- Home
- Skills
- Product Management
- business-case
Works with the AI Agents you already use
Business Case
Interviews you about a workflow automation or piece of software before you build it, does the arithmetic out loud, and ends in a build, buy, thin-slice, or don't-build verdict you can defend to someone holding a budget.
Free
business-case
Example session with this skill installed
Should I automate our invoice reconciliation? We get a lot of invoices and it feels like it takes forever. $250k has been floated as an upfront cost.
- Read your context and instructions
- Compiled the business-case
What the skill does instead of writing a plan
It restates the work as a plain sentence and asks you to correct it, then runs a round of numbered questions, each with a recommended answer attached:
❓
Q1 —
Instances per year: How many invoices came through last month, not a typical month? 100, 1,000, or 10,000?
➡️ Guessing ~1,000/month, steady.
❓
Q2 —
Person-hours per instance: How long did the last one take, hands-on, and how much of that was waiting on someone else?
➡️ Guessing 5 min hands-on.
You answer; it posts a running ledger. The ledger either builds the case or collapses it in front of you:
📊 Ledger (likely: 300 h/yr · $13,478/yr, payback ~111 weeks)
| term | low | likely | high | source |
|---|---|---|---|---|
| invoices / yr | 14400 | 14400 | 15840 | stated |
| exception rate | 40% | 35% | 25% | stated (412/1200 counted) |
| min / clean invoice | 2.00 | 2.92 | 3.67 | derived from the average |
| irreducible review, min | 1.5 | 1.0 | 1.0 | stated |
| hours / yr | 72 | 300 | 481 | derived |
The naive version of that case was 960 h/yr and $43k. Two questions took 69% of it off the table. The skill will keep going until every term in the spine has a value and a source, then land on exactly one verdict: build now, build this thin slice, gate the slice on one measurement, buy this instead, don't build, or not enough signal.
Connects securely to your tools. The creator never sees your data.
About this skill
business-case
A skill that interrogates you about a workflow automation or a piece of software before you build it, does the arithmetic out loud, and ends in a verdict you can defend to someone holding a budget.
It is not a scoping tool. It is built to answer one question — does this deserve to exist — and it is willing to answer no.
What it does
- Interviews you in rounds about frequency, person-hours, exception rate, review floors, adoption, upkeep, rot, and the price of whatever already exists.
- Posts a running ledger after every round, with low / likely / high and a source tag on every row (
stated,derived,assumed,unknown). - Lands on exactly one recommendation: build now; build this thin slice; build this thin slice gated on one measurement; buy this instead; don't build; or not enough signal.
- Names the assumption the case rests on, the cheapest way to test it this week, and kill criteria for 90 days out.
How it works
Every question has to move a term in one equation:
recoverable h/yr = instances/yr
× automatable share (1 − exception rate)
× (clean-instance person-hours − irreducible human time per instance)
annual value = recoverable h/yr × loaded hourly cost
− annual upkeep (maintenance + rot)
− annual cost to run (cost per instance × automated instances/yr)
A question that moves no term does not get asked. The skill is also built to avoid common traps that have been worth a large fraction of a real case: blended averages, multiplying people by elapsed time, folding the review floor into "automatable share," unrecapturable savings, phantom headcount, audiences mistaken for populations, and skipping the thing you already pay for.
It asks the buy-versus-build question in round one and, when a sub-agent is available, sends one looking while the rest of the round runs.
Where it stops
- It does not design the system. Value first, shape second. If you push it to the build twice, it writes the shape as a conditional rather than pretend the value question was settled.
- It does not price revenue, retention, or willingness-to-pay bets. If hours come out immaterial and the real case is a product bet, it says the spine does not apply and switches questions instead of computing a rigorous, meaningless number.
- It does not sum non-hour value into the dollar figure. It does force each item to an order-of-magnitude bound.
- Defaults (a Boston tech median loaded rate of about $98/hr, and thresholds such as 40 and 200 gross h/yr) are rules of thumb, not findings. It uses a rate you give it in preference to the default.
The skill is MIT licensed. The round-and-frontier interview mechanic is adapted, with thanks, from Matt Pocock's grilling skill.
Compatibility
Works as a SKILL.md through the Agensi agent connection with Claude, ChatGPT, Cursor, Claude Code, and other agents that can load Agensi skills. No extra runtime, API keys, or local install. When the agent has sub-agents, it uses one to check whether something already on the market does the job; if not, it does that check itself between rounds.
This is not a Claude Code plugin. Agensi delivers the skill text through MCP; there is nothing to install on the machine.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Product Management.
It will not design the build for you. Value first, shape second. If you insist on jumping to design, it will write the shape as a conditional on the value terms still being measured.
It will not invent a revenue case. If recoverable hours are immaterial and the real case is retention or willingness to pay, it says the spine does not apply and switches questions. It will not compute a rigorous, meaningless number.
Non-hour value is bounded, not summed. Faster deals, risk removed, or audit evidence stay in a separate column, each sized to an order of magnitude.
Defaults are starting points. If nobody supplies a loaded rate, it assumes a Boston tech median of $150,000 base and computes about $98/hr. Give it your rate, salary, or location and it will use that instead. Thresholds (40 and 200 gross h/yr, ~12 instances a year, ~0.7 FTE) are rules of thumb and can be moved.
It needs your numbers, not memory of averages. Exception rates and volumes should be counted from last month where you can. If you will not measure, it keeps your number, marks it assumed, and says which way the verdict flips if the count comes back worse.
Is this the same as a Claude Code plugin install? No. The public GitHub repo can also be installed as a Claude Code plugin, but this Agensi listing delivers only the SKILL.md through the agent connection. You do not need to install anything locally.
Why is it free? The skill is already published under the MIT license at https://github.com/Levison/business-case-skill. This listing makes the same skill easy to load through Agensi.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 2 days ago
- Free with an Agensi account