Putting AI to work in your business
What AI can do for an organization today, where your data goes when you use it, and how to adopt it without being tied to this year's tools.
Where things stand with AI
AI tools are changing month to month. What was true at the start of the year often isn't by the fall. Most business owners we talk with know AI matters. Few have a clear picture of what it can do for them today.
That leaves two worries pulling in opposite directions. If you wait, competitors may get ahead. If you commit too early, the business may be tied to a system that is obsolete within a year or two, through a product, a vendor, or a contract that is hard to leave. Underneath both is a third question: where your information goes when someone types it into an AI tool.
Two places AI helps
The first is in how software gets built. We use AI coding tools in our own work, and they have changed what custom software costs. Building something shaped to one organization's work used to be hard to justify against a license. Often it no longer is.
The second is inside the work itself. An AI system can read your documents, answer questions from your own records, prepare routine reports, and answer the phone, working from the information your business already keeps.
You don't have to work out which kind you need. That comes out of understanding the problem.
What AI does well today
These are the uses we recommend now. Each works reliably when it is built carefully: a person reviews any result that matters before it is used, and every result is logged.
-
Reading documents and correspondence
Pulling dates, amounts, names, and obligations out of email, invoices, letters, and forms, and turning them into records, schedules, and work queues.
-
Answering questions from your own data
Questions asked in plain English and answered from your own records and documentation, with the method shown so the answer can be checked.
-
Preparing routine reports
Spreadsheets, charts, and summaries drafted from your data.
-
Answering the phone and handling requests
A voice agent answers calls in a natural voice, takes messages, and answers questions from your own information. Connected to your systems, it can book appointments, look up an order, or update a record, reaching only what it has been allowed to reach. It is the caller's choice: anyone who would rather wait for a person or leave a voicemail can.
-
Training staff
An AI tutor that knows your business works through what a new employee needs to learn: your products, prices, policies, and procedures. It explains, asks questions, and checks understanding, and it can play a customer on a practice call. Each session is graded against your real information, with quotes showing what went well and what to work on.
Where your data goes when you use AI
Every AI system sends information to a model somewhere. Where that model runs decides who can see your data, what it costs, and how capable the model is. There are four broad options, and many organizations use more than one.
-
On your own hardware
Open models run on servers you own, in your office or data center, and nothing leaves the building. The costs are the hardware and its upkeep. These models are capable, though smaller than the largest hosted ones.
-
Dedicated hosting
A model runs on cloud servers reserved for you, in a region you choose. Your data leaves the building but isn't processed alongside anyone else's. You pay for the capacity whether or not you use it.
-
Inside your cloud account
The large cloud platforms offer leading models, including ones from other AI companies, billed by use inside your own account. Your data stays with a cloud provider you may already use, under the agreements you already have with it, including a business associate agreement for health information. The convenience comes with a pull toward building everything on that one platform.
-
Pay-per-use services
Requests go to a provider's shared service through its API and are billed by use. This gives access to the most capable models at the lowest starting cost. Your data goes to the provider with every request, under the provider's terms.
What to check before choosing a provider
Retention and training terms differ by provider and by plan: whether requests are stored, for how long, and whether they can be used to train future models. Business plans usually come with stronger terms than the free consumer apps, which staff may already be using on their own.
Independent certifications show how seriously a provider runs its operations. The ones to look for:
- SOC 2 Type 2
- An outside auditor tested the provider's security controls over a period of months, not just on one day.
- ISO 27001
- The provider runs a certified program for managing information security.
- ISO 27701
- That program extends to handling personal information.
- ISO 42001
- The provider manages its AI systems under a certified program, the newest of these standards.
A certification shows the controls were audited. What happens to your data is still set by the contract.
If you handle health information covered by HIPAA, the provider must sign a business associate agreement before any patient information is sent to it. Many providers offer one only on certain business plans, and the free consumer apps generally don't offer one at all. There is no official HIPAA certification, so a provider described as "HIPAA certified" is describing its own controls. The signed agreement is what counts.
Your own records count too. The documents, questions, and answers an AI system handles are business records, so we keep them in your systems under your retention rules.
We go through the terms and certifications with you and match each kind of data to a deployment that suits it. Sensitive records can stay on your own hardware while routine work goes to a pay-per-use service.
Keeping today's gains from becoming tomorrow's liability
Cloud hosting went through this cycle first. Hosting at scale brought real advantages: capacity on demand and no hardware to maintain. Organizations that built deep into one provider's own services also found that leaving became a project nobody wanted to fund. AI is on the same path, moving faster.
So we build for change. Your data, your processes, and the knowledge of how your business works are what last, and they stay in systems you own. The model is one replaceable part. When a better or cheaper one arrives, we swap it in, and you don't wait on a vendor's plans to get it.
Because every result is logged, a new model can be tested against your real cases before it goes into use.
See it working (coming soon)
We're building a public demonstration: an AI analyst working against the business systems of a fictional distributor, including its sales, purchasing, and accounting records. You'll be able to ask it a question, get the answer with its spreadsheets and charts, and read how the answer was reached.
Start with a problem
You don't need an AI strategy to begin. Tell us what isn't working, and we'll say plainly whether AI is part of the answer. Sometimes a simpler fix is better.
The first step is a conversation with Randall Smith, the engineer who would do the work. If the problem needs a closer look, we propose a paid assessment with a written report that is yours to keep.