Work
Selected work
Various workflows we've designed across industries.
Accounting
Tax document intake and missing items
In tax season, client documents arrive through a portal and by email, often incomplete.
Documents drop into an AI sort and land in three bins: received, missing, and needs a look. The one that needs a look pauses while staff review it, then moves on to the tax software. What AI does
- Sorts each document and spots what is missing.
- Prepares the request for missing items.
What people keep
- Nothing reaches a client until staff approve it.
- A person checks every number in a return.
Impact
- 150 hrsper tax season
- $9,000per season
- Same daymissing forms requested
For a practice with about 300 returns, with intake going from 45 minutes to 15 per return, at $60 an hour.
Bookkeeping exceptions and the posting cycle
Bank, payroll, and depreciation data arrive on three different schedules.
Bank and payroll feeds enter a recurring month-end loop. AI categorises a batch of transactions, the batch pauses while a reviewer approves it, then the loop continues to post it to the books. What AI does
- Puts the data into one consistent shape.
- Prepares entries for review.
What people keep
- A person approves before anything posts.
- Review stays the firm's professional judgment.
Impact
- 1,920 hrsper year
- $86,400per year
- 1 FTEof capacity
For a firm with about 40 clients, with each client's monthly work going from 6 hours to 2, at $45 an hour.
Finance
Screening new deals against market data
An investment firm screens offering memos and checks every broker's rent assumptions against nearby comps by hand, one memo at a time.
New offering memos and a set of comparables feed one AI comparison. Mismatches peel off into a flags tray. The result pauses while a partner approves the screen, then moves on to the shortlist. What AI does
- Reads each offering memo.
- Flags rents that run above nearby comps.
What people keep
- Partners decide which deals go to committee.
- Nothing goes back to a broker without approval.
Impact
- 600 hrsanalyst time per year
- $51,000per year
- Every memochecked against comps
For a firm screening about 20 memos a month, with each screen going from 3 hours to 30 minutes, at $85 an hour.
Monthly property financial review
An investment firm reviews a long monthly report from every property it owns, and someone has to find the exceptions.
Each month the owner reports go once around the AI review cycle, and one exception lifts off for attention. The result pauses while the asset manager approves it, then follow-ups go out. What AI does
- Compares each report with budget and last month.
- Flags exceptions, with the source page.
What people keep
- The asset manager approves every action.
- Property management keeps its own systems.
Impact
- 540 hrsper year
- $45,900per year
- 1 dayfrom report to follow-up
For a firm with about 20 properties, with each review going from 3 hours to 45 minutes, at $85 an hour.
Law
One contact list from every attorney's mailbox
Relationships with clients, adjusters, and referral sources sit in each attorney's mailbox.
Attorney inboxes and the client list feed scattered contact fragments, which drift together into one AI-built record. It pauses while an attorney confirms the merge, then joins the shared contact list. What AI does
- Builds one contact list from email metadata only.
- Suggests who owns each relationship.
What people keep
- Attorneys confirm every pipeline move.
- Anyone can correct a label in one click.
Impact
- 416 hrsper year
- $25,000per year
- One viewof who knows whom
For a firm with about 60 attorneys, with a coordinator's time going from 10 hours a week to 2, at $60 an hour.
Following up on overdue invoices
Overdue invoices often get followed up by hand, one at a time.
Overdue invoices get an AI reminder, and most resolve right there. The rest climb to an AI second notice. The last step pauses while an attorney approves it, then the account is resolved. What AI does
- Tracks every overdue invoice.
- Prepares the right next step for each one.
What people keep
- The billing manager approves every email.
- Attorneys handle the oldest balances.
Impact
- 450 hrsper year
- $27,000per year
- Every invoicegets its next step on time
For a firm with about 250 overdue invoices a month, with each follow-up going from 12 minutes to 3, at $60 an hour.
Construction
Scheduling service visits across crews
A builder with a service team books repair and maintenance visits every day, and someone has to fit each job to the right crew by skills, location and open time by hand.
New service requests drop into one AI step that plans the day's routes and assigns each job to a crew. The plan pauses while the dispatcher approves it, then each crew lane carries its job to that crew's schedule. What AI does
- Reads each new request and its priority.
- Proposes a day's routes that fit jobs to crews.
What people keep
- The dispatcher approves or changes the plan.
- Crews get their schedule in the tools they already use.
Impact
- 360 hrsper year
- $16,200per year
- Same dayplan ready each morning
For a team booking about 60 service visits a week, with routing going from 10 hours a week to 3, at $45 an hour.
Turning field changes into change orders
On a busy job, changes show up in emails, texts and site notes, and turning them into priced change orders before the work is done keeps slipping.
Field emails and site notes land in one change order that AI drafts section by section. The draft pauses while the project manager approves it, then it goes to the owner and into the books. What AI does
- Gathers field changes from email and site notes.
- Drafts the change order against the estimate and contract.
What people keep
- The project manager approves the price and scope.
- Nothing goes to the owner until it is signed off.
Impact
- 720 hrsper year
- $46,800per year
- Every changelogged
For a builder running about 10 active jobs with 4 change orders a month each, with drafting going from 2 hours to 30 minutes, at $65 an hour.
Distribution
Weekly retail sell-through report
Brands selling through several retailers get sales data in each retailer's own item numbers.
Six retailer portals feed one AI step that matches every item to one product list. The result pauses while the team approves it, then one report goes out. What AI does
- Pulls sales from every retailer each week.
- Matches items to one product list and drafts the report.
What people keep
- Unmatched items go to a person.
- Nothing reaches executives until it is approved.
Impact
- 364 hrsper year
- $21,800per year
- Mondayreport, every week
For a brand selling through about eight retailers, where the report takes about 8 hours a week, going to 1 hour, at $60 an hour.
Weekly product ranking report
Product teams want a weekly read on what moves on the largest online marketplace.
Marketplace rankings arrive out of order and AI sorts them into an ordered stack. The top two are highlighted while the team picks what to act on, then the weekly brief goes out. What AI does
- Sorts new products into your categories.
- Writes one analyst's pick in plain words.
What people keep
- No AI touches a number.
- If the pull fails, nothing is sent.
Impact
- 234 hrsper year
- $17,500per year
- Every weekon a fixed schedule
For a company where the report takes about 5 hours a week, going to 30 minutes, at $75 an hour.