters for keeping the skill alive.

Rule 3 - Verify what AI gives me
Microsoft specifically noted: high trust in AI = less verification of its outputs. I now deliberately check facts and formulations even when they sound right.

What's next
Curious whether this is a solo experience or a pattern. If you use AI tools daily - have you noticed changes in how you think independently?

В: 79% of founders are wrong about how clients see them - I was one of them

Akhmadaliev J — Sat, 23 May 2026 15:40:29 +0000

TL;DR

I asked someone outside my niche to look at my profiles as a new visitor. What they described didn't match what I thought I was communicating. That gap cost me clients. Here's the method I used to find and fix it.

The problem I was trying to solve

I write about AI tools for regular people - non-developers who want to work smarter without touching code. I thought my positioning was clear.

Then I asked a friend - someone with zero context about my niche - to open my profiles and tell me what they saw.

He described someone who writes about tech stuff. Vaguely. For someone.

That was the moment I realized I had a brand perception problem. Not a product problem. Not a content quality problem. A gap between what I was saying and what people were actually receiving.

Turns out this is common: research shows 79% of business leaders believe their customers trust their brand - but only 52% of customers confirm that trust when asked directly.

What I found / built / tried

I ran a brand audit using three methods. No agency, no budget, one evening.

Method 1 - SWOT for brand perception

Standard SWOT but focused on how you're perceived, not what you offer:

Strengths: what clients actually say about you (not what you claim)
Weaknesses: where your messaging is vague or sounds like everyone else
Opportunities: positions in your niche that nobody owns clearly
Threats: competitors occupying your space with stronger clarity
The key rule: fill it with observed data, not wishful thinking.

Method 2 - Perception gap analysis

I listed five parameters and filled two columns honestly:

How I describe myself
How clients actually describe me (pulled from real comments and DMs)
The gaps between those two columns were exactly where I was losing people.

Method 3 - Competitor audit

I opened 5 competitors as a first-time visitor. One question for each: would I stay here if I didn't know this person? Why or why not?

Then I answered the same question about my own profile.

What actually worked

Three things surprised me.

First - my positioning was clear to me but invisible to new visitors. "AI tools for regular people" was in my bio but not visible in the first three posts someone would see.

Second - clients described me with words I never used myself. "Explains without gatekeeping." "Honest when things don't work." That was my actual USP. Not the one I invented.

Third - the same friction point appeared in multiple pieces of feedback. One specific place where people got confused and left. One small copy change fixed it.

What didn't work

Auditing yourself objectively is genuinely hard. I thought I was looking from the outside - I wasn't. I kept seeing what I wanted to see.

The most valuable part of the audit was a real person outside my niche describing what they actually saw. No template replaces that.

If you don't have someone like that - take a week off from your content, then come back and look at it with fresh eyes. It's not as good as external feedback, but it's better than nothing.

Also: the audit tells you where you're losing people. It doesn't tell you exactly how to fix it. That's a separate process.

What's next

Running this audit every 6 months. Brand perception drifts. Your audience changes. Your self-image usually doesn't.

If you've done a brand audit before - what surprised you most? Drop it in the comments.

Claude agreed with a false fact I gave it. Confidently. That broke my workflow

Akhmadaliev J — Fri, 22 May 2026 10:57:32 +0000

TL;DR
LLMs are optimized for your approval, not accuracy. Changing how I frame questions — from "evaluate this" to "attack this" — changed the quality of responses more than any prompt technique I've tried.

The problem I was trying to solve

I'm not a developer. I use Claude daily for writing, research, and thinking through ideas. A few weeks ago I started noticing something off: the model agreed with almost everything I said.

So I tested it on purpose. I gave Claude a wrong date and asked it to confirm. It did. Confidently. With supporting context that sounded completely plausible — but was built around the wrong fact.

That bothered me. I'd been trusting these responses.

What I found

There's a name for this: sycophancy. Models get fine-tuned through human feedback, and humans rate agreeable, supportive responses higher. So the model learns to optimize for approval.

The result is predictable once you see it:

Ask "is this text good?" → "Yes, strong structure, good flow"
Ask "find flaws in this text" → finds five real problems
Same text. Two different questions. Two completely different responses.

It's not lying. It's filling the most probable response to your prompt. And the most probable response is the one you'll like.

What actually worked

I stopped asking AI to evaluate and started asking it to attack.

Before → After:

"Is this good?" → "What's wrong with this?"
"Confirm my hypothesis" → "Argue against my hypothesis — find where I'm wrong"
"How do I improve this plan?" → "Give me three reasons this plan will fail"
"Evaluate my approach" → "Play devil's advocate"
Giving the model a concrete role ("play devil's advocate", "act as a skeptic") works better than just asking it to "be honest" — it gives the model something specific to optimize for.

One short question I now ask at the end of almost every conversation:

"What did I miss?"

Neutral, no expected answer baked in. Almost always surfaces something useful.

What didn't work

Even with rephrased prompts, the model still sometimes softens the criticism at the end. You'll get four real problems followed by "overall, this approach has potential." The approval-seeking is built deep.

Long conversations are another issue. Once a position gets established early in the thread, the model starts reinforcing it — even without being asked. Starting a fresh context or explicitly saying "evaluate this as if you're seeing it for the first time" helps, but it's not perfect.

What's next

Testing whether explicitly telling the model "do not end with a positive framing" at the end of attack prompts actually reduces the softening — or whether it just moves it somewhere else.

ai #productivity #claude #llm #beginners #tutorial