Topic
Shadow AI: risks, examples, and a practical response
Shadow AI is any AI tool a person uses for work without approval, oversight, or an owner. This page covers the common examples in a revenue team, how to find them, and the response that works better than a ban.
Decision rule. Sanction faster than you ban. Every week a good tool stays unapproved is a week the work moves off your systems.
What you can do here
- Build your shadow AI inventory
Tool, owner, data entered, connected systems, next action.
- Set the ownership model
Shadow AI is a governance symptom before it is a security one.
- Read the single-player AI problem
Why individual wins never become team capability.
What shadow AI looks like in a revenue team
- A personal chatbot account used to draft customer emails, with account details pasted in.
- A meeting recorder joining calls without a data agreement.
- A browser extension with access to the CRM, installed by one rep.
- A no-code automation moving records between systems under someone's personal credentials.
- A team subscription paid on an expense card and invisible to IT.
Why it happens
It is almost never defiance. It is a person solving a real problem faster than the approved path allows. The sanctioned tool is slower, missing, or requires a request that takes two weeks. The behaviour is a signal about your intake process before it is a signal about the person.
Discovery
- Pull expense and card data for AI vendors over the last two quarters.
- Review identity logs for third-party sign-ins against corporate accounts.
- List the browser extensions and connected apps with access to the CRM and inbox.
- Ask the team directly, without consequence attached to the answer, and get more than any scan returns.
- Record each finding in an inventory rather than a message thread.
Data exposure, stated plainly
- What was entered: customer names, pricing, contract terms, personal data, unreleased plans.
- Where it went: which vendor, which region, under which terms.
- Whether inputs may be used for model training.
- What the tool can reach: read-only text, or live access to the CRM and inbox.
The response
A ban moves the behaviour somewhere you cannot see. The response that holds is an approved alternative that is genuinely good, a request path measured in days, a clear statement of what may and may not be entered, and a named owner for the inventory.
Amnesty helps. Ask people to register what they already use, without penalty, and you get an inventory in a week that a scan would not produce in a quarter.
Shadow AI inventory worksheet
Record what is actually in use before deciding what to do about it. Registration without penalty produces a better list than any scan.
| Tool | Person using it | Business use | Information entered | Connected systems | Approval status | Next action |
|---|---|---|---|---|---|---|
How to use this
- Run this as an amnesty. Consequences attached to the answer produce a shorter, less useful list.
- The connected systems column matters most: read-only text is a different problem from live CRM access.
- Close each row with a next action: approve, replace, restrict, or remove.
- Rows are held in this browser tab only. Nothing is saved or sent anywhere. Download the CSV before you close the page.
Questions readers ask
- What is shadow AI?
- Any AI tool used for work without approval, oversight, or a named owner. In revenue teams it usually appears as personal chatbot accounts, meeting recorders, browser extensions with CRM access, and team subscriptions on expense cards.
- How do you find shadow AI in a company?
- Combine expense and card data for AI vendors, identity logs for third-party sign-ins, a review of connected apps and extensions with CRM or inbox access, and a direct, consequence-free question to the team. The last one usually returns the most.
- Should companies ban unapproved AI tools?
- A ban moves the usage out of sight. What works is a genuinely good approved alternative, a request path measured in days, a plain statement of what may not be entered, and a registered inventory with an owner.
- What is the main risk of shadow AI?
- Data leaving under terms nobody reviewed, and tools holding live access to the CRM and inbox with no owner. The second is the one that turns a privacy question into an operational incident.
Any survey figures linked from this hub carry their own method and sample notes. Check those before repeating a number. Written by Jonathan Kvarfordt. Last reviewed September 19, 2026. Why trust this analysis?
What to look at first
- Unsanctioned AI charges appearing in expense reports
- How long a tool request waits for a decision
- Whether a win in one seat has been reproduced in a second
Issues
- The Single-Player Problem: Why Your AI Wins Never Leave the Room They Were Built In
Only 4 percent of revenue teams describe their AI work as fully integrated across functions. The other 96 percent have pockets of brilliance surrounded by organizational silence. Four structural decisions fix it.
- Why Does the Same AI Stack Produce Different Results at Two Companies?
AI is power. Everyone has access to the outlet. The gap between two teams running the same model is domain expertise, systems, and process, not the model.
- Nobody Owns AI in Your Revenue Org, and It Shows
AI in GTM sits between RevOps, IT, enablement, and whichever leader moved first. Ambiguous ownership is why capability stalls at the pilot line. Here is an ownership model that actually holds.
- Your CRM Was Built for Reporting. Agents Need It to Be True.
AI agents do not tolerate the fuzzy data humans quietly route around. Here is the readiness assessment to run before you point an agent at your pipeline, and what to fix first.
Research
Frameworks
Definitions
- Shadow AI Stack
The shadow AI stack is the set of AI tools reps already use outside the sanctioned roadmap: personal accounts, pasted call notes, buyer data in consumer tools. It is running your go-to-market whether or not it is governed.
- The Single-Player AI Problem
The single-player AI problem is a real AI win that stays with one operator. The workflow was never written down, owned, or wired into the system of record, so the gain never becomes a team result.
Open data
- The Shadow AI Survey
Aggregated findings from The Revenue AI Report's Shadow AI survey: unsanctioned AI usage inside revenue teams, by seat and motion. Methodology, schema, and CSV access. Free download, no signup, CC BY 4.0.
- The Tool Saturation Map
AI vendor density by revenue category: how many vendors compete in each seat and motion, and how the count is moving. Methodology, schema, and CSV access. Free download, no signup, CC BY 4.0.
Playbooks
- Centralized AI prompt library (L2)
L2 Assisted. A shared, versioned, reviewed prompt library. Boring infra; necessary.
- Sanctioned email + meeting assistant (L2)
L2 Assisted. One approved AI tool for the whole revenue team (e.g. Lavender, Regie, Copilot). Centralized billing, basic usage tracking, prompt library owned by enablement.
- Claude for Work shared projects (L2)
L2 Assisted. A whole team standardizes on Claude for Work (Projects + shared knowledge). PM, design, and eng all chat to the same context, product specs, design docs, codebase snippets, instead of copy-pasting into 5 different chats.
- AI council + intake process (L3)
L3 Integrated. Cross-functional council reviews AI tool requests weekly. Stops shadow IT, sets standards, owns budget.