An AI adoption studio for real businesses

AI, in plain daylight.

We put AI to work on the parts of your business that are genuinely repetitive, and we'll tell you, in writing, where it won't help. That second part is why people hire us.

See what it could save you Read what AI won't fix
Est. 2021 · Calgary & remote Scroll. The lines untangle further

We are not excited about AI. We are specific about it. No "transformation journeys," no robot pictures, no percentage signs we can't defend in a meeting with your accountant.

  • 41businesses shipped, 2021–2026
  • 7clients we told not to buy anything
  • 0press releases about the future of work
01

Where AI actually helps

Here's a real pattern: invoice handling at a 30-person firm. Flip the switch and watch the same work reorganize. The time numbers are from a 2025 engagement, rounded to the nearest five minutes.

One invoice, door to ledger 26 min
A person does this Software does this Human checkpoint, always

The approval step never left. Someone still signs off on every payment over $500. That's a design decision, not a limitation.

02

What could AI save you?

Three sliders, one honest range. This is back-of-napkin math to decide whether a conversation is worth your time, not a forecast.

12
2150
9 h
1 h25 h
$48
$22$160

A realistic annual range

$0–$0

assuming AI reliably absorbs 25–45% of that repetitive time. That's the band we've measured, not the one in vendor decks

0 hours a year back for work that needs a brain

The honest caveat: about a third of the audits we run find less than this. If that's you, we'll say so in week one and the audit fee is all you'll ever spend.

03

The Trust Ledger

Every engagement starts by sorting the work into three columns. This exact table goes in your contract, with your processes filled in. Nothing moves columns without your signature.

We automate

Repetitive, rule-shaped, checkable.

  1. Data entry & re-keyinginvoices, POs, intake forms, CRM hygiene
  2. First-draft anythingreplies, summaries, reports. Drafts, never sends
  3. Routing & triagewhich inbox, which person, which priority
  4. Reconciliationmatching what happened against what should have
  5. Status chasingthe follow-up emails nobody enjoys writing

Stays human

Judgment, relationships, accountability.

  1. Every send buttona person approves anything that leaves the building
  2. Exceptionsthe weird 8% that doesn't fit the rules
  3. Client conversationsAI preps the brief; a human holds the call
  4. Hiring, firing, pricingdecisions with names attached
  5. Quality verdicts"good enough to ship" is a human sentence

We never touch

Not negotiable, not for extra budget.

  1. Your data's ownershipnothing trains on it, nothing leaves your tenancy
  2. Your voiceno synthetic you, in writing, audio, or otherwise
  3. Final decisionssoftware recommends; it never decides
  4. Your team's dignitywe don't build surveillance and call it analytics
  5. Regulated judgmentmedical, legal, financial advice stays licensed

↑ this last column is the whole business model

04

AI won't fix this.

Somebody in this industry should keep a public list. Fine, we'll go first. If a vendor tells you otherwise about any of these, ask them to put it in the contract and watch what happens.

  • ✗

    A process nobody understands

    Automating chaos gets you faster chaos. If three people describe the workflow three ways, we fix the workflow first. That part is whiteboards, not GPUs.

  • ✗

    Judgment calls with real stakes

    Pricing a bid, reading a room, deciding whether to fire a client. Models produce confident text, not accountability. Confidence is not the same thing as being right.

  • ✗

    A product people don't want

    AI can write your marketing faster. It cannot make the thing worth marketing. We've declined four engagements on exactly these grounds.

  • ✗

    Being cheap at everything

    Some tasks are still done better and cheaper by a competent person with a checklist. Our audits mark those "leave it alone", usually about a third of the list.

  • ✗

    Your culture problem

    If people hoard information or dread Mondays, a chatbot will not heal that. It will, however, generate very polished passive-aggressive emails.

  • ✗

    Work that must be right the first time

    Language models are wrong a small but stubborn percentage of the time. Anywhere that percentage is unacceptable and unverifiable, we design the human in, or we stay out.

This page is our best sales tool. that's the whole point

05

Audit. Pilot. Scale.
In that order, every time.

Fixed fees, exit ramps between every phase, and deliverables you own outright. Roughly one in three clients stops after the audit, on our recommendation.

Audit

3 weeks · $9,500 fixed

We sit with your team (actually sit, screen-share, watch the clicks) and map where the hours go. Then we sort every task into the three ledger columns and price the boring ones.

You walk away with

  • A task-level time map of your operations, with real minutes
  • Your Trust Ledger, drafted and argued over together
  • A ranked shortlist: 3–5 automations with defensible ROI math
  • A written "don't bother" list: the things to leave alone

Exit ramp: take the report to anyone, including our competitors. It's yours.

Pilot

6 weeks · $24,000 fixed

We build the single highest-ranked automation from the audit and run it in the wild, next to the old way, with your people grading its homework weekly.

You walk away with

  • One automation live in production, on your accounts and infrastructure
  • A before/after timing study your CFO can interrogate
  • An error log: every miss, published to you weekly, uncurated
  • A trained internal owner who can pull the plug without calling us

Exit ramp: if the pilot misses the number we projected, phase three is free to walk away from, and we'll say so first.

Scale

quarterly · from $6,000/mo

We roll the pattern across the rest of the shortlist, one automation at a time, each with its own checkpoint. The retainer exists to maintain, measure, and retire things, not to invent reasons to stay.

You walk away with

  • A quarterly ledger review: what's earning its keep, what gets retired
  • Documentation your next hire can run without us
  • A standing rule: any automation your team distrusts gets paused, no meeting required

Exit ramp: 30 days notice, everything keeps running. We build ourselves out of the loop on purpose.

06

Case notes

Real engagements, details blurred where clients asked. Each one includes the part that didn't work, because those are the paragraphs you should actually read.

Accounting firm · 34 people

Month-end close, minus the re-keying

Client document intake went from a shared inbox and two full-time sorters to an extraction pipeline with human review on every flagged item. Close went from 9 business days to 6.

Hours back / month
212
Extraction accuracy
96.4%
Payback period
4.5 mo

What didn't work: handwritten receipts. Accuracy sat at 71% and we stopped pretending it would improve. A person still does those. It's four hours a month.

E-commerce · 11 people

Support triage without the canned feel

Every ticket now arrives pre-sorted with a drafted reply and the customer's full history attached. Humans edit and send. First-response time fell from 11 hours to 40 minutes.

Drafts edited
38%
CSAT change
+0.6 pts
Headcount change
0

What didn't work: auto-sending "simple" replies. We trialed it for two weeks, CSAT dipped, we killed it. Every reply has a human's name on it again.

Property management · 60 people

Maintenance requests that route themselves

Tenant requests get classified, prioritized, and matched to the right trade with photos attached, before a coordinator ever opens them. Emergency response time dropped 55%.

Requests / month
1,900
Misroutes
2.1%
Coordinator overtime
−80%

What didn't work: predicting maintenance before it's reported. Six weeks of trying, nothing beat the building managers' gut. We refunded that workstream.

07

Plain answers

The questions people actually ask once the slides are closed.

Will this replace my staff?

In five years and 41 engagements, zero clients have cut headcount because of our work. What changes is the shape of the day: less re-keying, more of the work that needs a person. If your actual goal is layoffs, we're the wrong studio and this was a cheap way to find out.

Where does our data go?

It stays in your accounts, your cloud, your API keys, your audit logs. Nothing we build trains a model on your data, and we put that in the contract with teeth: it's a termination-for-cause clause, not a values statement.

What if the AI gets something wrong?

It will. Roughly 2–5% of the time depending on the task, which is why everything we ship has a human checkpoint before consequences and an error log you can read anytime. We design for "wrong sometimes, caught always", not for a perfection nobody can deliver.

We tried ChatGPT and it didn't stick. Why would this?

Because a chat window isn't a system. Sticking requires plumbing: connections to your actual tools, checkpoints, an owner, and a number it's measured against. That's the unglamorous 80% of our job. The model is honestly the easy part.

How disruptive is the audit?

About 45 minutes per key person, spread over three weeks, mostly watching them work. No all-hands, no workshop with sticky notes. Most teams tell us it's the first time anyone asked where their hours actually go.

Which AI vendors are you tied to?

None. No reseller margins, no referral fees; we've turned them down in writing. We pick per task, we tell you why, and about a fifth of what we recommend isn't AI at all. Sometimes the answer is a form with required fields.

What does it cost, really?

The numbers on this page are the numbers: $9,500 audit, $24,000 pilot, retainers from $6,000 a month. Fixed, not "starting at." If scope grows, that's a new signed page, never a surprise line item.

Ready for a boring,
profitable conversation?

Thirty minutes. Bring your messiest process. We'll tell you on the call if there's nothing here for us. That happens about a third of the time, and it's a good call either way.

hello@daylight.studio

no newsletter pop-up appeared during this visit. you're welcome.