How to use the Channel Economics Analyzer agent skill

What the skill makes the agent do

An agent skill is a folder of instructions, reference material and scripts that an AI agent can load when a task matches the skill's description. This one covers reviewing direct versus partner-led channel economics. Its job is to answer three questions: what each channel costs to serve on a fully loaded basis, what each channel returns under three ROI lenses, and what channel mix fits your constraints.

The skill covers costs that are often missed, such as sales staff time, partner discounts, partner marketing funds (MDF), enablement time, support load and allocated overhead. Each channel receives a verdict of DOUBLE-DOWN, MAINTAIN, DEFUND or EXIT. The skill states that humans make the final decision, and it does not reallocate resources on its own.

• Cost to serve per deal and per dollar of ARR, with a 'true gross margin' line after channel-specific load
• ROI per channel under cash year-1, LTV-adjusted and marginal lenses, plus the diminishing-returns inflection point
• A constrained channel-mix recommendation with sensitivity scenarios, such as direct CAC rising 20% or partner discount widening 5 points

Install the skill from the ZIP or the Claude apps

Option one is a local install. Download the ZIP from this page and unzip it into your user skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill available only in that project. Option two is to upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

After installing, check that the folder contains SKILL.md, the scripts/ folder, the references/ folder and assets/channel_data_template.md. If any of these are missing, the skill will not have what it needs.

• Local: unzip into ~/.claude/skills/ for all projects, or into .claude/skills/ inside one project
• Apps: Settings → Capabilities → Skills, then upload the ZIP

Check that Python 3 runs the scripts

The three scripts need Python 3. They use only the Python standard library, so there is no package installation step. Open a terminal, move into the skill folder, and run the built-in sample. This confirms Python works before you enter any real data.

Every script accepts --help, --sample, --input and --output. Use --help to see the options for any script.

Fill in the channel data template

Before running the analysis, copy the data into assets/channel_data_template.md. The template takes about 20 minutes and gives a schema for each input, with notes on what to enter and why. The scripts read JSON, so the template shows the fields you need.

The inputs include deal count and ARR over the trailing twelve months, average deal size, gross margin, CAC, sales-cycle days, retention and expansion rates, partner discount percentage, and attributable costs such as SDR, AE, SE, channel manager, customer success, support, marketing, partner MDF, tooling and overhead allocation percentage.

If you do not know a value, the template says to leave it null or as an explicit '$0 unknown' and note it. The scripts report unknowns openly instead of quietly filling them in.

• Define a channel as a go-to-market motion such as direct, partner-led, marketplace, reseller or OEM, not a marketing source
• Use the same overhead allocation method across all channels
• Enter retention per channel, not pooled
• Count a deal as channel-sourced only if the partner originated the opportunity and brought it unqualified

Run the three analysis scripts in order

Step one calculates cost to serve for each channel. Run it once per channel. Its 'true gross margin' output feeds the next two steps.

Step two calculates ROI under the three lenses and gives each channel a verdict. Step three runs the constrained mix optimizer, which applies limits such as a minimum direct share and a maximum partner concentration. Use --profile to choose an industry profile for the two analyzers. The listed profiles are saas, api, enterprise-software, marketplace and hardware. A profile sets benchmark defaults but does not override your own numbers.

• Step 1: cost_to_serve_calculator.py with --input channel.json
• Step 2: channel_roi_analyzer.py with --input roi.json and a profile
• Step 3: channel_mix_optimizer.py with --input mix.json and a profile

Trigger the skill with natural requests

You do not need to type script names to use the skill. Ask the agent a question that matches its purpose, and it can load the skill and walk you through the intake. The skill's description gives examples of the kind of question it handles, such as 'which channel actually makes money, direct or partner?'.

Good requests name the decision, the channels and the data you have. The agent will ask for any missing inputs before running the scripts.

• 'We are 60% direct and 40% partner-led. Which channel makes money after CAC, support and partner discount?'
• 'Help me prepare the quarterly channel review. Load our channel data and give me verdicts per channel.'
• 'Should we enter the new region direct-first or partner-first? Run the channel economics and show the mix sensitivity.'
• 'The board asked what we got from the MDF spend. Compute the ROI of our partner channel.'

Read the outputs and make the decision

Each report gives you cost to serve, ROI under three lenses, the diminishing-returns inflection point, and a verdict per channel. The mix report includes a sensitivity table. Treat the verdicts as a structured recommendation. The output depends entirely on the numbers you entered, so check the inputs before you act on any verdict.

Take the three reports to the review. The skill recommends and the people in the room commit to the decision.

• Use the sensitivity table to test whether the recommendation holds when key assumptions move
• Compare per-channel LTV and payback instead of pooled figures, since pooled numbers can hide one channel funding another

Safety notes on scripts and network use

The skill runs local Python scripts. According to the skill's documentation, it does not use the network. Review the scripts in the scripts/ folder before running them on sensitive data, as you would with any script you did not write. Keep the channel data files on your own machine, and remember that the skill's reference materials are cited canon, not verified data.

• Runs scripts: yes, Python 3 standard library only
• Uses network: no, according to the skill's documentation
• Reference materials are cited sources, not checked datasets