For SMBs
AI for Small Businesses
Most companies under $3M do not have an AI problem. They have three people doing the work of five, and a process that was never written down.
This page is mostly about what you can do without us, because for a lot of businesses this size that is the correct answer and we would rather say it here than on a call.
The short answer
Under about $3M, the right move is usually a connector plus one narrow AI system built on your own history. A full assessment only pays for itself once one operation is costing you a salary or more per year.
This fits you if
- One operation is already costing you a full salary a year
- You have history to train on: sent quotes, replies, job records
- You want one thing gone, not a programme
- You are willing to name one person who will run it
It does not if
- The work is a clear rule with no judgement in it, buy a connector
- The process is broken rather than manual, fix that first, it is free
- You want to see what AI can do, that is a research budget not a project
- Nobody has time to sit with the build, it will not land
Start with the arithmetic, not the tool
Take the operation you were going to automate and put a number on it. Hours a week, times the loaded cost of whoever does it, times fifty. If that number is smaller than a junior salary, automate it with a $40 tool and stop reading.
At this size that is genuinely the answer more often than not, and there is no version of this page where we pretend otherwise.
The three that usually do pay at this size
- Speed to lead. If inbound waits hours and your competitors answer in minutes, that gap is revenue and it is measurable next week.
- Same-day invoicing. The invoice goes out when the job closes rather than the following week. Cheapest build, fastest to show up in the bank.
- The forty questions. Whatever your team asks you every week that they should be able to answer themselves.
All three are narrow, all three sit against a number you already track, and none of them require touching the parts of the business that are political.
What you can do yourself this month
- Write down the last twenty times you did the expensive work. Not the process, the actual cases. That is your training data and you do not have it yet.
- Find where your written policy and your real practice disagree. Every company has this gap and it is the reason help-centre chatbots underperform.
- Wire the plumbing with a connector. It is cheap and it will still be useful when something smarter sits on top of it.
- Pick the person who would own it. If there is no name, do not start.
If you do those four things and then decide to bring someone in, you will get a better build for less money, from us or from anyone.
When it is worth paying someone
When the work needs judgement. If you can explain the task to a new hire in one sentence, a rules engine handles it. If explaining it takes four minutes and the phrase it depends, that is the work worth replacing and it does not fit in a connector.
And when the thing standing between you and doing it yourself is not knowledge but weeks. That is the honest reason most owners at this size hire out.
Recent work at this size
Manufacturing
Manufacturer
Customer support system
3 agents
Workload absorbed by the system
Hours to minutes
First reply time
Home Services
Home Services Company
Sales system
4h to 4min
Time to first callback
19% to 28%
Close rate on inbound
Professional Services
Professional Services Firm
Onboarding system
3 days to 40min
New client setup
Same week
Delivery start, previously the next
Wholesale
Wholesaler
Collections system
12 days
Sooner that cash comes back
For SMBs, answered
Is this cheaper than the mid-market engagement?
The scope is narrower, so yes. One operation, no full company assessment. If it turns out the operation you picked is not the one costing you the most, that will show up quickly and we will say so.
Can we do it ourselves?
At this size, often. The list above is genuinely the order we would follow. The part people underestimate is preparing the history and testing against real past cases, which is unglamorous and takes longer than the build.
What if we grow past this?
Then the assessment starts making sense, because there are enough operations that choosing between them matters. That is the mid-market page.
Where this shows up
What you are probably weighing this against
Not sure which of the three you are looking at?
Bring the operation and a rough number on what it costs you a year. That is a fifteen-minute conversation and it usually ends with a clear yes or a clear no.