I Run 12 AI Tools Every Day. Here's How I Stop Them Running Me.

At some point over the last couple of years, I stopped choosing AI tools and started just accumulating them. n8n for workflow automation. Claude and ChatGPT for drafting, reasoning, and triage. OpenAI API powering a couple of in-house bots. GoHighLevel with its own AI layer. Pipedrive, HubSpot, and Close all bundling "smart" features into their CRMs. Instantly.ai and Smartlead.ai handling sequenced outreach. Clay for prospect enrichment. Apollo.io for lead intelligence. Help Scout running AI-suggested replies. Zendesk's AI summarisation sitting on top of all of it.

That's twelve tools with meaningful AI functionality, and I haven't even counted the ones embedded in Microsoft 365. Some I pay for directly. Some come with a CRM subscription I'd have anyway. Some I requested; others were pushed down from product decisions I wasn't in the room for.

This is what AI tool sprawl looks like from the inside. It's not dramatic. It's just slow accumulation, and by the time you notice it, you're already spending real money, carrying real risk, and doing a lot of duplicated work across systems that don't talk to each other well.

Why IT and Ops Teams Feel This First

The people who catch sprawl earliest are the ones who sit across the whole stack: IT, ops, and whoever's running the automations. When a sales leader adopts a new AI outreach tool, they're thinking about pipeline. When I find out about it three months later, I'm thinking about which data just left the building, which existing workflow now has a conflict, and whether the licensing model is compatible with what we already have.

According to research making the rounds this year, the average enterprise runs more than a dozen AI tools — and most were adopted without formal IT review. Among executives, the number using tools their IT department hasn't approved runs well above 90%. I believe it. I've watched it happen.

The problem with shadow AI isn't that individuals are lazy about security. It's that the sanctioned alternatives usually don't cover what they actually need to do. People go around IT because IT is slow, not because they're malicious.

That's worth sitting with. If you're the IT person who finds out about tools after the fact, the answer isn't better enforcement — it's a faster way to say yes, and a clearer picture of what you actually have.

What Sprawl Actually Costs

The money is the easy part to calculate. Add up subscriptions, API credits, per-seat fees across all the AI-layer tools, and you'll usually find a meaningful portion of your software budget going to overlapping capabilities — two tools that both summarise tickets, three that all score leads.

The harder costs are operational. When Claude is doing something different to what Help Scout's AI is doing, and neither result matches what Zendesk's summarisation produced, a support agent has to figure out which one to trust. That's friction. It slows decisions, erodes trust in AI suggestions generally, and often means the agent ignores all three and starts from scratch.

There's also the data handling side. Each of these tools has its own privacy terms, data residency behaviour, and model training policies. Some are very clear about it; others bury the relevant clauses. Every tool you add is another surface where customer data travels — and if you're working with an enterprise client or operating under GDPR, that's not a theoretical concern.

The Quarterly Audit I Actually Run

About a year ago I started running a structured review of the AI stack every quarter. It takes about half a day and has saved real money each time — not through dramatic cuts, but through finding tools that had drifted from their original purpose or been superseded without anyone noticing.

Here's the structure I use.

Step 1: Build the full inventory

List every tool that touches AI functionality, including the ones bundled into platforms you'd keep regardless. Be honest about what counts. If a CRM has an "AI-suggested next step" feature that someone clicks on regularly, it goes on the list. If that same feature exists in the CRM but has never been turned on, note it but bracket it separately.

For each tool, capture: who uses it, how often (roughly), what specific tasks it supports, and what it costs per month including API overages.

Step 2: Ask three questions about every tool

I ask the same three questions about each item on the list:

• Does another tool on this list already do this? Not could do it with some configuration — actually does it, in use, today. If the answer is yes, one of them is redundant. Figure out which one is used more and cut the other.
• If I removed this tomorrow, what would break and who would notice? The answer tells you whether it's embedded in a real workflow or just sitting there as a fallback nobody reaches for. Tools that nobody notices are gone are candidates for removal.
• Do I know where my data goes when I use it? Not "I assume it's fine" — do I actually know? Acceptable answers are: yes, I've read the DPA; or yes, it's self-hosted. "Probably" is not an acceptable answer.

Step 3: Consolidate around the tools that do the most work

After two rounds of this audit, my stack has stabilised around tools that do heavy lifting across multiple use cases rather than single-purpose tools that do one thing well but create integration overhead. n8n sits at the centre because it connects everything else. Claude handles the reasoning-heavy work — summarisation, triage logic, draft composition — because I can call it from within n8n and control exactly what context it sees. The outreach-specific tools (Instantly.ai, Clay, Apollo.io) stay because they're genuinely purpose-built and nothing else replaces them cleanly.

What I've cut or not renewed: standalone AI writing tools that got subsumed by Claude, a lead-scoring add-on that duplicated what a CRM already did natively, and a chatbot platform that cost more than the volume of interactions it handled could justify.

The Governance Side Nobody Talks About

Once you have the inventory, you need a process to prevent the next round of sprawl before it happens. This doesn't need to be complicated. A single shared document where anyone proposing a new AI tool answers the same three questions — what does it do, what does it overlap with, where does the data go — is enough. It creates a paper trail and, more importantly, it forces the proposer to think before adopting.

The goal isn't to slow down adoption. The goal is to make adoption deliberate. There's a real difference between choosing twelve tools because you've thought through what each one does that no other tool does, and having twelve tools because each one seemed reasonable in isolation when someone signed up for it.

Right now, a lot of teams are in the second situation. The sprawl happened fast, the review never happened, and nobody's quite sure what they've got or what it's costing.

A Sign This Is Becoming Mainstream

When I first started talking about AI tool audits internally, it felt like a niche concern — something that only mattered if you were deep in the automation weeds. That's changed. Gartner and others are publishing on it. SaaS security vendors are building sprawl-detection products. The conversations I see in IT and ops communities have shifted from "what AI tools should I add?" to "how do I manage what I've already got?"

That's a healthy shift. The novelty phase of AI adoption — where adding a new tool felt exciting regardless of whether you needed it — is ending. What comes after it is the operational maturity phase, where the orgs that did deliberate adoption will outperform the ones that accumulated and hoped for the best.

Half a day per quarter to audit the stack is a small investment. The alternative is finding out the hard way, usually when a vendor changes their pricing model, a security incident surfaces a tool you forgot existed, or you're onboarding a new team member and realise you can't actually explain what all these things do.

If you're running a similar audit or working through your own version of this problem, I'm happy to compare notes. The link below goes to my contact page.