contextric Book a call

Selected work.

Admin reviews, engineering reviews, AI training, AI enablement and independent supplier reviews, alongside more than 25 years of delivering technology for UK Government and FTSE 100 companies.

  1. AI review of the admin function for a food manufacturer
  2. Agentic engineering review for a global systems integrator
  3. AI training for an engineering team of 500+
  4. AI capability discovery for a large engineering organisation
  5. Independent review of an AI reporting tool for a wealth manager
Client
Food manufacturer making traditional baked snacks
Scale
60 staff
Need
Find where AI could make the admin function more efficient
Role
Independent reviewer and trainer
Duration
A day with the admin team, then one-to-one training

AI review of the admin function for a food manufacturer

The business makes traditional baked snacks and employs around 60 people. Its admin function handles orders, sales and invoicing, and produces the cook and wrapping lists that tell the factory what to make. Leadership wanted to know where AI could make that work quicker and more reliable.

What we did

James spent a day alongside the admin lead, following real work through order intake and entry, sales and invoicing, and the business's own process for producing cook and wrapping lists for the factory staff.

We mapped where the time went, where errors crept in and had to be put right, and which steps AI could take on, with a person checking anything important. Alongside the review, James ran one-to-one AI training with the CEO, focused on leadership work.

What we found

  • Order intake and entry, sales and invoicing all had steps that AI could take on or speed up
  • The bespoke process for producing cook and wrapping lists was a strong candidate for automation
  • Errors in manual steps were causing rework that AI-assisted checks could reduce
hours a week saved, estimated
11 to 22
staff
60
day with the admin team
1

Outcome

  • Recommendations for each process, with estimated savings of 11 to 22 hours a week and less rework from errors
  • One-to-one training for the CEO, after which the CEO's use of AI for leadership tasks increased significantly
Client
Global systems integrator
Sector
Public sector, legacy modernisation
Scale
6 projects, around 150 people
Role
Independent reviewer, reporting to the leadership team
Duration
Six weeks

Agentic engineering review for a global systems integrator

The client was modernising legacy systems for several public sector customers, and building its own AI engineering capability alongside. Each project had adopted AI tools in its own way, costs were rising, and leadership wanted to know what good looked like and how far they were from it.

What we did

We worked with people in every role, from architects to engineers, business analysts and testers, sitting in on ceremonies and following real pieces of work from requirement to release.

We looked at how each team used AI: which tools and models, what context they gave them, and what came back. Then we compared it with current practice in agentic engineering.

What we found

  • The most capable, most expensive models were used for every task, including simple ones
  • Agents were given whole repositories as context, so most tokens went on code that didn't matter
  • Each project had built its own prompts and workflows, with no way to share what worked
  • Analysts and testers were largely left out, so gains in coding were lost further down the line
  • Nobody was measuring what AI delivered, only what it cost
lower token spend, pilot teams
40%
more delivered per sprint, pilot teams
25%
projects reviewed
6

Outcome

  • A report for the leadership team with prioritised recommendations: models matched to tasks, scoped context, a shared playbook, analysts and testers in the workflow, and simple measures of value
  • Pilot teams cut token spend significantly while delivering more
  • The recommendations became the basis of the organisation's AI engineering standards
Client
Organisation with a 500+ strong engineering team
Scale
500+ engineers
Role
Designed and delivered the programme, then trained the trainers
Duration
Nine months

AI training for an engineering team of 500+

The organisation had rolled out AI coding tools to more than 500 engineers, but adoption was patchy: some teams used them every day, others had barely opened them. Leadership wanted every engineer confident with AI, quickly, and in a way that would last.

What we did

We designed a bespoke programme around the client's own tools, codebase and policies. Each cohort started with two weeks of supported self-development, with short exercises and drop-in help, then came together for a hands-on classroom day: prompting for engineers, AI in the IDE, agentic workflows, reviewing AI-generated code, and security and policy.

James delivered the classroom days in person to more than 200 engineers. We then trained a group of the client's own engineers to deliver it, with materials, trainer notes and observed sessions, so the programme could reach everyone.

engineers trained in person
200+
average course rating
4.5/5
cohorts
14
internal trainers
8

Outcome

  • Extremely positive feedback from attendees after every cohort
  • Leadership saw the training as key to increasing AI adoption and understanding
  • Handed over to internal trainers, who now run it for new joiners
Client
Large engineering organisation
Scale
30+ interviews across 12 roles
Role
Discovery and training review
Duration
Three weeks

AI capability discovery for a large engineering organisation

Part way through an AI enablement programme, the organisation needed an honest picture of where its people really were: what they used AI for, what held them back, and whether the training was working.

What we did

Over three weeks we interviewed more than 30 people, from junior staff to senior leaders, across engineering, operations and support functions.

We reviewed the existing AI training and mandatory modules, usage data and the programme's budget, and set what we found against what people said they needed.

What we found

  • Novel, valuable uses of AI in several roles, invisible to the rest of the organisation
  • Mandatory training that was generic, so completion was high but confidence was low
  • No owner for usage-based AI costs, so some teams overspent while others held back
  • A strong appetite to learn, especially from colleagues
people interviewed
30+
roles covered
12
weeks
3

Outcome

  • Role-based learning paths to replace the generic mandatory module
  • A champions network to spread the best uses of AI
  • Clear ownership and reporting for AI spend
  • A plan for the next stage of the enablement programme
Client
Wealth management company
Sector
Financial services
Need
Decide whether to buy an AI tool that drafts client reports for financial advisers
Role
Independent reviewer
Duration
Two days

Independent review of an AI reporting tool for a wealth manager

The company was close to buying a package that used AI to draft client reports for its financial advisers. The supplier's pitch rested on a novel approach to security, which it said made it safe to send client data to AI models. Before committing, the company wanted an independent view of whether that was true.

What we did

We reviewed the supplier's technical documentation, security architecture and data flows, and tested what the sales material claimed against the evidence behind it.

We asked for the assurances a regulated firm should expect: independent certification, penetration test results, and clear answers on where client data goes and who can see it. We set what we found against FCA expectations and UK GDPR.

What we found

  • No independent evidence that the security approach worked: no certifications, audits or penetration tests to back it up
  • Several claims in the sales material were misleading, including what client data left the company's environment
  • The novel approach added complexity without reducing risk compared with standard, proven controls
  • No clear answer on where client data was stored and processed, or which AI models it was sent to
  • No way for advisers to check the accuracy of AI-drafted reports before they reached clients
  • The contract gave the company little protection if things went wrong
days
2
key findings
6
alternative products recommended
2

Outcome

  • Advised the company not to proceed with the package
  • Recommended two alternative products already established in financial services
  • Set out best practice for using frontier AI services safely without novel security measures: enterprise agreements that rule out training on client data, UK or EU data residency, access controls and audit logging
  • A clear set of questions for the company to ask any AI supplier in future

Book a free 30-minute intro call.

Tell us what you're working on. You'll get an honest view of what would help and what a sensible first step would be. No obligation.

Choose a time

Or email hello@contextric.com