Service 03 / 10AI ideas, agents and rules

AI strategy and AI agents that do everyday tasks for you.

We find where AI helps you most, then build AI agents that draft quotes and chase invoices. We test each agent on your own examples. A named person approves every risky step.

An illustrative AI task board for “Your company”. Six tasks sit in four columns: queued, agent working, needs approval and done. Each card shows the agent and the person who owns the work. It also shows the level, spend against a budget and progress through the steps. A billing dispute waits for the Service lead to approve a $240 credit before the agent adds it.

What each agent may do alone

  1. L1Suggestsuggests only
  2. L2Drafta person sends it
  3. L3Act with approvala person approves
  4. L4Act within limitselse asks a person
Typical length
Strategy in 3 weeks · first agent live by week 8
Scope
AI ideas · agents · AI helpers
Standard
Tested before going live

01Why now

Why AI trials stall, and what we agree before we build.

AI trials rarely fail because of the AI. They stall for the 4 reasons below. Screens marked Illustrative are examples, not client results.

What happens to 100 AI ideas

Illustrative
  1. Ideas raisedEvery team has a list 100
  2. 01 Stalls on: No owner, no measure −58
  3. Demos builtTried on sample data 42
  4. 02 Stalls on: Data not ready −24
  5. Trials with usersLive data, a few people 18
  6. 03 Stalls on: Risk unclear −10
  7. Approved for wider useOwner, risk and cost agreed 8
  8. 04 Stalls on: Cost unknown −3
  9. In daily useMeasured every week 5
A presenter walks colleagues through a chart on a large screen in a glass-walled meeting room
The demoThe demo is easy. Daily use is the hard part.

Agreed before we build

  • A named owner for the work
  • A starting number and a target
  • How ready your data is
  • A risk level and AI limits
  • A cost limit per task
  1. 01

    No owner, no measure

    Nobody owns the work the demo changes, or the number it should move.

    How we fix itEach idea gets a named owner, a starting number and a target first.

    Ranked AI ideas
  2. 02

    Data not ready

    The demo used a clean sample. Live data is scattered, out of date or hard to reach.

    How we fix itWe score the data for each idea, and every gap goes on the plan with an owner.

    The workshop
  3. 03

    Risk unclear

    Nobody can say what the agent may touch or who approves it, so legal says no.

    How we fix itEach agent gets a risk level, limits, named approvers and a log of every action.

    Safe AI rules
  4. 04

    Cost unknown

    Nobody knows what each task costs at full volume compared with doing it by hand.

    How we fix itFrom day 10, we compare the cost per task with doing it by hand.

    How we test

02Ranked AI ideas

Your AI ideas ranked, with the top 3 costed.

We score each idea on value, ease, risk and time. You get a plan by quarter and business cases for the top 3.

Handwritten sticky notes on a printed canvas listing customer problems and pains
Value before buildEvery idea starts from a named problem and what fixing it is worth, before anyone builds.Photo: Daria Nepriakhina / Unsplash
  • Value

    Hours saved or sales affected, using your own volumes.

  • Ease

    How easy it is to build: where the data lives and how clean it is.

  • Risk

    EU AI Act risk level, personal data and the harm if the agent is wrong.

  • Time to results

    Weeks to a measured trial, data work included. Each week takes 0.05 off the score.

A chart of twelve AI ideas by value and ease to build. Three in the high-value, easy-to-build corner are marked to build first. They are reconcile vendor invoices, triage inbound RFPs and draft renewal quotes. The ranked list that follows holds the same information.

Bubble size = hours of work a month · axes 3–10Illustrative portfolio

score = 0.45·value + 0.35·ease + risk − 0.05·weeksrisk: +1 minimal · +0.4 limited · −1 high

#AI ideaScoreRisk

What you take away

  • Ranked planAll 12 ideas, in order, by quarter
  • Business case × 3Today’s numbers, target, costs and payback
  • Owners and measuresOne owner and one number to move per idea

Selected · rank 01 · score 7.5

Reconcile vendor invoices

Finance · owner Finance lead · 4–6 wks to a measured trial

Value
360 hours a month matching about 2,900 invoices to orders and receipts.
Ease to build
Invoices and orders already sit in the accounts system. About 18% arrive as PDFs to be read first.
Risk Minimal risk
Minimal tier. Supplier data only, and payment release stays with a person.

DecisionBuild first

Trial starts atL2 · Draft

Top level allowedL4 · only after 90 days at L3

03Agent demo · Illustrative

How much an agent does alone is up to you.

Pick a task, set the level and limits, and run it. The level only changes the last step. If the agent hits a limit, it passes the task to a person.

agent-runner · Your company · test space Live demo · touch any control to take over

Interactive agent run: choose a task, an autonomy level and limits, then press Run. The trace lists each step; the run ends in a suggestion, a draft, an approval request, an action within limits, an escalation or a halt, and adds an audit row. Outcomes are pre-written; no live model is called.

1 · Task
2 · Autonomy level
3 · Limits
Agent will propose 8%

run_7f3a21 · renewals-agent · prompt renewals v14.1

Renew a customer contract

L3 · Act with approval
Latency
3.9 s
Tokens
4.6k → 800
Records
3
Spend
$0.020

Run complete · approved by Sales ops lead · quote sent · audit entry written

  1. Planplannerrouter-small

    inGoal: renew account 20-418 before 14 Oct

    out4 sub-tasks · tools kb, crm, calc

    640 ms1.2k → 180$0.0003

  2. Retrieve policykb.searchreadembed

    in“renewal discount policy”

    outPricing policy v4 §3.2 · discounts up to the approval limit

    410 ms40 → 0<$0.0001

  3. Read systemcrm.get_accountread

    inaccount_id = 20-418

    outARR $48,000 · 3 seats added · usage +18%

    280 ms——

  4. Calculatecalc.renewal

    inARR, seats, 24-month term

    outPrice $51,840 · discount 8%

    35 ms——

  5. Draftllm.draftreasoner-large

    inQuote and cover email in the account’s tone

    outQuote Q-7731 · email of 146 words

    2100 ms3.4k → 620$0.020

  6. Policy checkpolicy.check

    indiscount, records, spend vs limits

    outDiscount 8% ≤ 10% · records 2+1 ≤ 25 · spend $0.020 ≤ $0.25 · cites §3.2

    90 ms——

  7. Write actioncrm.update_quotewrite

    inQ-7731 → status “sent”

    outApproved by Sales ops lead · Q-7731 sent

    320 ms——

Ready

Press Run agent to start. Change a limit first to see a guard step in.

Suggestion · L1 changes nothing

Renew account 20-418 for 24 months at $51,840 with an 8% discount. Sources: pricing policy v4 §3.2 and CRM usage.

Sent to Sales ops lead. A person does the work.

Draft saved · crm.save_draft

Editable. Nothing leaves until Sales ops lead sends it.

Approval needed · Sales ops lead

Send quote Q-7731 to the customer?

Evidence: pricing policy v4 §3.2 · CRM usage +18% · 3 records affected

Executed after approval

Quote Q-7731 sent. Approved by Sales ops lead.

Write
crm.update_quote · Q-7731 → status “sent”
Approver
Sales ops lead
Waited for a person
14 s
Audit row
run_7f3a21 · append-only

Run stopped by a guard

The run passed its limit and stopped before acting.

Handed to Sales ops lead with the trace so far.

Stopped at
—
Written
Nothing
Spend so far
—
Audit row
—

Audit log · append-only · newest first

  1. 10:42:07run_7f3a21L3Renew a customer contractExecuted after approval

    Approver
    Sales ops lead
    Models
    router-small · embed · reasoner-large
    Prompt
    renewals v14.1
    Tools
    kb.search:r · crm.get_account:r · crm.update_quote:w
    Records
    3
    Spend
    $0.020
  • Approved tools only

    Only the tools each level allows.

  • Limited access

    It can read a lot but change little.

  • Spend and record limits

    Checked before each step and each change.

  • Policy checks

    Each change checked against your written rules.

  • A person approves

    Risky steps wait for a named person.

  • Full action log

    Model, prompt, tools, cost and approver.

OWASP Top 10 for LLM Applications — Security risks in generative AI applications

LLM06 · Excessive Agency. OWASP’s list of AI security risks warns against giving AI too much power. We give few tools, ask approval for risky steps and check outside the model.

04How it works

What a live AI agent is made of.

A demo needs a model and an instruction. A live agent needs 8 parts, each tested, watched and easy to swap. We pick the tools for each part with you.

  • Models

    plan

    A small, fast model for easy steps and a bigger one for hard ones. From a provider, or open models in your region.

    • Anthropic
    • Google Gemini
    • Meta Llama
    • Mistral AI
    • OpenAI
    • vLLM
    • Hugging Face
  • Task steps

    plan

    Each task saves progress, retries and picks up after a failure.

    • LangGraph
    • LangChain
    • Temporal
    • Python
  • Memory & knowledge

    plan

    Short-term memory in Redis. Your documents indexed, with a link to each source.

    • PostgreSQL
    • pgvector
    • Qdrant
    • Redis
    • LlamaIndex
  • Logins & access

    act

    Each agent has its own login, with only the access it needs, for a short time.

    • Okta
    • Auth0
    • Vault
    • OpenID

A diagram of an agent loop: plan, act, observe, repeated until the task is done. Eight parts sit around it. On one side: models, task steps, memory and knowledge, and logins and access. On the other: your systems, automatic checks, testing and monitoring. Each part lists the technologies used for it.

  • Your systems

    act

    MCP, a standard AI link, connects your systems once for all agents, within limits.

    • Model Context Protocol
    • FastAPI
    • OpenAPI Initiative
    • HubSpot
    • Salesforce
  • Automatic checks

    act

    Your rules checked on every action, personal data hidden, inputs and outputs checked outside the model.

    • Open Policy Agent
    • Microsoft Presidio
    • NeMo Guardrails
    • Llama Guard
  • Testing

    observe

    Every change runs on your own examples before release. An AI scores each case, checked against your team.

    • promptfoo
    • Ragas
    • MLflow
    • GitHub Actions
  • Monitoring

    observe

    Every step recorded with its speed, cost and inputs, in the tools your engineers already use.

    • OpenTelemetry
    • Langfuse
    • Grafana
    • Datadog
  • Models you can swap

    Changing a model is one setting, and all the tests run again before release.

  • Your data stays yours

    Paid business terms, or models on your own servers. Nothing trains on your data.

  • Your region if needed

    Models and data stay in the region your policy or GDPR (Europe’s data law) requires.

05How we test

Every change is tested on your own examples.

Each agent has a test set made from your own cases. A change goes live only if it passes. A person still signs off each release.

test-runner · renewals-agent · test set v10 · 218 cases Tests · release allowed

An illustrative test dashboard for one agent. It shows five pass marks and success per scenario against its mark. It compares three models. It shows the cost and energy saved by using a small or a large model per step. A test history follows. An instruction change lets the agent act before its rules check, and the test check blocks the release. The rule moves into code and the re-run passes.

  • Task success

    94.2%

    pass ≥ 92%

  • Correct tool use

    98.1%

    pass ≥ 97%

  • Policy violations

    0

    must be 0

  • Cost per task

    $0.021

    manual $4.80

  • 95% of tasks done in

    6.8 s

    pass ≤ 10 s

Success by scenarioIllustrative

  1. Standard renewal64 cases 98%pass ≥ 95%
  2. Seats changed mid-term38 cases 96%pass ≥ 92%
  3. Discount over the limitmust hand off25 cases 100%pass = 100%
  4. Usage data missing25 cases 88%pass ≥ 85%
  5. Hidden instructions in account notesmust refuse · LLM0130 cases 100%pass = 100%
  6. Ambiguous contract termmust ask36 cases 84%pass ≥ 80%
Scorer agreement
κ 0.84 vs human labels
Marked by 2 people
40 of 218 cases
Last calibrated
release 0418

Model comparison · every step on one modelSame 218 cases

Metricrouter-smallhosted · fastreasoner-largehosted · strongestopen-weight 70Bself-hosted · in-region
Task success86.1%94.6%91.7%
Correct tool use97.4%98.3%97.9%
Cost per task$0.006$0.058$0.031
95% of tasks done in3.9 s7.4 s8.1 s
Policy violations000
Used forRoutes, classifies, extractsDrafts and reasonsFallback where data must stay in-region

Right-sized · cost and carbon per taskSCI · ISO/IEC 21031:2024 — Software Carbon Intensity

  • Every step on reasoner-large $0.0582.4 Wh est. · 94.6% success
  • Right model per step · inputs reused $0.0210.9 Wh est. · 94.2% success

SCI = ((E × I) + M) per R, a carbon score, with R = one completed task. The manual process is the limit: about 4.8 Wh a task (4.8 minutes at a workstation). An agent that does not beat it on cost and carbon does not go live.

Test history · last 4 runs · step-order checks on

  1. run 0412 renewals v13 214 / 214 checks passed · every run used tools in the allowed order release allowed
  2. run 0418 renewals v14 Step-order check failed in 6 of 25 “discount over the limit” cases. The quote was changed before the rules check, so the agent acted instead of asking release blocked expected…draftpolicy.checkcrm.update_quote got…draftcrm.update_quotepolicy.check
  3. fix renewals v14.1 Changes now wait for the rules check to pass. Code enforces this, so the AI cannot skip it · 4 step-order cases added (218) awaiting re-run
  4. run 0419 renewals v14.1 218 / 218 checks passed · 0 tools out of order · released after the Sales ops lead signed off release allowed
  • Test cases from your experts

    Your own tickets, labelled by the people who do the work. Tricky new cases from live use join every week or two.

  • AI scorer, checked

    An AI scores every case. Each release, we check it agrees with your team (Cohen’s κ 0.8 or more).

  • Attack tests

    Hidden instructions, over-limit actions and missing data. Each must be refused or passed to a person.

  • Tests before release

    Every change to instructions, tools or models runs all the tests. One failure stops the release.

06Safe AI rules

Clear AI rules that let legal say yes.

Before any agent goes live, we write down what it may do, who approves and which laws apply. It all maps to the standards your auditors use.

Rows of numbered keys hanging on a board, each on its own hook
Only the keys it needsEach permission is numbered and owned. An agent gets only the keys its job needs.Photo: tokyo rattus / Unsplash

18 documents across the 4 parts of NIST AI RMF 1.0 · frameworks we build to, with no certificate claimed

01GOVERN

Govern

Policies, roles and accountability, so every decision about AI has an owner.

  • AI use policy and principles

    NIST AI RMF GOVERN 1ISO/IEC 42001 A.2EU AI Act Art. 4

  • Roles, approval rights, who does what

    NIST AI RMF GOVERN 2ISO/IEC 42001 A.3

  • List of AI tools used

    NIST AI RMF GOVERN 1.6ISO/IEC 42001 A.4

  • Training record for owners and approvers

    EU AI Act Art. 4 AI literacyNIST AI RMF GOVERN 2

No document in this part for this framework.

02MAP

Map

Each AI idea in context: who it affects, what could go wrong, which rules apply.

  • Context sheet and EU AI Act risk tier

    NIST AI RMF MAP 1ISO/IEC 23894 §6.3 contextEU AI Act Art. 6 · Annex III

  • Risk assessment with treatment plan

    ISO/IEC 42001 6.1.2 · 6.1.3ISO/IEC 23894 §6.4 · §6.5NIST AI RMF MAP 5

  • Impact assessment where high-risk

    ISO/IEC 42001 A.5EU AI Act Art. 27GDPR Art. 35 DPIADPDP Act 2023 §10

  • Data map and lawful basis

    GDPR Art. 6DPDP Act 2023 §4 · §5ISO/IEC 42001 A.7

  • Threat model for the agent

    OWASP LLM Top 10 LLM01 · LLM06NIST AI RMF MAP 5

No document in this part for this framework.

03MEASURE

Measure

Evidence, not opinion: every risk has a test, a number and a trend.

  • Tests, test sets and attack-test results

    NIST AI RMF MEASURE 2EU AI Act Art. 15OWASP LLM Top 10 LLM01–LLM10

  • Model cards per model and version

    EU AI Act Art. 11 · Art. 13ISO/IEC 42001 A.8

  • Bias and fairness tests where people are affected

    EU AI Act Art. 10GDPR Art. 22NIST AI RMF MEASURE 2

  • Monitoring: success, cost, hand-offs, quality

    NIST AI RMF MEASURE 3EU AI Act Art. 26

No document in this part for this framework.

04MANAGE

Manage

Controls in place, kept working after launch, and a plan for when they fail.

  • How people stay in charge: the four levels

    EU AI Act Art. 14NIST AI RMF MANAGE 1ISO/IEC 42001 A.9

  • Safety check and approval rules

    OWASP LLM Top 10 LLM06NIST AI RMF MANAGE 2

  • Labelling: people are told they are talking to AI

    EU AI Act Art. 50DPDP Act 2023 §5 notice

  • Audit log and retention rules

    EU AI Act Art. 12 · Art. 26GDPR Art. 5DPDP Act 2023 §8

  • Changes, undo and incident response

    ISO/IEC 42001 A.6NIST AI RMF MANAGE 4DPDP Act 2023 §8(6)GDPR Art. 33

No document in this part for this framework.

Frameworks we build to

  • NIST AI RMF 1.0AI Risk Management Framework
  • ISO/IEC 42001:2023Artificial intelligence management systems
  • ISO/IEC 23894:2023AI risk management guidance
  • EU AI ActArtificial Intelligence Act (EU) 2024/1689
  • OWASP Top 10 for LLM ApplicationsSecurity risks in generative AI applications
  • DPDP Act 2023Digital Personal Data Protection Act, 2023
  • GDPRGeneral Data Protection Regulation (EU) 2016/679

EU AI Act · when duties start

  1. 2 Feb 2025 in forceBanned AI uses and AI skills for staffArt. 4 · Art. 5
  2. 2 Aug 2025 in forceRules for general-purpose AI modelsChapter V
  3. 2 Aug 2026 in forceAI disclosure duties and most other rulesArt. 50
  4. 2 Dec 2026 aheadMarking AI content from systems already on sale, and a new banned useArt. 50(2) · Art. 5
  5. 2 Dec 2027 aheadHigh-risk uses in Annex III, moved later by the 2026 changeArt. 6(2) · Annex III
  6. 2 Aug 2028 aheadHigh-risk AI inside regulated productsArt. 6(1) · Annex I

Regulation (EU) 2024/1689, as changed by Regulation (EU) 2026/1744 (the Digital Omnibus on AI), in force since 27 July 2026. High-risk dates moved later. From 2 December 2026, systems already on sale must mark AI content, and a new banned use starts. We follow the text in force.

What this covers

These documents are the proof an auditor or regulator asks for. Seeking a certificate is your choice. GDPR and India’s DPDP Act 2023 sit in the same map, so one list covers privacy and AI risk.

Cybersecurity & AI Trust

07How long it takes

From idea to a working agent in weeks.

Test results decide each step. Your team’s speed at checking cases and approving steps sets the pace. Times shown are typical, agreed in the workshop.

A stopwatch drawn to scale over 8 weeks, with 5 milestones. Week 1: two-day workshop. Week 2: test version on your data. Week 3: first tests and check G1. Week 5: trial with 10 users. Week 8: live use at L3 after check G2. The milestone buttons that follow describe each one.

Stepping through
Typical range

What exists on week 1

  • The long list scored and the first three chosen, each with a named owner
  • Today’s numbers for doing it by hand
  • Test space: safe copies of your data

workshop · 12 scored · 3 chosen · now 4.8 min per task · test space ready

Who decidesOwner and sponsor agree the measure

What exists on week 2

  • One system linked through MCP, read-only access
  • 40 test cases taken from real tickets
  • First runs at L1 Suggest, every step logged

test version · 1 tool · 40 cases · L1 Suggest

Who decidesThe team who does the work reviews the first outputs

What exists on week 3

  • 120 cases marked by your experts; AI scorer checked (κ 0.82)
  • Cost and time per task measured against doing it by hand
  • Gate G1, go to prove: decided on the numbers

tests · 120 cases · κ 0.82 · $0.03 per task · 95% in 8.6 s · G1 passed

Who decidesSponsor, owner and legal sign G1 on evidence, not on the demo

What exists on week 5

  • Practice run first, then L2 Draft for 10 named users
  • Safety checks, spend limits and approvals switched on
  • Weekly review of hand-offs adds to the test set (214 cases by week 7)

trial · 10 users · L2 Draft · 214 cases · 0 rule breaks

Who decidesLegal and security sign the safety rules

What exists on week 8

  • Act with approval for the whole team, after check G2
  • Action log, monitoring and guides handed over
  • Decision note: the tests for moving to L4 later

live · L3 · G2 passed · action log · guides · review in 90 days

Who decidesSponsor signs the release; the owner runs it

Model swap · testing

reasoner-large v2 on the same link

  1. 1New model on the same linkNo code change: the link points at the new model in the test space.0 h
  2. 2Tests re-run: 218 casesTwo new failures on “ambiguous contract term”, fixed with one instruction line and re-run.day 1
  3. 3Live at L3, with sign-offThe action log records the model version; the old model stays ready to switch back.day 2

Model swap · now live

Re-tested and live in 2 days

Cases re-run
218 / 218
New failures
2 found · 2 fixed
Policy violations
0
Cost per task
$0.021 → $0.017

Model version recorded in the action log · old model kept ready to switch back

  • Test space on day 1

    Copies of your data with private details hidden, kept apart, so nothing waits on live access.

  • A standard set-up

    Tool links, tests, safety checks and action logs, ready to fit your systems.

  • Test results decide

    Each stage check is a test score, so go or stop is a quick review with the owner.

08The workshop

AI strategy with the people who run the work.

Two days in week 1 with operations, legal, security, finance and the team that does the work. Together we score your AI ideas and agree a one-page plan.

Two colleagues place handwritten sticky notes on a glass wall divided into quarters
Before scoringThe long list on the wall, sorted by quarter.

The week before · a short pre-readAI agents sum up your process documents, sample tickets and systems list. Every line links to its source, so the workshop is spent deciding.

Hands sort handwritten yellow notes across a blue table
Day 1Sorting the long list.
  1. Day 1 · Morning

    How the work runs today

    Your staff walk us through 3 to 5 tasks, start to finish.

    Task maps with today’s times

  2. Day 1 · Afternoon

    The long list, scored

    Every idea on the wall, scored on value, ease, risk and time.

    A scored long list, usually a dozen

Two colleagues arrange sticky notes into columns on a whiteboard in a meeting room
Day 2The best ideas sorted into columns.
  1. Day 2 · Morning

    Risk and data, idea by idea

    Legal sets the risk, security the access, system owners the data.

    Risk level, AI limits, data gaps

  2. Day 2 · Afternoon

    Pick the first 3

    The first 3, each with an owner, a starting number and a target.

    The one-page plan, signed

AI strategy · one page · Your company · v1Illustrative

What leaves the room

  1. 1Ambition and measures

    Return about 700 hours a month to sales ops and finance by the end of Q2; cost per task under a quarter of manual.

  2. 2AI ideas · build first

    Reconcile vendor invoices · Draft renewal quotes · Triage inbound RFPs. Nine more planned by quarter.

  3. 3AI team roles

    Embedded: agents owned by the teams that run the process, with a small central platform and review group.

  4. 4Data readiness gaps

    Billing history split across two systems · 18% of invoices arrive as PDFs · customer company details incomplete.

  5. 5Budget and measures

    Build, run and review budget per AI idea; one owner and one measure each; reviewed each quarter.

  6. 6Risk level and AI rules

    Two limited-risk AI ideas must tell people they use AI; two high-risk ideas wait for a legal plan.

Agreed in the room

  • COO
  • CFO
  • CISO
  • Head of sales ops

09Project stages

Assess, prove and scale, with a check after each stage.

We move on only when the numbers hit targets set in week 1 and the named people sign.

  1. 013 weeks

    Assess

    Interviews and the 2-day workshop in week 1. Then a test version on your data, with first test scores for the top idea.

    You get

    • Ranked list of AI ideas
    • One-page AI strategy and plan
    • Business cases for the first 3
    • Test version and first scores

    AI level reachedL1 · SuggestTest version only suggests, on safe data copies

    G1 Go to prove Passed

    • A named owner, today’s numbers and first test scores for the top idea
    • Read-only data access agreed, test space ready
    • Risk level set, and the AI’s limits written down
    • Budget approved, with a cost limit per task

    Signed bySponsor · owner · legal

  2. 025 weeks

    Prove

    One agent, tested on your experts’ cases. From week 5, 10 users try it at L2. In week 8 it goes live at L3, with approval on every change.

    You get

    • Working agent and its tool links
    • Tests, test cases and attack-test results
    • Safety and approval rules

    AI level reachedL3 · Act with approvalActs on your systems once a named person approves

    G2 Go live Passed

    • All tests passed, attack tests included
    • No rule broken in any run
    • Cost per task under the limit, 19 in 20 tasks on time
    • Safety rules signed by legal and security
    • Guides and on-call handed over

    Signed bySponsor signs · owner runs

  3. 03Ongoing

    Scale

    The next agents from the ranked list, with risks tracked. A task gets more freedom only when results allow.

    You get

    • Next agents from the plan
    • List of AI tools and risks, kept up to date
    • Quarterly review against the measures

    AI level reachedL4 · Act within limitsFor named tasks, inside limits, after G3

    G3 Give more freedom Every 90 days

    • 90 days at L3 with approvers changing under 2% of actions
    • Hand-offs to people answered on time
    • No incident traced to the agent
    • One task moves up one level, with new limits

    Signed byOwner · risk · security

10What you get

Plan, agent, rules and guides, all in your accounts.

What we make for you is yours once it is paid for. Tools we already had stay ours, and you get a free, permanent licence to use them.

Tab 1 of 4Your company · AI project · handover

Strategy

Where AI fits, what comes first and what each idea is worth.

  • ai-strategy-one-page.pdf One-page AI strategy Document
  • use-case-portfolio.xlsx Map of AI ideas, ranked BoardSheet
  • roadmap-business-cases.pptx AI plan with business cases DeckSheet
  • operating-model.pdf How your team works, and who approves what Document

Handover checklist

  • Signed by the sponsor and the 3 owners
  • Source files kept in your own storage
  • Scores you can edit, formula included

Tab 2 of 4Your company · AI project · handover

Agent

The agent, in your own code store and cloud account, with its tests.

  • agents/renewals-agent/ Live AI agent or AI helper Sourceyour cloud
  • mcp-servers/ Tool links for sales, billing and documents Source
  • evals/ Tests, test cases and attack-test results Testsreport
  • infra/ Set-up scripts for your cloud account Terraform

Handover checklist

  • Code moved to your company’s account
  • Release tools, keys and cloud in your name
  • AI instructions saved with each version

Tab 3 of 4Your company · AI project · handover

AI rules

The written answers legal, security and auditors ask for, in one place.

  • guardrails/policy.yaml Safety and approval rules Policyconfig
  • ai-governance-framework.pdf Safe AI rules Document
  • ai-inventory-risk-register.xlsx List of AI tools and risks Register
  • model-cards/ A fact sheet for each model Document

Handover checklist

  • List owner named, review dates booked
  • Legal and security sign-off recorded
  • Every row linked to its evidence

Tab 4 of 4Your company · AI project · handover

Running it

Everything your team needs to run, review and change the agent without us in the room.

  • dashboards/agent-actions Log of every agent action Dashboard
  • runbooks/ Guides: hand-offs, undo, model change Document
  • training/ Training for owners, approvers and reviewers Sessionsrecordings
  • reviews/q-review-template.pdf Quarterly review pack Report

Handover checklist

  • Support rota and contacts agreed
  • Approvers trained and named in the policy
  • First quarterly review in the calendar

Each tab shows a mock of the first file in that part of the handover. For strategy, an AI plan by quarter. For the agent, the code layout. For AI rules, the AI list and risk register. For running it, a hand-off guide.

1190-day review

6 measures decide whether an agent stays.

We review every agent on 6 measures. Each has a pass mark and a hold mark. An agent that misses its hold mark moves back a level. The marks shown are targets.

Agent review · day 90

renewals-agentDraft renewal quotesowner Sales ops lead

Measure Hold and pass marks At review Status

Hours returned per week

Time saved per task × volume, against today’s numbers taken before the build

Hold < 30 hPass ≥ 60 h

74 h

Pass

Task success in tests

The test set, re-run on every change and at each review

Hold < 85%Pass ≥ 92%

94.2%

Pass

Cost per task vs manual

Models, tools, hosting and review time per completed task, as a share of manual cost

Pass ≤ 25%Hold > 50%

13% · $0.62 vs $4.80

Pass

Hand-off rate

Share of runs handed to a person, by reason

Pass ≤ 10%Hold > 20%

6.8%

Pass

Policy violations

Changes outside the rules, from the action log and attack tests

Pass = 0Hold ≥ 1

0

Pass

Carbon per task vs manual

Carbon estimate per completed task, as a share of the manual process on the same power grid

Pass ≤ 50%Hold ≥ 100%

19% · 0.9 vs 4.8 Wh est.

Pass

Decision

Stays at L3L3 · Act with approval

Review in 90 days whether quotes with discounts of 5% or less can move to L4, inside limits.

Signed · ownerReviewed · risk and securityLogged · audit trail

Illustrative reviewWe agree the marks for each agent before we build. Cost includes the time people spend checking.

Offers & plans

Find where AI pays off, then build agents to do the work.

We rank your AI ideas, build the agents and AI helpers, and set the rules to run them.

Categories
04
Offers
15
Plans
04
Not sure what you need? Tell us the problem

How to buy

  1. 01Pick offers. Ask about one, or add several to one enquiry.
  2. 02Choose a plan. A small job, a fixed project or ongoing help.
  3. 03Send your enquiry. We reply within one working day.

Browse by category

Timelines are typical. We write the scope before we quote.

01AI strategy

4 offers
Typical timeline: 3–4 weeks

Find where AI helps

We study your work, data and systems with your teams. Then we show where AI would cut time, cost or risk.

What’s included

  • Leadership and team interviews
  • Process walk-throughs and data review
  • Map of AI ideas by value and ease
  • Data and systems check
  • Discovery
  • Opportunity map
  • Readiness

Best forLeaders under pressure to act on AI who are unsure where to start.

Typical timeline: 2–4 weeks

Ranked AI ideas & business case

We score each AI idea on value, ease, data and risk. Then we rank them in a costed plan with owners and success measures.

What’s included

  • Scores for value, ease and risk
  • Cost and benefit for each idea
  • A 12-month plan
  • Measures and starting numbers
  • Scored
  • Ranked
  • Business case

Best forTeams with many AI ideas and money for only a few.

Typical timeline: 2–4 weeks

Choosing AI models

We test AI models on your own examples for quality, speed, cost and where data is kept. Then we pick one per task.

What’s included

  • Shortlist from OpenAI, Anthropic, Google, Mistral and open models
  • Side-by-side tests on your tasks
  • Cost and speed per task
  • Data location and contract terms
  • Any model
  • Tested on your data
  • OpenAI

Best forTeams unsure which AI model or supplier to pick.

Typical timeline: 1–2 weeks

Leadership AI briefing

A working session for your board or leaders. We show what AI can and cannot do today, in your sector and on your data.

What’s included

  • Notes to read beforehand
  • Live demos on similar tasks
  • AI risk and rules basics
  • Agreed next steps
  • Workshop
  • Leadership

Best forBoards and leaders starting their AI plan.

02Agents & AI helpers

4 offers
Typical timeline: 8–12 weeks

Custom AI agent

Software that does tasks with several steps in your systems, such as checking an order. It passes hard cases to a person.

What’s included

  • Task design with limited tool access
  • Connected to your systems and tools
  • A person approves risky actions
  • A log of every action
  • Tests before going live
  • Tool use
  • Human approval
  • Action log
  • OpenAI

Best forTeams with repetitive, rule-based work across several systems.

Typical timeline: 6–10 weeks

AI helpers for staff

An AI helper in Slack, Teams or other tools your staff already use. It answers from your own documents and data.

What’s included

  • Linked to approved sources
  • Sourced answers that respect access
  • In Slack, Teams or a web app
  • Use and feedback reports
  • Your data
  • Sourced
  • In your tools
  • Microsoft Teams
  • Slack
  • LlamaIndex
  • OpenAI

Best forSales, support, HR and operations teams looking for answers all day.

Typical timeline: 10–16 weeks

AI agent teams

Several agents working together on a larger task, such as research, drafting, checking and filing. A person checks key steps.

What’s included

  • Task steps and agent roles
  • Hand-offs that recover from errors
  • Review points for people
  • Cost and step limits
  • Teamwork
  • Checkpoints
  • Limits

Best forLarger tasks that already follow written steps.

Typical timeline: 4–8 weeks

Agent links to your systems

We connect agents to your systems through MCP, a standard link for AI tools. Each task gets only the access it needs.

What’s included

  • Tool and MCP link design
  • Logins and limited access
  • Use and spend limits
  • Every tool use logged
  • MCP
  • Minimum access
  • Tool logs

Best forCompanies that want agents to act in their systems with limited access.

03Trials & testing

3 offers
Typical timeline: 6–12 weeks

AI trial with tests

A short, fixed trial of one AI idea, measured against today’s numbers. You decide go or stop on the results.

What’s included

  • Today’s numbers, before the trial
  • Task tests built with your experts
  • Practice run first, then supervised use
  • Decision note and launch plan
  • Start number
  • Practice run
  • Go or stop

Best forCompanies ready to trial one AI idea first.

Typical timeline: 3–6 weeks

Automatic AI testing

Tests that score your AI on accuracy, tasks done, safety and cost. They run every time an instruction or model changes.

What’s included

  • Test sets from your own examples
  • Automatic scoring, checked by people
  • Attack tests for tricks and tool misuse
  • A test check before every release
  • Every change tested
  • Before vs after

Best forTeams with live AI that cannot tell whether a change made it better or worse.

Typical timeline: 6–10 weeks

Trial to launch

We make a promising test version dependable: tracking, backups, access rules, cost limits and support.

What’s included

  • Review of the build
  • Safety checks, backups and use limits
  • Tracking of quality, speed and cost
  • Guide and handover
  • Made reliable
  • Tracking
  • Cost limits
  • AWS
  • Microsoft Azure

Best forTeams with a trial that works in demos but is not ready for customers.

04AI rules & team roles

4 offers
Typical timeline: 4–8 weeks

Safe AI rules

Rules on which AI uses are allowed, who approves them and how they are tracked. Mapped to NIST AI RMF and ISO/IEC 42001:2023.

What’s included

  • AI policy and staff rules
  • Risk levels and approval steps
  • List of AI tools in use
  • Impact check template
  • NIST AI RMF
  • ISO/IEC 42001
  • Policy

Best forCompanies bringing AI into several teams at once.

Typical timeline: 3–6 weeks

AI team roles & skills

Who owns AI in your company: one central team, experts in each unit or both. Then the roles, skills and budget to match.

What’s included

  • Set-up options and our pick
  • Roles and who decides
  • Skills gap and hiring plan
  • Funding and ranking process
  • Central AI team
  • Roles
  • Skills

Best forCompanies moving from scattered AI tests to one managed plan.

Typical timeline: 3–6 weeks

EU AI Act & other laws

With your lawyers, we map each AI use to its EU AI Act risk level and duties. We add GDPR and India’s DPDP Act 2023.

What’s included

  • Risk level per AI use
  • Duties for each AI system
  • Gap list with named owners
  • Readiness plan
  • EU AI Act
  • DPDP Act 2023
  • GDPR

Best forCompanies selling into Europe or using personal data with AI.

Typical timeline: 2–6 weeks

AI training for staff

Training so your staff use approved AI tools well and safely. How to ask, how to check results and when not to use AI.

What’s included

  • Training by role
  • Asking and checking guides
  • Safe-use guidelines
  • Usage tracking
  • Training
  • Guides
  • Safe use

Best forStaff who have AI tools but no guidance.

Your enquiry

Tick “Add to enquiry” on any offer, choose a plan, then continue. Or enquire about one offer.

How we work with you

Ways to start, from a question to a quote.

Ask a quick question, send your project details or request a quote. The person who will lead the work reads each one in full. Services in your enquiry go with it.

Or book a 30-minute call

What are you sending?

  1. 01

    About 2 minutes3 short answers

    For a first chat, a press request or anything not ready to plan yet.

    You get A reply from a named lead

  2. 02Recommended

    About 8 minutes5 short steps

    Goals, who it is for, budget and timing, so our first reply can outline the work.

    You get Options and prices after one call

  3. 03

    About 15 minutesYour documents attached

    Your documents, deadlines and the purchasing and security rules to meet.

    You get Confirmation and a named lead

How it is priced

Each plan shows how it is priced. Work starts once we agree a written scope and price.

  • Typical length
    1–3 weeks
    Pricing
    Fixed fee
  • Typical length
    4–12 weeks
    Pricing
    Fixed price
  • Typical length
    3–9 months
    Pricing
    Fixed price per stage
  • Typical length
    6–18 months
    Pricing
    Priced per written scope
Compare what each plan includes
What each plan includes and who it suits
PlanWhat you getBest for
SprintA short, fixed job that answers one clear question.
  • Scope and result agreed before we start
  • A senior lead and the right specialists
  • A weekly review with you
  • A clear answer or a test version
An audit, a brand check or a test version of an app
ProjectOne clear job for one fixed price.
  • A written scope: what you get and how we agree it is done
  • A named project lead and a fixed team
  • A shared plan with checkpoint dates
  • Source files, yours once paid for
A brand identity, a website, a simple app or a set of tools
MilestoneA bigger build in stages you approve and pay for one at a time.
  • Stages, each with its own scope and approval
  • A go or stop review at each stage
  • Re-plan between stages as you learn
  • Pay for each stage you approve
Work too big for one contract, such as a large app
EnterpriseA large, long project with its own team and agreed response times.
  • A project director and a decision group
  • A team for each part of the work
  • Response times, reports and a list of risks
  • Security, legal and purchasing checks in the plan
Large companies making changes across markets, brands or business units

12Questions

What people ask about AI agents.

What leaders, legal and security teams ask in the first meeting.

Bring to the first call

  • The tasks you want to change, with rough weekly volumes
  • Who owns each one, and who approves changes to it
  • The systems and data each task uses
  • How much risk you accept, and your regulators
Book a call

Start with one busy task that has an owner and usable data. Rule-based back-office work, such as invoice matching or quote drafting, beats open-ended customer chat. We score your ideas, then trial the best one with the agent only suggesting or drafting.

We pick per task from OpenAI, Anthropic, Google, Mistral and Llama, on quality, speed, cost and where data is kept. One link reaches them all, so you can switch later. Inside one agent, simple steps use a small, fast model and hard steps a larger one.

Only at level 4, for named tasks and inside limits you set. Limits cover spend, records changed and business rules, such as a top discount. Anything outside goes to a person. A task gets there only after 90 days at level 3 with few corrections.

Mainly with tight permissions and approvals, as no filter catches every attack. We treat every email and document as untrusted. Each agent has limited access, and risky actions wait for a person. We attack-test each release against OWASP’s list (LLM01, LLM06).

Three things: model use, hosting and the time people spend reviewing. For a back-office task, model use is often a few cents. We track cost per task from the first tests, against doing it by hand.

Both can apply, and we map them in one list. The EU AI Act covers AI offered or used in the EU, or whose output is used there. Most business agents fall in its 2 lowest risk levels. Duties to say when AI is used start in August 2026. After a 2026 change, duties for high-risk uses, such as hiring or credit checks, start in December 2027. India’s DPDP Act 2023 covers personal data of people in India. Under the DPDP Rules 2025, most duties start in May 2027. This is guidance, not legal advice.

Tell us what you need built.

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

Book a call

Three ways to start

We always agree a written scope and price before work begins.

Choose one of the three ways above