Approach

AI does the repetitive work. A named person approves it.

We make brands, websites, apps and AI for new and growing businesses and enterprises. AI agents draft and check the work and ask a person when unsure. You can read the action log.

Stages
Discover → Improve
Approvals
One per stage
Action log
Every action

A sample screen. AI agents drafted 12 campaign versions: 11 passed and 1 unproven claim was held. The work waits for the brand lead.

How we use AI

5 controls between AI and your brand.

AI agents bring speed. The other 4 controls make their work safe to approve, on every project.

ControlWhat it doesIn the logWho owns it
01AI agents Do the repetitive work: research summaries, drafts, versions, code changes, checks and monitoring. agent.localise · 4 markets · 38 assets Built and run by us
02Tests Score every output against a written standard, so failed work never reaches an approver. eval.tone · 0.92 ≥ 0.85 · pass Pass marks set with you
03Safety checks Hard limits the AI cannot cross: unproven claims, personal data, wrong tone and hidden instructions. guard.pii · 2 fields redacted Rules owned by you
04Approvals A named person approves any step that goes live, spends money or speaks for you. gate.legal · approved · A. Rao Your approvers
05Action logs Every instruction, model, output, score and decision, recorded with who and when. log · 1,284 entries · exportable Readable by you, always
A woman at a laptop taking a printed document from a colleague standing beside her desk.
Named approverAI drafts. A named person reads and approves it before it goes live.

People decide what to make, for whom, what it may claim and whether it goes live.

How it works

How a job moves from request to approval and release.

Pick a sample job and watch it move. At the approval step, approve it or send it back with a note.

  1. 01Request

    Launch plan approved

    Spring range · 4 markets · 38 files

    plan v4 · locked

  2. 02Agent

    Market versions agent

    Adapts copy and layouts to approved parts per market

    152 drafts · 6 min

  3. 03Tests

    Brand + language tests

    Tone, wording, text size, contrast 4.5:1

    149 / 152 pass

  4. 04Safeguards

    Claims safeguard

    Holds claims like “best” with no proof on file

    3 held · reason logged

  5. 05Approval

    Market lead approves

    Reviews held items and a 10% sample

    Approver · market lead

  6. 06Launch + learn

    Published + watched

    Weekly check that live ads still match the brand

    results → next plan

Action log append-only

  1. brief.lock · v4 · by brand lead
  2. agent.localise · start · 4 markets
  3. agent.localise · 152 drafts
  4. eval.brand · 149 pass · 3 fail → redraft
  5. guard.claims · 3 held · "best-selling" unsourced
  6. gate.market · approved 149 · 3 rewritten by hand
  7. ship.publish · 152 assets · monitor on

Sample runs for a fictional “Your company”. Timings are illustrative.

The 6 stages

6 stages, each approved by a named person.

Each stage shows what AI does, what you keep and what counts as done.

  1. 01

    Discover

    Typically 1–3 weeks

    What is true today, and what is it costing you?

    AI does

    Sort interviews, analytics, reviews and competitor data into themes, with sources.

    You keep

    • Findings map with sources
    • Team interview notes
    • Opportunity shortlist

    Approval · Problem framedDone when: Agreed problem statement and success measures

    ApproverYour executive sponsor

  2. 02

    Define

    Typically 1–2 weeks

    What will we build, and how will we know it worked?

    AI does

    Draft options for what to build, estimate effort and flag risks from the findings.

    You keep

    • Scope and how we agree it is done
    • Measurement plan
    • Risks and blockers

    Approval · Scope signedDone when: Scope, cost and success measures signed

    ApproverSponsor + product owner

  3. 03

    Design

    Typically 2–6 weeks

    Does it work for the people who will use it?

    AI does

    Draft versions within your brand rules and check every screen for accessibility and content.

    You keep

    • Clickable draft and design kit
    • Accessibility review
    • User test findings

    Approval · Design approvedDone when: Tested draft, no open accessibility issues

    ApproverProduct owner + brand lead

  4. 04

    Build

    Typically 4–16 weeks

    Is it correct, fast and secure?

    AI does

    Propose code with tests, then run speed and security checks on every change.

    You keep

    • Working software, ready to test
    • Test reports
    • Security risks and scan results

    Approval · Release approvedDone when: All checks passed, code reviewed by a person

    ApproverTech lead + your IT/security

  5. 05

    Run

    Typically ongoing

    Is it staying healthy after launch?

    AI does

    Flag downtime, slow pages, rising costs, brand mistakes and old content, with proof.

    You keep

    • How-to guide and on-call rota
    • Live dashboards
    • Problem and change log

    Approval · Service acceptedDone when: Agreed response times met for a set period

    ApproverYour operations owner

  6. 06

    Improve

    Typically quarterly

    What should change next, and why?

    AI does

    Rank ideas by likely value, draft versions and add results to the to-do list.

    You keep

    • Experiment log
    • Quarterly outcome review
    • Updated plan

    Approval · Plan resetDone when: Results reviewed and next experiments agreed

    ApproverSponsor, quarterly

Quality standards

4 quality standards every launch must meet.

Built into how work is signed off, and checked on every change. These are standards we work to, not certificates we hold.

Speed · Google’s Core Web Vitals

“Good” on at least 3 in 4 visits

  • LCP≤ 2.5 sHow fast the main content shows
  • INP≤ 200 msHow fast the page answers a tap
  • CLS≤ 0.1How little the page jumps as it loads

Core Web Vitals — Loading, interactivity and visual stability

Accessibility

Usable by people with disabilities: WCAG 2.2 AA

  • 4.5:1 text contrast, 3:1 for large text
  • 24 × 24 pixels minimum for tap targets
  • Focus highlight never hidden by fixed menus
  • Keyboard and screen-reader tests before launch

WCAG 2.2 AA — Web Content Accessibility Guidelines

Security and privacy by design

Risks mapped before it is built

  • Defences chosen for each risk and checked in review
  • AI tools tested for tricks with hidden instructions
  • Data kept minimal, with consent, under Indian and EU law
  • Each change scanned for passwords

OWASP ASVS — Application Security Verification StandardOWASP Top 10 for LLM Applications — Security risks in generative AI applicationsDPDP Act 2023 — Digital Personal Data Protection Act, 2023

Carbon · Software Carbon Intensity

Measured per unit of use, then reduced

SCI = ((E × I) + M) per R

E
Energy used by the software, kWh
I
Carbon intensity of that energy, gCO₂e/kWh
M
Carbon from making the hardware it uses
R
The unit of use: a visit, a call, a user

SCI · ISO/IEC 21031:2024 — Software Carbon Intensity

How we work with you

Weekly reviews, one to‑do list and a decision log.

You see the same board and scores as we do. Every decision is logged with its reason.

  1. MonPlan

    Sort the to-do list

    AI ranks new items by value and effort. Your product owner sets the order.

  2. TueBuild

    Working sessions

    Joint sessions on the hardest problems, while AI tests each change.

  3. WedCheck

    Risk and quality

    Test scores, held items and budgets reviewed. Anything failing gets a named owner.

  4. ThuBuild

    Demo prep

    Only working software or finished work goes into the demo.

  5. FriDecide

    Weekly review

    A demo, decisions recorded and next week agreed, in 30 to 45 minutes.

What you always have access to

  • Shared to-do list. One list, ranked by your product owner, visible to everyone on both sides.
  • Decision log. What was decided, why, by whom and what it replaced.
  • Quality board. Test scores, page speed, open risks and cost, updated live.
  • Action log export. Every AI action, on request, in a format your auditors can read.
Seen from above, four colleagues around a table covered in printed charts, one pointing at a pie chart.
FridayThe weekly review: the same charts on both sides, and decisions written down before anyone leaves.

Decision log · Your company · sample

IDDecisionBecauseOwnerState
D-041Launch 3 markets first, 1 laterLegal review time in market 4SponsorApproved
D-042Raise tone test pass mark to 0.85Two drafts passed that missed the brandBrand leadApproved
D-043Keep human review on all refund answersLow volume, high riskSupport leadApproved
D-044Move search to a faster hosted serviceCost vs speed trade-offTech leadOpen

Responsible AI

6 commitments we write into the contract.

Each matches a lane of the board above and is a contract clause you can hold us to.

Frameworks we build to

ISO/IEC 42001:2023 — Artificial intelligence management systemsNIST AI RMF 1.0 — AI Risk Management FrameworkEU AI Act — Artificial Intelligence Act (EU) 2024/1689OWASP Top 10 for LLM Applications — Security risks in generative AI applicationsMITRE ATLAS — Adversarial Threat Landscape for AI Systems

  1. Your data stays yours

    Your data never trains shared AI models. How long we keep it is agreed in writing.

    Enforced on the board lane: Request signal.* · your data, your retention

  2. Disclosed

    You always know which work AI touched and which model and instructions it used.

    Enforced on the board lane: Agent agent.localise · model + prompt v4 logged

  3. Checked for accuracy and bias

    We test for accuracy, bias and harm before launch and regularly after.

    Enforced on the board lane: Tests eval.grounding · 1.00 · pass

  4. Attack-tested

    Tested for hidden instructions that trick an AI, data leaks and rule-breaking, against standard risk lists.

    Enforced on the board lane: Safeguards guard.llm01 · input clean

  5. Owned by a person

    Every AI agent has a named owner and never approves its own work.

    Enforced on the board lane: Approval gate.market · approver · market lead

  6. Reversible

    Your team can pause, undo or switch to manual any automatic action.

    Enforced on the board lane: Launch + learn ship.rollout · rollback armed

Close-up of a hand signing a printed form with a fountain pen on a light wooden desk.
Owned by a personNo AI agent signs for itself. Each works under a named owner who is responsible for it.

Plans

6 plans, all with the same controls.

Plans differ in the stages they cover, what is fixed up front and how long we stay.

  1. 01

    Sprint

    A short, fixed job that answers one clear question.

    Typical length
    1–3 weeks
    Pricing
    Fixed fee
    Best for
    An audit, a brand check or a test version of an app

    Stages covered

    1. Discover — covered
    2. Define — covered
    3. Design — not covered
    4. Build — not covered
    5. Run — not covered
    6. Improve — not covered

    Ends with a decision at the “Problem framed” or “Scope signed” approval.

    Ask about a sprint plan
  2. 02

    Project

    One clear job for one fixed price.

    Typical length
    4–12 weeks
    Pricing
    Fixed price
    Best for
    A brand identity, a website, a simple app or a set of tools

    Stages covered

    1. Discover — not covered
    2. Define — covered
    3. Design — covered
    4. Build — covered
    5. Run — not covered
    6. Improve — not covered

    Scope signed up front. Ends when the launch is approved.

    Ask about a project plan
  3. 03

    Milestone

    A bigger build in stages you approve and pay for one at a time.

    Typical length
    3–9 months
    Pricing
    Fixed price per stage
    Best for
    Work too big for one contract, such as a large app

    Stages covered

    1. Discover — not covered
    2. Define — covered
    3. Design — covered
    4. Build — covered
    5. Run — covered
    6. Improve — not covered

    You approve and pay for one stage at a time, and can stop after any.

    Ask about a milestone plan
  4. 04

    Retainer

    A monthly team that runs and improves what we built.

    Typical length
    Ongoing · 6-month minimum
    Pricing
    Monthly fee
    Best for
    Monthly social media and ads, or care for live sites and apps

    Stages covered

    1. Discover — not covered
    2. Define — not covered
    3. Design — not covered
    4. Build — not covered
    5. Run — covered
    6. Improve — covered

    Starts after launch: the Run and Improve stages, reviewed quarterly.

    Ask about a retainer plan
  5. 05

    Enterprise

    A large, long project with its own team and agreed response times.

    Typical length
    6–18 months
    Pricing
    Priced per written scope
    Best for
    Large companies making changes across markets, brands or business units

    Stages covered

    1. Discover — covered
    2. Define — covered
    3. Design — covered
    4. Build — covered
    5. Run — covered
    6. Improve — covered

    Every stage for each part, with agreed response times.

    Ask about an enterprise plan
  6. 06

    Squad

    Our specialists join your team and work in your tools.

    Typical length
    Ongoing · 3-month minimum
    Pricing
    Paid by time used
    Best for
    Teams with a clear plan that need senior people fast

    Stages covered

    1. Discover — not covered
    2. Define — not covered
    3. Design — covered
    4. Build — covered
    5. Run — covered
    6. Improve — covered

    Works to your schedule. Your people approve each stage, and we meet their standard.

    Ask about a squad plan

Tell us what you need built.

You will talk to the lead who would run your work.

Book a call

Three ways to start

  1. 01About 2 minutes

    A quick question

    You get A reply from a named lead

  2. 02About 8 minutesRecommended

    Your project details

    You get Options and prices after one call

  3. 03About 15 minutes

    A request for a quote

    You get Confirmation and a named lead

We always agree a written scope and price before work begins. How each plan is priced