A bid for the Department for Business & Trade AI Capability Programme

Zühlke's proposal to help DBT use AI safely.

A 24-month programme to help DBT's 3,200 staff use AI safely in real work: skills, governance, culture and measurable adoption.

Two phases · 24 months · £7.5M · Proof before scale

What DBT asked suppliers to solve

Help 3,200 people use AI safely in daily work.

DBT: the UK Department for Business & Trade
~3,200 staff · 14 UK offices · 8 overseas posts
Two phases over 24 months — £7.5M
Not a software purchase: skills, governance, culture
Measurable proof by Dec 2026; the annual report, April 2027

The lesson from Meridian, DBT's previous programme

Meridian cost £2.2M. Six months later, capability had evaporated.

of 3,200 staff still learning, 6 months on

active champions left, of ~98 at peak

71%

used tools exactly as before

234

views of the governance framework — adopted by no one

Why Meridian matters: DBT has already bought activity that did not change behaviour.

Activity is not
capability.

Capability means people repeatedly doing real work better and safer. Meridian did not fail through effort. It failed because training, governance and ownership were not built into the work.

Why the last programme did not change behaviour

The fixes are built into our delivery model.

Training sat alongside the work, not inside it
Senior leaders didn't show up (~12% director attendance)
A 180-module catalogue — too broad to be relevant
No consequence for opting out
Champions volunteered, unfunded, unsupported — the network collapsed on exit

Our answer

Make AI useful, governed and owned by DBT.

Embed in the work

We redesign real workflows from the inside — capability built into how the task is done, never standalone training.

Govern as we go

Governance grows from the live use cases teams actually run — so it's usable, and used. We adopt and tailor the AI Playbook, not a shelf document.

Design our exit

Named DBT owners, a funded community of practice, a month-18 sustainability gate. What we build survives us.

The commercial difference

We tie our fee to change that lasts.

Meridian was paid for a catalogue. We're asking to be paid for behaviour that's still there after we leave. We set a baseline in the first weeks — then a slice of our Phase 2 fee, and the decision to continue, ride on adoption that persists.

If the change isn't real, you shouldn't pay us as though it were.

Phase 1 · Prove the approach · months 1–6 · ~£1.5M

Six months to prove value and de-risk scale-up.

Weeks 1–2

Mobilise. Lock in Director-level sponsors. Set the baseline — you can't prove value without it.

Weeks 3–8

Refresh your 2022 segmentation. Discovery across directorates. A shadow-AI audit — surfacing real risk and real value.

By day 90

A governance framework people actually follow, and live lighthouse pilots in the directorates that want it.

Deliverables expressed as outcomes, not documents. A clear go / no-go at month 6 — you are not locked in.

Phase 2 · Scale what works · months 7–24 · ~£6.0M

Eighteen months to scale proven patterns.

Embedded squads — joint Zühlke + DBT pairs redesign real workflows in situ
A community of practice with a funded coordinator, protected time, self-selected champions
Exit by design — a named owner per workstream; a month-18 sustainability gate
Governance matures with use; outcomes measured six-monthly

Adoption sequence

Start with likely wins, then bring cautious teams in safely.

Investment and your data analysts go first — that's where proof comes fastest. We approach Trade Policy and Legal last, on their terms: accuracy-first, the human expert always in the loop.

We treat the 2024 Legal incident as the reason to be careful everywhere — not a footnote.

Why this fits the current government context

Use the AI Playbook and public-sector evidence.

We build on the government's AI Playbook (Feb 2025) — not a framework we invent
We learn from what works: Consult analysed 50,000 responses in ~2 hours; Caddy ~halved response times
The Public Accounts Committee: ~70% of departments struggle for AI skills

You're facing a national challenge — with a method to answer it.

Proof you can see — not discovered at the end

What success looks like by December 2026.

Investment

18% → 41% · target 55%

Data & Analysis

35% → 62% · target 75%

Ministerial & Strategy

11% → 22% · target 45%

Trade Policy (last, on purpose)

5% → 6% · target 25%

Sustained active use in named teams, against a baseline, on a live dashboard. Dashed line = where you started.

Built into delivery, not bolted on

Social value that's specific.

Apprenticeships & early-careers hiring near your 14 UK offices
A remote-first, low-carbon programme with a Carbon Reduction Plan
Deliberately bring the 54% with least confidence along — no two-tier department
Augment, not replace — an honest promise to staff, made with the unions

The social value commitments are delivered through the programme itself.

The delivery team

A team that tapers.

Not AI experts seeing a department for the first time — people who understand how a large, policy-heavy organisation actually changes.

Programme & Delivery · Engagement · Technical & Data · Discovery & Measurement · Capability & Engineering · Enablement & Tooling

The team shrinks as your capability grows. We get cheaper as you get better.

Programme risks we will manage openly

We'll name the hard things.

Staff resistance — Trade Policy and Legal especially
Union concerns if AI reads as job cuts
Whether Director-level sponsorship really materialises
Even the ways our own fee-at-risk model could be tested

They're written down, with mitigations — because the risks that sink programmes are the ones nobody said out loud.

The test we're happy to be judged on

Real AI capability, owned by DBT.

In twelve months: named teams doing real work measurably better, safely, and still doing it after we've stepped back. Less dependency over time is the point.

Zühlke. DBT AI Capability Programme