Small businesses are getting the AI back office big companies already have

Operators running a business solo or with a tiny team are the slice of the economy that McKinsey's State of AI finds least likely to have AI in production: 29% of companies under $100 million in revenue, against roughly half of companies above $5 billion. The gap is documented; the mechanism for closing it has not been. This month, HoneyBook shipped what its category has not had before: a connector that lets an AI assistant act on an operator's records, not just answer questions about them.

HoneyBook is a customer-relationship platform founded in 2013 for independent service businesses, the photographers, event venues, and consultants who run enquiries, contracts, scheduling, and payments in one tab. Its new offering, HoneyBook MCP, is a Model Context Protocol (MCP) connector: a standardized way for an AI assistant (in this launch, Claude) to read and write against HoneyBook's underlying data. The "read" part answers the questions a small operator might already ask an AI tool: did client X pay the invoice; how many leads are stuck at the proposal stage. The "write" part is what has not existed in this category before. The assistant can update a pipeline stage, send an invoice from a template, generate a contract, or fire a reminder email. The wire write-ups frame that as a productivity story. The structural question is whether one connector in one CRM can move the 29%.

The mechanism matters because the adoption gap is not a capability gap. The model can already do the work. What an SMB operator cannot easily get is permissioned, structured access to the systems of record where the work actually happens, including the CRM, the invoicing tool, the contract store. MCP is a standardized protocol: a connector declares what it exposes, the assistant declares what it wants, the operator approves. HoneyBook's connector is one of the first CRM-shipped implementations aimed at the small-business segment rather than at enterprise software buyers.

The pain it is aimed at is concrete. A HoneyBook survey of 455 US venue owners, released July 27, found 97% say they lack an easy system to organize bookings. That is the population the 29% adoption figure is averaging over: operators whose bottleneck is operational, not technical, and whose time is the scarce resource. The connector's argument is that the data already exists inside HoneyBook. The operator just does not have hours to use it.

A second HoneyBook study, reported by the trade press, claims small business owners who have adopted AI earn almost five times as much revenue as those who have not. That figure comes from HoneyBook's own research, not an independent benchmark, and the comparison does not control for the kind of operator most likely to buy an AI subscription in the first place. It is useful as a vendor-stated ceiling, not as a forecast for what the connector will deliver.

The falsifier for the gap-closing claim is the next quarter of rollouts. If the 29% number moves in McKinsey's next update, and if the SMB end of the distribution gets there without a parallel flood of integration-maintenance work for operators, the connector pattern is doing structural work. If the number stays flat, the connector will have added a feature rather than a category. The framing should avoid implying a partnership beyond the MCP standard itself.

The wire treated this as a HoneyBook product launch. Whether the 29% moves in McKinsey's next update is the test of whether the connector pattern is structural, or just another integration for operators to maintain.