What Salesforce's SMB AI data says for mid-market operators

In December 2024, Salesforce published its 2025 Small & Medium Business Trends Report based on responses from 3,350 SMB leaders across North America, Latin America, Asia-Pacific, and Europe. The top-line numbers got the press: 91% of SMBs with AI say it boosts revenue. 78% expect AI to materially change how their company operates. 75% are at least experimenting with AI.

Those are real numbers. They are also unhelpful without context. The headline numbers do not tell a 100-person operator what to do Monday morning.

Two findings buried two-thirds of the way down the report do. They explain the entire gap between SMBs that grow with AI and SMBs that fall further behind. The pattern matches what shows up across 140 conversations with mid-market operators.

74% of growing SMBs are increasing data management investments. 47% of declining SMBs are doing the same. A 27-point gap on the single most predictive variable in the survey.

This is not a "growing companies have more money" finding. The survey covers companies up to 200 employees and the gap holds across size brackets. What it captures is sequence. The growing SMBs decided to fix their data layer before they bought AI tools. The declining ones bought AI tools and waited for the data layer to follow. It does not.

That sequence pattern shows up in every engagement I run. A construction group running seven concurrent projects bought project management software with "AI insights." The crews stayed on their existing system. The insights never got made. The CFO still could not answer "are we making money on this project?" without calling three people and waiting two weeks.

We started with the data layer. Consolidated the cost data. Cleaned the project tracking. Connected the field reports from WhatsApp into a structured form. The fix usually starts with data consolidation before AI. Then the AI on top of that data started returning answers the executive team could trust. The order matters. You cannot put AI on top of broken data and expect anything but worse-quality wrong answers, faster.

Growing SMBs are twice as likely to have an integrated tech stack: 66% vs 32%. That is the largest gap anywhere in the survey.

Integration is not a feature. It is a posture. The 66% cohort looked at their tool sprawl and said: these need to talk to each other or none of them are worth keeping. The 32% kept buying tools and let the integration problem sit.

A $14B wealth advisory firm I work with had 90% of its institutional knowledge in Outlook. The other 10% was scattered across SharePoint, file servers, and one partner's personal cloud drive. The firm had bought tools. None of them talked to each other. Information was visible to whoever happened to be in the right meeting, invisible to everyone else. When a new prospect walked in, the partner blocked four to eight hours rewriting a proposal from scratch because they could not find the close-enough version that existed somewhere in the archive.

The fix was not another tool. It was making the existing knowledge searchable. One source of truth the AI agent reads on every draft. Now proposals are 80% drafted in minutes.

The 91% revenue-boost number is great for a LinkedIn headline. It tells you nothing about what to do Monday morning.

The data-management and integration gaps are the actionable numbers. They tell you, in order:

1. Fix your data before you buy AI. The 27-point gap is the cost of skipping this step.
2. Connect what you have before you add what you do not. The 34-point integration gap is the cost of letting tool sprawl continue.

Do those two things and the 91% revenue-boost finding becomes available to your business. Skip them and it does not. The order is the variable.

Across 140+ conversations with operators in construction, legal, financial services, manufacturing, distribution, and architecture, the same pattern shows up.

A construction CFO does not know real-time project margins. The data exists, but it lives in spreadsheets, WhatsApp messages, paper forms, and the project manager's head. AI cannot help here. Data consolidation can.

A law firm partner takes a call at 11pm because there is no system filtering qualified leads from junk. An AI agent could help here. But only after the firm decides which CRM is the source of truth and connects the WhatsApp inbox to it.

A packaging manufacturer operating in four countries cannot tell a customer whether the SKU they want is in stock. The data exists in four separate country systems that do not talk. AI cannot help here. Integration can.

In every case, the AI use case sits downstream of a structural problem. Salesforce's data shows the SMBs who figure this out grow at materially higher rates than the ones who do not.

The Salesforce survey defines SMB as 200 employees or fewer. That overlaps with the high end of the Work-Smart ICP (20 to 200 employees). The findings hit even harder at the upper end.

In the 50-200 employee range specifically, the data-and-integration sequence is more consequential than in the 5-50 range. A 10-person company can run on Excel and email forever. A 100-person company hits a structural ceiling. Either the data and integration question gets answered, or growth stalls because the executive team cannot see what is happening fast enough to act on it.

That is the call I get most often: "We grew past what spreadsheets can hold. We have not done anything about it yet. We do not know where to start." Sometimes the company already bought AI tools nobody uses. Sometimes they have not yet and the team is debating which one. Either way, the answer is the same: the tools are not the problem. The data and integration layer is.

Salesforce's data does not name the failure modes by name. The corpus does. Across the engagements I have run, four patterns repeat. The full breakdown is in why 95% of AI projects fail.

1. Buying AI before fixing data. Microsoft Copilot, ChatGPT Enterprise, an agent platform. Six months later, nobody uses it. The 32% of stagnant SMBs without an integrated stack are largely in this group.
2. Buying integration as a project, not a posture. A three-month MSP engagement that consolidates two systems and stops. The other six systems keep sprawling.
3. Hiring a fractional CTO to "lead AI strategy." Strategy decks, working groups, no production system six months later. The team rotates.
4. Skipping the audit. Buying tools based on which industry conference made the loudest claim, instead of which problem actually costs the company money.

The growing-SMB cohort in the Salesforce data is not avoiding AI. They are sequencing it correctly. Audit, data, integration, then AI. That order is the entire difference.

An AI Ops Audit takes two to four weeks. It is fixed-fee. The output is a map of your data, your tools, your processes, and a ranked list of where you are losing time and money with a "can AI help here" rating against each.

For a 100-person company, the audit usually surfaces three to five high-leverage starting points and another five to ten that should wait. The high-leverage ones are usually not the obvious ones. A construction CFO came in convinced she needed an AI agent for client communication. The audit showed the real cost was three weeks of delay in surfacing project margins. We built the cost dashboard first. The communication agent is still on the roadmap. It will be useful when we ship it. It would have been useless if we shipped it first.

That sequencing is the whole game. Salesforce's data gives the cleanest macro signal yet that the SMBs getting this right are pulling away from the ones getting it wrong.

A diagnostic audit runs $5K to $12K, scoped to company size and complexity. The fee applies toward a build if you move forward.

A foundation build that addresses the data and integration layer first, then adds AI on top, runs 4 to 16 weeks fixed-fee. Most builds qualify for the IRC Section 41 R&D Tax Credit, which counts 65% of contractor payments as Qualified Research Expenses. The effective cost after credits and Section 174A immediate expensing drops 18-23%. Florida companies with 50 or fewer employees that include training can stack the Florida Incumbent Worker Training grant on top for an additional offset.

The tax-credit framing matters specifically because the Salesforce data shows growing SMBs invest more in data management. That investment is what qualifies as qualified research. Your tax credits are largest exactly on the work the growth data tells you to prioritize.