The MSP AI Build List

Chicago Peer Group · Hosted by Greg

The MSP AI Build List is a fifteen-item, three-tier playbook distilled from a Wednesday morning Chicago peer group of MSP owners — honest conversations about what is working, what is broken, and what they are building. Every item ships with a copy/paste prompt you can drop directly into Claude or ChatGPT, so you leave this page with something you can act on this week. No strategy deck, no committee, no waiting on a vendor.

Context

Why this document exists

Wednesday morning we spent several hours as a group doing something most of us don’t do enough of — slowing down to compare notes. Not vendor pitches. Not keynotes. Just MSP owners being honest about what’s working, what’s broken, and what we’re building.

This document is a distillation of that conversation. Every idea below came from someone in the room. Some of you are further along. Some are just getting started. That’s fine — this list is designed to meet you where you are.

The goal is simple: you leave here with something you can actually act on this week. Not a strategy deck. Not a committee to form. Something you can open Claude or ChatGPT with today and start. To make that easier, every item below includes a copy/paste prompt you can drop directly into your AI tool of choice to get moving immediately.

Before you dive in

A few ground rules

Use Claude.

Nearly everyone in the room converged on Claude — for its reasoning, its ability to connect to APIs and data sources, and its MCP connector ecosystem. If you are on free tools or only using ChatGPT, upgrade. The gap is real.

Read-only first.

When you connect AI to any internal system — your PSA, your accounting software, your CRM — start with read-only API access. Do not give it write permissions until you have tested extensively and looped in your security lead.

Bring security into every build session.

One MSP in the room accidentally granted global admin M365 access to an entire client org because an AI-generated script wasn’t reviewed. It happens fast. Security and automation need to be in the same room from day one.

Done beats perfect.

The fastest-growing MSP referenced has one dedicated AI person and a weekly cadence of: look at last week’s problems, automate as many as possible, repeat. They’re two weeks ahead of everyone else and widening the gap.

Tier 1 · Build this week

The highest-ROI lowest-barrier items

Each one can be started today with a prompt and an existing tool you already pay for.

PSA Natural Language Query Layer

What it is

Connect Claude to your PSA (ConnectWise, HaloPSA, Autotask, etc.) via a read-only API. You and your managers ask plain English questions instead of building reports.

Why it matters

Every MSP in the room is sitting on a mountain of ticket data they can’t easily query. This removes the Bright Gauge dependency for day-to-day questions and puts answers in seconds instead of hours.

Examples of what you can ask

"What are the top 5 clients by ticket volume this month?"
"Which tickets have been open more than 7 days with no update?"
"What’s the sentiment trend for Acme Co over the last 30 days?"
"Which technician has the highest reassignment rate this week?"

Copy-paste prompt

I want to connect you to my PSA tool via its API so I can ask natural language questions
about my ticket data. My PSA is [ConnectWise / HaloPSA / Autotask — fill in yours].

Step 1: Walk me through how to generate a read-only API key in [your PSA].
Step 2: Show me how to connect that API to you using an MCP server or direct API calls.
Step 3: Once connected, I want to be able to ask questions like:
  - What are my top open tickets by time spent this week?
  - Which clients have the most unresolved tickets right now?
  - What's the average first-response time by technician this month?

Start with Step 1 and confirm you understand the goal before proceeding.

Vendor Bill Variance Report

What it is

Connect Claude to your accounting software (QuickBooks, Xero, Sage, etc.) and ask it to flag vendor bills that changed significantly month over month or quarter over quarter.

Why it matters

One person in the room described watching their controller spend half a day manually reviewing vendor invoices for cost creep. Connected to QuickBooks via MCP, Claude returned the same report in 30 seconds — listing every vendor with a change above 10%, the delta, and the direction. The controller now runs financial reviews through Claude.

Copy-paste prompt

I want to connect you to my accounting software so I can run financial queries
without manual report building. My accounting tool is [QuickBooks / Xero / Sage — fill in yours].

First, help me connect to it. I want to use the MCP connector if available,
or walk me through the API setup if not.

Once connected, my first query is:
"Show me all vendor bills from the last 3 months where the amount increased
or decreased by more than 10% compared to the prior period. List the vendor name,
the change amount, and the percentage change. Sort by largest change first."

After we run that, I want to set this up as a recurring Monday morning report.

Weekly Top-10% Ticket Automation Review

What it is

Every Monday, pull the top 10% of tickets by time consumed from the prior week. Have Claude categorize them, identify patterns, and flag which ones are candidates for automation. One person owns execution that same week.

Why it matters

This is the exact playbook used by the fastest-growing MSP referenced in our session. They saved over 700 hours per month with one dedicated person running this loop weekly. They are not adding headcount to scale — they are automating the repetition out of the business.

Copy-paste prompt

I want to build a weekly automation review process. Here's how it works:

Every Monday, I'll pull ticket data from my PSA for the prior week.
I'll paste that data here (or connect via API — let's start with paste for now).

Your job is to:
1. Identify the top 10% of tickets by total time spent
2. Group them by issue type or category
3. Flag which ones appear repetitive (same issue, multiple occurrences)
4. Recommend which ones are strong candidates for automation or scripting
5. Suggest a specific next step for each flagged item

Here's last week's ticket data: [paste your export here]

Let's start there. I'll run this every Monday and we'll build a running log
of what we've automated over time.

Internal SOP Chatbot

What it is

Upload your internal SOPs, runbooks, escalation paths, and process documentation into a Claude project or knowledge base. Staff ask questions in plain English and get the right answer instantly.

Why it matters

New hires spend weeks asking the same questions. Senior techs get interrupted constantly. This puts your institutional knowledge on call 24/7 and compounds every time you add a document.

Copy-paste prompt

I'm going to upload my company's internal SOPs and process documents.
I want you to act as an internal knowledge assistant for my IT team.

When my staff ask you questions, you should:
1. Search the uploaded documents first
2. Answer in plain, direct language — no fluff
3. If the answer isn't in the documents, say so clearly and suggest who to ask
4. If a process has multiple steps, format them as a numbered list

To start, I'm going to upload [your first SOP document].
After I upload it, ask me a few test questions so I can verify you understood it correctly.

[Upload your documents, then test with questions like:]
- "What's the escalation path for a client with a down server after hours?"
- "What do I need to gather during a new client onboarding call?"
- "Who owns the Check Point firewall renewal process?"

Tier 2 · This month

Higher value with a bit more setup

Validated by people in the room. Pick one and run it as a focused project.

Client Onboarding Discovery Chatbot

What it is

A client-facing chatbot that guides new clients through the technical discovery process. They drop in their prior MSP invoice, answer questions, and the system extracts and organizes all the information your team needs.

Why it matters

Technical discovery is one of the most time-consuming parts of onboarding. This turns a 3-hour back-and-forth into a structured async process. The chatbot asks follow-up questions, reads invoices, and flags gaps.

Copy-paste prompt

I want to build a client-facing onboarding discovery chatbot.
Here's what it needs to do:

1. Welcome the new client and explain what information we need
2. Ask them to upload or paste their prior MSP invoice — then read it and
   extract: services listed, pricing, any hardware referenced
3. Ask follow-up questions based on what it finds
   (e.g., "I see you have a SonicWall listed — do you know the model?")
4. Collect: number of users, number of locations, key contacts,
   current pain points, and any known issues
5. Output a structured discovery summary I can paste into our PSA

Let's build the conversation flow first. Draft the opening message and
the first 5 questions the bot should ask every new client.

Rack Photo to Network Diagram Overlay

What it is

Engineer takes an NMAP scan of the network and photos of the server rack. AI overlays the scan results onto the physical image — putting the IP address and device name directly over the image of each piece of hardware.

Why it matters

Documentation that used to take hours now takes minutes. One MSP in the room built this in-house and it was the most visually impressive demo of the day.

Copy-paste prompt

I want to build a tool that combines a network NMAP scan with physical rack photos
to create an annotated network diagram.

Here's the concept:
1. I run an NMAP scan and export the results (XML or CSV)
2. I take photos of the physical server rack
3. The tool matches devices from the scan to devices visible in the photos
4. It outputs an annotated image with device name and IP overlaid on each piece of hardware

Let's start simple. I'll paste an NMAP scan result below.
Read it and give me a clean list of every device found: name, IP, MAC address,
and any open ports. Then we'll figure out the photo overlay step.

[Paste your NMAP scan output here]

Project Quoting Brain

What it is

Feed your past project quotes and actual delivery hours into Claude. It builds a knowledge base that helps scope future projects faster and more accurately — and gets smarter with every project you complete.

Why it matters

Complex project scoping often lives in one or two people’s heads. When they’re unavailable, the quote either stalls or comes out wrong. This extracts that institutional knowledge and makes it available to anyone.

Copy-paste prompt

I want to build a project quoting assistant that learns from our past work.

Here's the goal: When we scope a new project, I want to be able to describe
what the client needs and get a draft quote that reflects what we've actually
done before — including real hours, common pitfalls, and licensing decisions.

To start, I'm going to paste in [one past project quote or scope of work].
Read it and extract:
1. Project type
2. Scope of work items
3. Hours estimated per item
4. Any tools, licenses, or vendors involved
5. Any notes about complexity

After we do a few of these, we'll start building a reference library
I can query when scoping new projects.

[Paste your first past quote here]

Ticket Sentiment and Client Risk Dashboard

What it is

Claude reads the last 30 days of ticket notes for each client and outputs a tone summary and risk score. Flags accounts showing frustration, repeated issues, or escalation patterns.

Why it matters

Clients rarely call to complain before they churn. The signals are in your tickets. This surfaces them before a QBR conversation becomes a cancellation conversation.

Copy-paste prompt

I'm going to paste ticket data from my PSA for a specific client —
the last 30 days of ticket notes and resolutions.

I want you to:
1. Summarize the overall sentiment (positive / neutral / frustrated / at risk)
2. Identify any recurring issues (same problem appearing more than once)
3. Flag any language in the tickets that suggests client frustration
4. Give me a risk score from 1-10 (10 = high churn risk)
5. Suggest 2-3 talking points for my next QBR with this client

Here's the ticket data for [Client Name]: [paste ticket export]

Resume Screener

What it is

Upload your job description and a batch of resumes. Claude scores each candidate, ranks them, and generates custom interview questions based on each person’s background.

Why it matters

Screening resumes is pure time. This cuts the manual review from hours to minutes — and it’s immediately packageable as a service you can offer clients who are hiring.

Copy-paste prompt

I'm hiring for [job title]. I'm going to give you our job description
and a set of resumes. For each resume, I want you to:

1. Score the candidate 1-10 based on fit with the job description
2. List their top 3 relevant strengths
3. Flag any gaps or concerns
4. Generate 3 custom interview questions based specifically on their background

Here's the job description:
[Paste job description]

Here are the resumes (paste them one at a time or upload as files):
[Paste or upload resumes]

Start with the first one and I'll confirm the format works before we do the rest.

Tier 3 · Queue for your AI developer

Bigger builds — spec them, then hand them off

Don’t try to DIY them without technical resources. The starter prompts below get your developer moving on day one.

Client Lifecycle Portal

What it is

A unified portal that surfaces client health — agreement status, firewall alerts, open tickets, upcoming renewals, CSAT scores — in one place. Built with AI-assisted code, pulling from your PSA, RMM, and documentation tools.

Why it matters

Write a one-page spec describing what you want to see on the screen for a single client. Hand that spec to your most technical person and have them use Claude to build it. One MSP in the room built a version of this in 7 days.

Copy-paste prompt

I want to build a client health dashboard. Here's what it needs to show for each client:
- Open ticket count and average age
- Agreement expiration date and MRR
- Last QBR date
- Any active alerts from our RMM
- A health score (we'll define the formula together)

Tech stack I want to use: [your preferred stack — Next.js, React, etc.]
Data sources: [your PSA API, your RMM API]

Start by scaffolding the data model. What tables or API endpoints do I need
to pull to populate each field above?

AI-Powered Website Flywheel

What it is

An autonomous marketing system that analyzes your site traffic, your competitors, and what people are searching for — then creates new SEO-optimized pages without you asking. Connected to Google Analytics, Search Console, and your CMS.

Why it matters

One MSP in the room built this and generated 7 qualified inbound leads in 10 days. His sales rep hasn’t had to do outbound in over a week. It compounds over time.

Copy-paste prompt

I want to build an autonomous SEO content system for my MSP website.
Here's the concept:

1. An orchestrator runs daily and asks: "What keywords are my competitors ranking for
   that I'm not? What questions are people asking that my site doesn't answer?"
2. It generates a brief for a new page or blog post
3. It writes the content (2,000+ words, SEO-optimized)
4. It publishes it to my site automatically

Connections needed: Google Search Console API, Google Analytics API,
SEMrush or Surfer SEO API, my CMS (WordPress / Webflow / custom — fill in yours)

Let's start with Step 1. Help me connect to Google Search Console and
pull a list of keywords where I'm ranking between position 5 and 20 —
those are my fastest opportunities.

Automated Call Flow Analyzer

What it is

Transcribe client phone calls, feed them into Claude, and get back: first-touch resolution rate, sentiment by call type, and recommendations for restructuring the call flow to reduce transfers and improve outcomes.

Why it matters

One MSP in the room delivered this as a paid engagement for a nonprofit client and doubled their first-touch resolution rate. It’s a billable service with clear ROI.

Copy-paste prompt

I have call transcripts from a client's phone system. I want to analyze them
to improve their call routing and resolution rates.

For each transcript I upload, I want you to:
1. Identify the caller's issue in one sentence
2. Note how many times the call was transferred
3. Flag whether the issue was resolved on the first touch
4. Identify the sentiment at the end of the call (resolved and satisfied / unresolved / frustrated)

After we process a batch, I want a summary:
- What % of calls were resolved on first touch?
- What are the top 5 issue categories by volume?
- What call flow changes would most improve first-touch resolution?

Here's the first transcript: [paste transcript]

External & revenue-facing

Turn what you build into client revenue

These came up in the second half of our session. Less about internal efficiency, more about how this work creates new MRR.

AI Enablement Assessment (Not "Readiness")

What it is

Send 5 questions before the meeting, ask 5 more in person, deliver a roadmap with 3 to 5 specific recommendations, and offer a monthly engagement to execute on it.

Why it matters

One MSP in the room ran 10 of these and converted 3 to monthly recurring revenue at $700/month each. The word matters: "Readiness" implies AI is optional. "Enablement" implies it’s already happening in their org — with or without you.

Copy-paste prompt

I run an MSP and I want to offer an AI Enablement Assessment to my clients.
The goal is to understand where they are with AI adoption and sell them
a monthly engagement to help them get more value from it.

Help me build 10 assessment questions. The first 5 should be sent ahead
of the meeting — they should be easy to answer and get the client thinking.
The second 5 should be asked in person based on common follow-up patterns.

The assessment should uncover:
- What AI tools they're already using (knowingly or not)
- Where they're losing time to manual processes
- What data they have that's not being used
- Security and governance gaps
- Leadership stance on AI adoption

Draft the 10 questions. Then help me build a one-page summary template
I can use to deliver findings after the meeting.

DNS Shadow AI Report

What it is

Pull your clients’ DNS traffic to show them every AI tool being accessed on their network — ChatGPT, Claude, Gemini, Midjourney, and dozens of others. Present it without judgment.

Why it matters

The reaction in the room matched what multiple MSPs have seen in practice: clients say "we don’t use AI" until you show them the report. Same aha moment as showing them leaked passwords. It opens every subsequent conversation.

Copy-paste prompt

I have a DNS traffic or web filtering log from a client's network.
I want to identify every AI-related tool or service being accessed.

Here's the log: [paste or upload your export]

From this data:
1. List every AI tool or service accessed (ChatGPT, Claude, Midjourney, Perplexity, etc.)
2. Show how frequently each was accessed
3. Flag any that could pose data security or compliance risks
4. Give me a one-paragraph executive summary I can share with the client's leadership

Format the output as a clean report I can present in a client meeting.

Build #15

Co-Managed AI Build Engagement

Some of your clients already know what they want to build with AI — they just can’t deploy it securely. You become their technical partner. They own the logic and the prompts. You own the backend: Azure hosting, GitHub, patch management, security review, uptime.

This positions you as an AI infrastructure partner rather than a vendor. It is stickier than anything you can sell as a subscription and it deepens the relationship. A trucking company in one of our peer group conversations had 35 AI projects on their wishlist and the internal talent to build them — they just needed the MSP to make them production-grade.

How to price it: Treat it like a managed application. Flat monthly fee for hosting, monitoring, and maintaining the infrastructure. Project fee for each new build you support.

Worth a 30-minute leadership conversation

Strategic questions to resolve first

Vertical vs. horizontal.

Do you go deep on one industry (legal, healthcare, construction) or build tools that work for most clients? The room leaned toward vertical — but only if you already have 30%+ concentration in a vertical. If you don’t, start with horizontal and let the verticals emerge from real client conversations.

What’s in scope.

Copilot is already billed. ChatGPT questions are already coming in. Where does AI support end and project work begin? Define it now, in writing, before your team starts making promises.

MSA language.

If you’re rolling out any AI tool to clients — especially on an opt-out basis — update your MSA first. The cleanest approach discussed: assume-acceptance model tied to monthly billing. Talk to your attorney.

R&D tax credit.

If you have engineers spending time building AI tools — internally or for clients — that work likely qualifies for R&D tax credits. One MSP in the room did a 3-year look-back and it was significant. Talk to your accountant this month.

Employee transparency.

There are probably people on your team who have already automated parts of their job and aren’t telling you — because no one has said publicly that it’s safe to do so. Make the announcement. Tell your team that getting more efficient doesn’t mean they lose hours — it means they get to spend those hours somewhere more valuable.

One last thing

Rebuild the process — don’t bolt AI on

Someone in the room said it well near the end: the smartest companies aren’t bolting AI onto their existing processes. They’re rebuilding processes with AI as the foundation.

That’s the shift. Not how do we add AI to what we do? — but if we were starting this workflow today, how would we build it knowing AI exists? That question is worth asking about every manual process in your business. And the answer, more often than not, is faster and cheaper than you think.

Want help building one of these?

BRITECITY runs the same playbook internally — orchestrators, prompts, cron jobs, and a weekly automation review. If you want a partner to spec the next one with you, book a call.

Book a call See AI IT Services

Compiled from our Wednesday morning Chicago peer group session. Hosted by Greg. All ideas sourced from MSP owners in the room.