Tooling Sprawl

There is a specific failure pattern that hits small businesses about eighteen months into their AI adoption, and once you know what to look for it is impossible to miss. It is the tooling sprawl pattern.

Symptoms: the company is paying for fourteen AI tools. Three of those subscriptions are charged to a credit card belonging to a person who left six months ago. Two more were signed up for during a workshop and never used. The team uses different tools for the same task — one person on Claude, another on ChatGPT, a third on a third-party wrapper — and the outputs vary enough that comparing them is impossible. Nobody can give you a complete list of what the company is paying for. The IT manager has stopped trying.

The visible cost — the credit-card bill — is the smallest part of the damage. The real cost is what sprawl does to your capability.

Tooling sprawl is not a single bad decision. It is the consequence of a sequence of individually reasonable ones.

First, the team decides to experiment. Experimentation is good. They sign up for three tools. Each person picks their favourite. Nobody picks the same one. Now you have three.

Then a vendor comes through with a vertical solution — AI for sales, AI for legal, AI for support. Each one is genuinely useful in its niche. Each one is bought. Now you have six.

Then somebody attends a webinar and signs up for the platform the webinar was about. Seven.

Then the team building an integration uses a different model than the one the team using the chat uses. Eight.

Then one of the existing tools releases a new product that overlaps with one you already have, but it has a feature you want. Nobody cancels the old one. Nine.

Eighteen months in, you are at fourteen, and nobody can quite remember why most of them were chosen.

The subscriptions are a rounding error. The real costs are quieter and larger.

Capability fragmentation. Each tool requires the team to learn it, build trust with it, develop taste for its outputs. Spread that across fourteen tools and the team is competent at none of them. They produce mediocre output from each, instead of excellent output from one.

Inconsistent voice and standards. Different tools produce different outputs. The team's external communications, internal documents, and customer-facing material drift apart. The brand and the standards become un-enforceable because no single tool is producing the work.

Security and credential risk. Fourteen tools means fourteen access surfaces, fourteen sets of API keys, fourteen places customer data could leak. Most SMBs have no real audit of this. When something goes wrong — and it will — nobody knows where to look.

Decision fatigue. Every new project starts with which tool should we use for this? That question, repeated three hundred times a year, drains the team's energy for the actual work.

The fix is unglamorous and politically difficult. It is: pick two or three tools, deeply. Cancel the rest.

The criteria for which two or three: one general-purpose conversational tool the whole team standardises on; one tool that handles your largest single workflow well (writing, research, customer support, whatever it happens to be); optionally one specialist tool for a niche where the depth genuinely matters. That's it. Three tools. The rest go.

The team will protest. The CFO will be relieved. Six months later, the team will produce noticeably better output, because they have built real depth with three tools rather than shallow familiarity with fourteen.

If you can only have three AI tools, here is the test for each one.

Could I remove this tool today and would the business notice tomorrow? If the answer is no — and for most tools, the answer is no — remove it. If the answer is yes, dig deeper: would the business notice in a week? In a month? In a quarter? Most tools fail this test at one month. Keep only the ones that would be missed within a week.

Run this audit annually. Sprawl re-accumulates. The team that runs the audit relentlessly is the team that compounds. The team that doesn't will be back at fourteen tools in eighteen months, having paid for the lesson twice.