Move work to AI, one tested step at a time.
Tower finds the next step AI could do, tests it on your own records, and shows the hours saved next to what it costs.
Nothing moves to AI until it is tested on your own records and someone on your team approves.
A newer AI cleared the test on your own records.
Tested on 200 of last quarter's payment lines: 197 handled correctly.
Half of adults have tried AI. Almost nobody's work runs on it.
Computers took about 40 years to reach almost everyone. The internet took about 30. Generative AI was tried by more than half of US adults in 3.
People don't know where to start, how to iterate, or how to work with AI. And every few months a better, cheaper model makes last quarter's setup out of date.
Sources: Bick, Blandin & Deming (St. Louis Fed); Gallup US employee surveys, 2024 to May 2026; Pew Research Center, Sep 2025; US Census Bureau; Pew internet adoption.
It sees how your work runs, and what's changing outside.
Tower maps each workflow from the tools you already use and reads each new AI release. It suggests how the work should change, tests the change on your own work, measures it, and keeps going.
AI arrives. Some steps are automated, built once, and then left alone.
Models keep improving. Tower keeps the work improving with them. Each new model is tested on the step where a person sits, using your own work. It passes, and a person says yes.
Approves stays with a person on purpose. Sign-off, advice and trust stay human. Every change is measured in hours and dollars.
This isn't about replacing people. Tower coaches each person with honest, private feedback and shows them the next thing worth handing to AI.
Where Tower works, department by department.
Built first for teams of 20 to 200 in mortgage and lending, accounting and professional services, real estate and insurance. Pick a department to see one workflow in detail: which steps AI does, where a person stays, and what it adds up to.
AI does the matching and the math. People handle exceptions and sign-off.
Match lender payments to bank deposits
Each month a few hundred payment lines arrive by email and lender portal. AI reads every line and matches it to a deposit. A person only sees the lines the AI isn't sure about, and each decision teaches it.
Accountants using generative AI supported 55% more clients and closed the books 7.5 days sooner.
Choi (Stanford) & Xie (MIT), 2025. 277 accountants at 79 firms. A link found in the data, not proof AI caused it.
- Calculate commissions and payoutsAI does 90% · saves ~6 h/week
- Prepare month-endAI does 45% · saves ~8 h/week
Stays with a person: Month-end sign-off and disputes
The relationship stays human. AI does the work around it.
Brief before every client call
Before each call, AI pulls the client's history, current rates and options into a one-page brief. The salesperson reads it, has the conversation, and AI drafts the follow-up for them to send.
- Draft follow-ups after callsAI does 60% · saves ~6 h/week
- Shortlist options for a clientAI does 40% · saves ~5 h/week
Stays with a person: The conversation and the advice
AI chases and checks. People make the judgment calls.
Chase missing documents
AI checks each open file for what's missing, asks the client by email or text, reads what comes back and files it. A person only steps in when a document is unclear or doesn't fit.
Workers given an AI assistant spent 2 fewer hours a week on email.
Dillon, Jaffe, Immorlica & Stanton, NBER, 2025. Randomized, 7,137 workers at 66 firms. Email time only. Some authors work at Microsoft, which made the tool.
- Book appointmentsAI does 80% · saves ~5 h/week
- Route requests to the right personAI does 75% · saves ~4 h/week
Stays with a person: Exceptions and judgment calls
AI builds and checks the file. The credit decision stays with a person.
Gather and check documents for a file
AI collects the documents, reads income and ID, checks them against the application and flags anything that doesn't add up. The underwriter starts from a clean, packaged file and makes the decision.
Mortgage lenders that lean on technology process applications about 20% faster, with no rise in defaults.
Fuster, Plosser, Schnabl & Vickery, Review of Financial Studies, 2019. Measured before today's generative AI. In 2023 only 7% of lenders used generative AI (Fannie Mae).
- Flag what doesn't add upAI does 60% · saves ~6 h/week
- Package files for reviewAI does 75% · saves ~4 h/week
Stays with a person: The credit decision
AI drafts and runs the schedule. The team decides what goes out.
Draft and schedule content
AI looks at what's working, drafts posts and emails in the company's voice and schedules them. A person edits for facts and tone, and nothing goes out without their approval.
Professionals using ChatGPT finished writing tasks 40% faster, and the quality was rated 18% higher.
Noy & Zhang, Science, 2023. Randomized, 453 professionals. Short writing tasks in a test, not a full job.
- Track competitors' rates and campaignsAI does 85% · saves ~4 h/week
- Schedule postsAI does 90% · saves ~3 h/week
Stays with a person: Approving what goes out
AI answers the routine questions. People take the ones that matter.
Answer routine questions
AI reads each question, finds the answer in the client's file or company policy and replies to the routine ones. Upset or complex customers go straight to a person, and AI learns from how they answer.
Support agents with an AI assistant resolved 14% more issues an hour, and newer agents 34% more.
Brynjolfsson, Li & Raymond, Quarterly Journal of Economics, 2025. 5,179 agents. One company's chat support team.
- Route callsAI does 80% · saves ~5 h/week
- Send status updatesAI does 90% · saves ~4 h/week
Stays with a person: Upset or complex customers
Numbers on cards marked EXAMPLE are modelled, not measured: a company of about 50 people, with time valued at $40 an hour. Research figures link to the published studies. Across the US economy, people using generative AI saved 5.4% of their work hours (St. Louis Fed, 2025, self-reported).
One system, three views.
Every AI dollar, tied to what it changed
What AI costs by department, the time it gives back, and the tools nobody uses.
Responsible for the team doing the work
Which of their team's workflows are improving, in hours and dollars, and what to test next.
Their own numbers, privately
How much of their work AI does now, private coaching, and one new change to try.
Tower looks at how work flows, not at people's screens.
Your data is yours. Tower reads the records your tools already keep (tickets, documents, code changes, AI sessions), with the hours and costs attached.
Never screens or keystrokes. No screen recording, no keystroke logging, no rankings of people.
Who sees what. Coaching is private to each person. Managers see how workflows are doing in hours and dollars.
Start with one workflow.
Pick the workflow
Choose one workflow and agree which records Tower can use. We check what it finds with the people who do the work.
Test one change
Record the baseline, agree what success means, test one change on your own work, and review the result together.
Keep it improving
After a person approves, the change goes live and is measured. As models improve, Tower tests again.
Before you join the waitlist.
Is this employee monitoring?
No. Tower measures workflows: which steps AI does, which people do, and what that costs. It doesn't record screens or keystrokes, and coaching is private to each person.
Who decides what moves to AI?
A person at your company. Tower proposes a change and tests it on your own work; it never switches a step to AI on its own.
What happens when a new AI model comes out?
Tower checks whether it changes any of your workflows, tests it on the steps where a person sits today, and shows you the result and the cost before anything changes.
Is Tower available today?
Yes, on our own team, every day. We're taking on a small number of companies of 20 to 200 people next, one workflow at a time. Join the waitlist and we'll be in touch.
Is our AI spend paying off?
Tower sets what AI costs against the time it gives back, by department and task, and flags tools nobody uses.
Could a cheaper or open model do this step?
Tower tests models on your own records and shows quality and cost side by side. A person approves any switch.
What does it cost?
A monthly subscription per company, plus a one-time setup priced on the value it creates. We share the numbers when we talk.
Where Tower came from.
How we found the problem
We built an AI workflow that cut a client's monthly commission calculations from two weeks to about three hours. A person reviews only what the AI isn't sure of, and it learns from every correction.
What was missing: as AI models got better, nothing checked how that workflow should change. That's Tower.
One change on our software team, measured
On Oct 20, a person started reviewing the risky changes before they went live. Tower measured the 11 days either side.
Only 31 of 110 risky changes got the review, and activity also fell, so we don't claim the review alone did it.
Tower runs on our own team today.
Two brothers, building it full time.
Jeetesh Rup
CEO & Co-founder
Former M&A lawyer and KPMG management consultant. Built benefits-realization frameworks that measure whether big changes actually deliver, the method Tower runs on.
VR
CTO & Co-founder
Former CPA who taught himself to build with AI. Co-founded the software agency where Tower was built; now full time on Tower.