Put AI to work on one workflow at a time, with your team in control.
We add AI to the systems you already use: reading documents, answering from your own knowledge, drafting replies and summaries. Each use starts as a measured pilot, with a person reviewing the output.
invoice-2291.pdf
Draft entry
Best for
Teams losing hours to documents, repetitive emails or searching for information
Typical timeline
Assessment in [X] days, pilot in [X–Y] weeks
Engagement
Fixed-price assessment, then a fixed-price pilot
Starts with
An AI opportunity assessment
Where AI usually pays off first
People copy data from PDFs, forms and emails into your systems.
The same customer questions get answered by hand every day.
Answers live in documents nobody can find.
Staff already paste company data into public AI tools, and there are no rules yet.
What we build with AI
Document intake
Invoices, forms and contracts read into structured data, with a review step before anything is saved.
Knowledge assistant
An internal assistant that answers from your policies, manuals and past tickets, and shows the source of each answer.
Drafts and summaries
Replies, summaries and reports drafted inside the tools your team uses, for a person to edit and send.
Workflow automation
Sorting, routing and follow-ups between your systems, with logs and limits.
The rules every AI feature follows
Privacy and mistakes are the two reasons most companies hold back on AI. These rules address both, and they’re written into every proposal.
AI used for decisions about jobs, credit, housing, insurance, education or health care carries extra legal obligations in several states. We flag these during the assessment, and we recommend you confirm them with your counsel.
Data
Processed through business APIs, with training on your data turned off where the provider allows it. We document each provider’s terms for you.
Access
The AI only sees what the workflow needs, with the same permissions as the person using it.
Human review
A person approves anything that changes records, sends messages or affects a decision about someone.
Logging
Inputs, outputs and approvals are logged, so you can check what happened.
Costs
A monthly spending cap and alerts, agreed before the pilot starts.
Disclosure
If customers talk to an AI assistant, it says so at the start and offers a way to reach a person.
Off switch
Every AI feature can be turned off without breaking the rest of the system.
What you get
From the assessment to a measured pilot, with a clear decision at the end.
Assessment report
Your workflows ranked by time saved, risk and effort.
Pilot plan
Success measures, a baseline of how the work runs today, and a budget cap.
Working pilot
Connected to your systems, with a review screen for your team.
Accuracy testing
Checked against your real examples before the pilot starts and throughout it.
Data-handling record
Providers, data used, retention and settings, documented for your records.
Results review
Measured against the baseline, with a recommendation to scale, change or stop.
How it works
Assessment
Interviews and a review of your workflows, data and tools.
From you: an hour with the people who do the work.
Pilot design
One workflow, success measures, a baseline and a cost cap.
From you: sample documents or data, anonymized if needed.
Pilot
The AI feature running with real work and a person reviewing every result.
From you: a small group of users and their feedback.
Decide
We compare results with the baseline and recommend whether to scale, change or stop.
From you: the decision.
For your IT team
Which models, where the data goes and who owns what.
Models and providers
Business APIs from OpenAI, Anthropic or Google, or models you host yourself. We document the data terms of whichever we use.
Security
Prompt-injection and data-leak risks tested before launch, API keys in a secrets manager, and no personal data in prompts unless the task needs it.
Ownership
Prompts, code, test sets and configurations are yours, so you can change providers later.
Questions about AI projects
Will AI replace our staff?
That’s not how we approach it. We target the repetitive part of a job, so people spend their time on the parts that need judgment.
Is our data used to train AI models?
We use business APIs and turn off training on your data wherever the provider allows it. We document each provider’s terms so you can check them.
What happens when the AI gets something wrong?
That’s why there is a review step and a test set. We measure the error rate before launch and keep measuring it afterwards.
How much does it cost to run?
Model costs depend on volume. The assessment estimates them, and the pilot runs under a monthly cap you approve.
Which task would you hand to AI first?
Tell us about it. We’ll tell you honestly whether AI is the right tool for it.