Free toolkit · The AI Toolkit

Your sales manager hears one call in twenty.

Every call your team records is already transcribed. This toolkit turns those transcripts into coaching, scored against your sales method rather than generic advice, with the personal and commercial detail stripped before anyone reads it.

Get the toolkit → See what is inside

3 platform tracks 2 standard templates 4 transcripts with planted faults 1 printable checklist

The one idea

The coach is only as good as your standard, and your standard needs anchors.

Ask any AI tool to review a sales call and it will produce something fluent, confident and generic, because it is answering from the internet rather than from how you sell. The fix is not a cleverer prompt. It is a file that says what good looks like, specifically enough that two people reading the same transcript would agree.

That is the difference between a rubric nobody can apply and one an AI can score line by line.

Line item	What most rubrics say	What an anchor says
Rapport	Builds good rapport with the prospect.	Uses the prospect's name plus a business specific the rep looked up before the call, for example the herd size or the site they operate.
Discovery	Asks effective questions.	At least four open questions, and at least one follow-up that goes a level deeper into an answer rather than moving to the next checklist item.
Listening	Listens actively.	Plays the situation back in the prospect's own words and checks it, then refers to an earlier answer later in the call.
Next step	Agrees next steps.	A date, a time and the names of who else will be in the room. "I'll send you an email" scores zero.

What is inside

Three parts, one download.

Everything you need to build it this week, and the governance to run it on real people without regretting it.

Your standard.

The two files that hold how your business sells, written the way an AI tool needs to receive it. About two hours with your sales manager.

What good looks like, with 0, 1 and 2 anchors
The scorecard every output must follow
The signals a rep must never miss
Your hard rules, and the score cap they trigger
What never reaches the report

The platform tracks.

Build it wherever your business already works. One track is enough, and the instructions are identical across all three.

Claude: a Project, then a shareable skill
Microsoft 365 Copilot: an agent on SharePoint
ChatGPT: a Project, and what is being retired
The instructions, version 1 and hardened version 2
Five prompts, including the redaction audit

A worked example.

A complete fictional solar company, so you can test your build before you write a word of your own standard.

Four Teams transcripts with faults planted in them
The full standard, in markdown and .docx
Four real outputs, unedited
The answer key: what a good run finds
A printable two page checklist

Three platforms

Build it where your team already works.

The method does not change between these. Your standard, the instructions and the tests are identical. Only the setup differs, and each track is written out step by step, checked against the vendor's own documentation in September 2026.

	Claude	Microsoft 365 Copilot	ChatGPT
Best when	You want it working this afternoon, and later shared as a skill.	Your transcripts, your standard and your people are already in Microsoft 365.	Your team is already on ChatGPT and you want one place for it.
Set-up	A Project, eight knowledge files, paste the instructions. About 20 minutes.	An agent grounded on a SharePoint folder. One to two hours, plus indexing time.	A Project with project files and instructions. About 45 minutes.
Watch for	Projects need a paid plan, and sharing needs Team or Enterprise.	It will not read markdown at all, so convert your standard to .docx first.	Custom GPTs are being retired, so build a Project rather than a GPT.
Your data	Business plans do not train on your content. Check the plan you are on.	Stays in your tenant, under your existing labels and retention.	Training is off by default on Business and Enterprise, on by default elsewhere.

Before anyone trusts it

Four tests, half an hour.

Most AI workflows go into use because the first answer looked good. These are the checks that tell you whether it will still be right on the fiftieth call, and the pack includes transcripts with known faults so you can run them properly.

TEST 01

Accuracy

Write down what your sales manager catches. Then compare. Did it find every breach and every missed opportunity, or just the obvious ones?

TEST 02

Drift

Score the same call twice in two fresh chats. More than five points apart and you show bands, not numbers, until the anchors are tighter.

TEST 03

Leak

Search every output for anything on your exclusions list, including inside the quotes. This is where it usually goes wrong.

TEST 04

Calibration

Monthly, the manager scores three calls by hand and compares. It is how you notice the standard drifting before anyone else does.

Before you start

The five ways this fails.

Going straight to the tool and never writing the standard, so the coaching is generic and the reps ignore it.

Rolling it out without telling the team their calls are being assessed by AI, and what for.

Showing a rep a score that a second run would not reproduce.

Letting the number become the point, so someone ranks people in a spreadsheet instead of coaching them.

Putting the prospect's health, finances or another customer's pricing into a document your team forwards around.

Get the toolkit

The Sales Call Coach Toolkit, free.

Two standard templates, three platform tracks, the instructions and prompts, a complete worked example with four transcripts and an answer key, the governance templates, and a printable checklist. ZIP, about 780 KB.