Skip to content

Knock Twice · AI setup for product and data teams

AI, set up properly for your team.

In two weeks I connect your AI tools to the systems and data your team already uses, with read-only access and spending caps. Then I write the guidelines that keep output consistent and build three workflows your team will use every week.

Two weeks · Fixed price · Read-only by default · You own everything

You work directly with Adam, Head of Product with ten years in enterprise software.

demo:~$ psql -c "select grantee, privilege_type, count(*) as tables from information_schema.table_privileges where grantee = 'ai_readonly' group by 1, 2"
grantee | privilege_type | tables
-------------+----------------+--------
ai_readonly | SELECT | 11
(1 row)
demo:~$ claude -p "How much revenue did we make in 2025, and from how many customers?" --allowedTools "Bash(psql *)"
In 2025 we made $450.58 in revenue from 46 customers, across 80 invoices.
Query: select sum(total) as revenue_usd, count(distinct customer_id) as customers, count(*) as invoices, min(invoice_date), max(invoice_date) from invoice where extract(year from invoice_date) = 2025
Row count: 1
Replay of a real session · demo data

Everyone has AI. Few teams have it set up.

It lives in private chats.

Everyone prompts differently, so nothing is shared and nothing compounds.

It can't see your work.

Without access to your data, tickets and designs, it guesses.

Nobody owns the risk.

Broad permissions and open-ended spend make leadership nervous, so rollout stalls.

The fix isn't another tool. It's setup: access, guidelines and a few workflows that fit how your team already works.

What you get in two weeks

Connected, safely.

Your AI tools connected to the systems your team uses, such as your data warehouse, your docs, Linear or Jira and Figma. Read-only by default, scoped roles and a monthly spending cap on every connection.

A team guidelines file.

One shared file that tells your AI how your team works: your terms, your standards, what to ask before acting and what never to do. Output gets consistent across people.
CLAUDE.md
# Chinook team guidelinesShared by every AI tool the product team uses. Chinook is a demo company: an online music store. Read this before any task.## Our terms- Revenue: the sum of invoice totals, in USD.- Market: the billing country on the invoice.- Customer: anyone with at least one invoice.- Business customer: a customer with a company on the account.- Year: the calendar year of the invoice date.- Catalog: tracks, albums and artists. Genre belongs to the track.

Three working workflows.

Built on your real work and tested with your team before handoff.
  1. Answers from your data.
  2. Call notes to spec draft.
  3. Design system audit.

Training and handoff.

A live session with your team, a short written playbook, and a record of every connection and how to turn it off. Everything runs on your accounts.
  • A live session with your team
  • A short written playbook
  • A record of every connection and how to turn it off
  • Everything runs on your accounts

Three workflows, shown with real output

Examples from a demo setup. Yours are chosen with you on day one, based on where your team loses the most time.

Answers from your data.

Ask in plain English and get the answer with the query shown, so anyone can check it. It runs through a read-only role with a monthly spending cap, so nothing can be changed and costs can't run away.

DEMO INPUT

Which five markets brought in the most revenue in 2025, and how many customers bought in each?

REAL OUTPUT

SELECT billing_country AS market, SUM(total) AS revenue_usd, COUNT(DISTINCT customer_id) AS customers FROM invoice WHERE EXTRACT(YEAR FROM invoice_date) = 2025 GROUP BY billing_country ORDER BY revenue_usd DESC, market LIMIT 5;
marketrevenue_usdcustomers
USA85.1411
Canada72.277
France40.594
Brazil37.624
Czech Republic36.752
Assumption: "customers" means anyone with at least one 2025 invoice billed to that market.

The latest invoice in the warehouse is dated 22 December 2025. I don't know if later sales are missing or if there were none.

Call notes to spec draft.

Drop in a customer or stakeholder call. Get a first-draft spec in your team's format: the problem, requirements, open questions and acceptance criteria. Your PM edits instead of starting from a blank page.

DEMO INPUT

Dana: Thanks for making time. What made you write in?
Priya: Month end. I lost two hours to your invoices again.
Dana: Walk me through those two hours.
Priya: Our editors buy tracks for client videos. Thirty or forty a month.
Priya: Every purchase is its own invoice, in its own email.
Priya: My accountant wants one statement a month. I build it by hand.

REAL OUTPUT

## Problem
"Every purchase is its own invoice, in its own email", so a video studio builds its monthly statement by hand.
Its accountant rejects invoices that "show my name, not the studio or our tax number", and the studio "would shop elsewhere" after its books close in December.

## Requirements
1. Offer one monthly statement per account, listing that month's invoices (assumption: calendar month).
2. Save a company name and tax number once, then print both on every later invoice.
3. Let buyers type a free text reference at checkout that prints on the invoice.
4. Let customers get old invoices on the site, not only by email.

## Open questions
1. Should the statement be a PDF, a spreadsheet or both? Owner: the customer's accountant.
2. Must we check the tax number, or is typing it in enough? Owner: Finance.
3. How many of our 10 business customers have this problem? Owner: Dana.

## Acceptance criteria
1. Given several invoices in one calendar month, when the month ends, then one statement lists them all.
2. Given a saved company name and tax number, when the next invoice is issued, then it shows both.
3. Given a reference typed at checkout, when the invoice is issued, then it shows that reference.

Design system audit.

Point it at a Figma file. Get the list of problems your design system hides: duplicate colors, raw values that should be tokens, broken variants. Designers fix the list instead of hunting for it.

DEMO INPUT

Audit the mock design file designs/checkout-kit.mock.json against our design system rules.

REAL OUTPUT

Checkout kit has problems in all three groups: 2 duplicate color pairs, 5 layers with raw values and 3 broken variants.

## Duplicate colors: 2 pairs
- `color/text/muted` and `color/gray/600` are both #52525B. `color/text/muted` is bound on `Label` in the two Disabled `Button` variants. No layer in this file is bound to `color/gray/600`.
- Style `Brand/Blue` and variable `color/brand/primary` are both #1F6FEB. The variable is bound on four layers: `Type=Primary, State=Default`, `Type=Primary, state=Focus`, `State=Focus` in `Input`, and `Promo banner`. No layer in this file uses the style.

## Raw values: 5 layers, 6 values
- `Order summary`: fill #F4F4F5 is not bound. Same value as `color/surface/subtle`.
- `Price`, inside `Order summary`: fill #52525B is not bound. Same value as `color/text/muted` and `color/gray/600`.
- `Promo banner`: corner radius 6 is not bound. No radius token is 6. The tokens are 4, 8 and 999.
- `Payment form`: left padding and right padding are 13 and not bound. That is two values. No space token is 13. The nearest is `space/3` at 12.
- `Pay button`: the instance overrides the fill with #1F6FEA and binds it to nothing. Its component uses `color/brand/primary`, #1F6FEB. The two are one hex digit apart.

## Broken variants: 3
- Name off the pattern: `Type=Primary, state=Focus`. It says `state` in lower case. The other five variants say `State`. So the set has a second property, `state`, with Focus as its only option.
- Missing state: `Type=Secondary, State=Hover`. Primary has Hover.
- Missing state: `Type=Secondary, State=Focus`. Primary has Focus, in the misnamed variant.

Built so your security lead says yes

Read-only first.

Write access only where you approve it, scoped to a single task.

Spending caps.

Every connection has a monthly limit you set.

Your accounts, your keys.

Everything runs on your accounts. I don't keep copies of your data or credentials.

Documented and reversible.

A one-page record of every connection, what it can access and how to turn it off.

How the two weeks run

  1. Day 1 · Kickoff.

    We find where your team loses the most time and pick the three workflows.

  2. Days 2 to 4 · Connect.

    Tools and data connected with scoped, read-only roles and spending caps.

  3. Days 5 to 8 · Build.

    Guidelines written, workflows built and tested on your real work.

  4. Day 9 · Train.

    A live session with your team, plus a short written playbook.

  5. Day 10 · Hand off.

    Everything documented and handed over.

Your time: about three hours from one owner on your side, plus the training session.

Who you'll work with

I'm Adam, Head of Product at a multi-tenant commerce platform, where this setup runs every day: AI connected to our warehouse through a read-only role with a monthly spending cap, design audits run against our Figma files, and specs and tickets drafted in our own format. I've spent ten years building enterprise software, and I set your team up the way I set up my own.

Knock Twice is my studio.

Email

Brands Knock Twice has shipped work for

  • Alexis Lauren
  • Barbara Katz
  • Brightwood
  • Chused & Co
  • EY
  • Five Guys
  • Jack Henry
  • Madrinas
  • MediaCom
  • MetLife
  • MOD Pizza
  • Omura
  • P.F. Chang's
  • R&R
  • Táche
  • Wendy's

Case study · Restaurant commerce platform

An AI data assistant for restaurant brands

Plain-language questions about a brand's app orders, answered straight from its Snowflake data. Every figure in an answer is filled in from a query result or checked against one.

Standard question, median time
2.4 s
Down from 4.6 s at first release
Analysis question, median time
7.5 s
Down from 42 s at first release
Model cost of a standard question
0.02 cents
Down from about 0.8 cents
Model cost of an analysis question, with a warm cache
2.4 cents
Down from about 15 cents

Case study · Restaurant commerce platform

Ordering from a restaurant inside ChatGPT

In three days, a planning document became a working ChatGPT app where a guest orders from a restaurant in plain language. In testing, an AI model built 20 of 20 orders correctly and never invented a menu item.

From a planning document to a working app
3 days
Test orders built correctly
20 of 20
Target: 95% or more
Invented items or options
0
Target: 0
Median tool response time
0.46 s
Target: 2 to 5 s

One scope. One price.

The Two-Week AI Setup

$4,500fixed

Founding price for the first three teams: $3,000, in exchange for a case study.

Book a 20-minute fit call

Includes

  • Tools and data connected with read-only roles and spending caps
  • A team guidelines file
  • Three workflows, built and tested on your real work
  • Live training and a written playbook
  • A record of every connection and how to turn it off
  • 30 days of email support after handoff
Terms
50% at kickoff, 50% at handoff.
Guarantee
If any of the three workflows isn't working at handoff, I keep going until it is, at no extra cost.

Optional monthly support is available after handoff.

Is this a fit?

Good fit

  • A company of 10 to 100 people in software, ecommerce, restaurants, retail or hospitality
  • A product, data, design or engineering lead who owns the rollout
  • A data warehouse, and tools like Linear or Jira and Figma
  • You want workflows that stick, not a strategy deck

Not a fit

  • You need an AI product built for your customers
  • You want an AI strategy presentation
  • Nobody on your side can own it after handoff

FAQ

Which AI tools do you set up?

The ones your team already pays for. Most of my work is with Claude and the standard connectors for tools like Linear, Figma and common data warehouses. If something you use can't do the job safely, I'll tell you before we start.

Is our data safe?

Access is read-only by default and scoped per connection, every connection has a spending cap, and everything runs on your accounts. I don't keep copies of your data or credentials, and you get a written record of every permission.

How much of our time does it take?

About three hours from one owner on your side, plus the team training session.

What if the team doesn't use it?

We pick workflows where your team already loses time, test them on real work before handoff, and train on those exact tasks. If one isn't working at handoff, I keep going until it is.

What happens after the two weeks?

You own everything and can run it on your own. Optional monthly support is there if you want help adding workflows.

Do you take on custom builds?

Sometimes, after a setup. If your team needs something built on top of it, such as a data assistant, I scope it as its own project with a fixed scope and a fixed price.

Which industries is this for?

Software, ecommerce, restaurants, retail and hospitality. My own background is commerce. If your business is different, tell me on the call and I'll say whether it fits.

Do you work with engineering teams?

Yes. The setup works the same way for engineers. The scope stays at three workflows, so we pick the three that matter most.

Why a fixed price?

You know the cost and the outcome before we start, and I'm paid for finishing, not for hours.

Do you sign NDAs?

Yes.

Set your team up properly.

A 20-minute call to see if it's a fit. If it isn't, I'll tell you what I'd do instead.

Book a 20-minute fit call

Know a team that needs this?

Book a 20-minute fit call