All field notes

The contract is where outsourcing risk is either handled or hidden. This checklist covers scope and acceptance, IP, data protection, security, change control, exit, liability, and governing law for AI and software engagements.

Collections operates under consumer protection rules on contact, conduct, and the treatment of people in difficulty. This guide covers contact strategy within those rules, detecting vulnerability and routing it to trained staff, handling disputes properly, and why the regulatory constraints shape the system rather than surrounding it.

HVAC service lives or dies on first-time fix, which depends on sending the right engineer with the right part. This guide covers dispatch with skills and certification matching, predicting the part before the visit, maintenance contract profitability, quoting accuracy, and why system design and safety decisions stay with qualified engineers.

Cleaning contracts are won on price and lost on quality complaints, with margin decided by labour hours per site. This guide covers scheduling across sites and shifts, verifying quality without constant supervision, planning consumables, and understanding contract profitability at site level rather than in aggregate.

Landscaping is weather-dependent work delivered across many small sites under seasonal demand swings. This guide covers weather-aware scheduling that survives contact with reality, route planning, seasonal labour planning, quoting accuracy, and evidencing contract delivery to clients who do not see the work happen.

Pest control combines routed service delivery with regulatory record-keeping about treatments and substances. This guide covers scheduling across routes and priorities, treatment records that satisfy audit, monitoring data from connected devices, recurring infestation analysis, and why treatment decisions stay with qualified technicians.

Security services run on coverage schedules, alert volumes that exceed human attention, and incident records that are frequently unusable. This guide covers scheduling to contracted coverage, triaging alerts to reduce false alarms, improving incident report quality, and why response and detention decisions remain human.

Most AI projects are lost in the first two weeks, when outcomes stay vague and access never arrives. This checklist covers what must be true at kickoff: outcome, owner, data, access, security, evaluation plan, team, and cadence.

An AI RFP built from feature checklists rewards marketing. One built to reveal method rewards the vendors who will deliver. This checklist covers outcome, context, method questions, security, operations, terms, artifacts, and scoring.

Service levels for AI systems must cover quality and change, not only uptime. This checklist covers availability, latency, quality thresholds, oversight response times, incident handling, change notification, reporting, and remedies.

AI teams fail from the wrong roles at the wrong time and from hiring on credentials rather than production evidence. This checklist covers roles by phase, evidence to require, candidate evaluation, domain fit, build-versus-augment, and onboarding.

Models inherit every flaw in their data. This checklist covers provenance and rights, representativeness, labeling quality, leakage prevention, privacy, splits and versioning, and documentation for training and evaluation data.

Buying AI means buying a method and a team as much as a product. This checklist gives buyers the questions and evidence to demand across capability, method, security, IP, operations, stability, and references.

The EU AI Act applies to many organizations outside Europe. This checklist walks through scope and role, risk classification, prohibited practices, high-risk obligations, transparency duties, general-purpose model duties, and timelines.

A forward deployed engineer is only as effective as the first two weeks allow. This checklist covers mission scope, business owner, access, embedding, discovery plan, specification, and the first review, for both the client and the engineer.

Palantir invented the forward deployed engineer. How the model worked, why it spread across the AI industry, and what it means when you hire one today.

The real reasons US companies outsource AI—talent scarcity, cost, and speed—and how to do it without losing coverage, security, or control of your IP.

Evaluation is how a probabilistic system is proven correct. This checklist covers specification coverage, the golden dataset, task metrics, calibrated graders, safety tests, CI gates, production sampling, and reporting.

Fairness is a tested property, not a stated value. This checklist covers scope and protected groups, metric selection, data assessment, pre-launch and production testing, remediation, documentation, and governance for AI systems that affect people.

AI governance is a set of structures and controls, not a policy document. This checklist covers the register, ownership, policies, risk tiering, evaluation, oversight, audit, vendor risk, and cadence an enterprise needs.

Chambers serve members with wide-ranging needs, issue export documentation, and hold local economic knowledge nobody else has. This guide covers member enquiry handling, export documentation accuracy, event and networking operations, local economic intelligence, and why advocacy positions stay member-led.

Childcare is administration-heavy, ratio-constrained, and built on parent trust. This guide covers reducing the administrative load on practitioners, planning staffing against ratio requirements, improving parent communication, and why safeguarding and any concern about a child must route immediately to qualified people.

Credit bureaus hold data that determines people's access to credit, housing, and sometimes employment. This guide covers identity matching and data quality, dispute handling that treats consumers fairly, fraud signals, and why fairness testing and explainability are obligations rather than good practice in this sector.

Driving schools sell instructor hours against pupils whose progress is uneven and whose test readiness is judged subjectively. This guide covers scheduling across instructor and pupil availability, tracking progress consistently, supporting test readiness conversations, and why the readiness judgement belongs to the instructor.