Closing the Tech Opportunity Gap With Data, Not Slogans

Programmes like ours attract a great deal of goodwill and very little scrutiny. That is a bad combination. This piece sets out where representation actually breaks down in the technology pipeline, which interventions have measurable effects, and the numbers by which BTMP should be judged — including the ones that do not flatter us.

The pipeline metaphor is wrong, and it matters

'Pipeline problem' implies a single channel with an input and an output, where the fix is to push more people in at one end. The metaphor survives because it is convenient: it locates the problem before the employer, and therefore outside their responsibility.

What the data consistently shows is not a narrow pipe but a series of exits, distributed across the whole length of a career. People leave at school, at university, at first application, at eighteen months in role, at the first promotion gate and at the return-to-work point after caring responsibilities. Each exit has a different cause, and an intervention aimed at the wrong one produces no measurable effect whatever its budget.

Read that table as a diagnostic. A programme that runs school talks and then reports on hiring outcomes is measuring an exit it did not address.

What works, what does not, and what we cannot tell yet

The evaluation literature on diversity interventions is uneven, but some patterns are robust enough to act on.

Interventions with reasonably consistent evidence

• Structured mentoring with a written plan and review cadence. Effects on retention and progression appear across multiple studies. Unstructured mentoring shows much weaker and less consistent effects.
• Sponsorship — someone spending credibility on your behalf. Strong association with promotion. Also the intervention most unevenly distributed in the first place, which is part of the mechanism.
• Financial support that removes a specific barrier. Connectivity, hardware and stipends show clear effects on completion, because they address an actual binding constraint rather than a motivational one.
• Structured interviewing with published criteria. Reduces variance in evaluation. Unstructured interviewing is close to a coin flip with a confidence interval attached.
• Sustained exposure over years, not events. Repeated contact with practitioners changes subject choice; single assemblies do not.

Interventions with weak or contested evidence

• One-off unconscious bias training. Reliable short-term attitude shifts, little durable behaviour change, and some evidence of backlash effects when mandatory.
• Short bootcamps without placement support. Completion is not the bottleneck; the transition to employment is.
• Awareness campaigns without structural change. Visibility without a changed process tends to move applications rather than offers.
• Diversity targets without allocation audits. Hiring targets met while work allocation stays uneven produce recruitment gains and retention losses.

How BTMP measures itself

We publish four numbers annually. They were chosen before we knew what they would say, which is the only way this exercise means anything.

1. Completion rate by track. Not overall — by track, because averaging hides the track that is failing.
2. Placement rate at twelve months. Twelve months, not three. Three-month figures measure how motivated people are immediately after a programme; twelve-month figures measure whether anything durable happened.
3. Mentor renewal rate. Our best proxy for whether the mentoring experience is actually good rather than merely well-intentioned.
4. Shipyards phase progression, including deliberate stops. Reported separately from dropouts, because a founder who correctly decides not to continue is a success and a founder who disappears is not.

The numbers that do not flatter us

• Completion in the Extended Reality track sits well below our other tracks. The most likely cause is hardware access; we are testing a loan expansion and will report whether it moves.
• Placement for career-changers over forty is meaningfully below that for under-thirties with equivalent portfolios. That gap is in the employer layer, not ours, but reporting it is our responsibility.
• Mentor renewal in the first cohort of any new track runs low, because the first cohort of anything is an experiment the mentors did not sign up for.
• We have never successfully measured what happens to people who apply and are not accepted. This is the largest hole in our evaluation and we have no good plan for it yet.

How to hold a programme like ours to account

If you are considering joining, funding or partnering with any programme in this space, these are the questions we would want asked of us.

1. What is your twelve-month outcome, and how is it defined? 'Placement' can mean a full-time role, an internship, or a freelance invoice. Ask which.
2. What is your completion rate, and what happens to people who do not complete? A programme that reports only on completers is reporting on survivors.
3. What have you stopped doing? Any programme running for five years that has never shut down an initiative is either extraordinarily lucky or not measuring.
4. Who is not served well by your model? Every design has a population it fits badly. A programme that cannot name theirs has not looked.
5. Can I speak to someone who left? The most informative conversation available, and the one programmes are least willing to arrange.

We answer all five publicly, and the fifth is genuinely uncomfortable. It should be.

What employers can actually do this quarter

Most of the exits in the first table are inside employers, not inside training programmes. Five changes that are within a single engineering organisation's control and that show measurable effect:

• Audit work allocation, not just headcount. Who gets the visible projects, the incident leads, the customer-facing work? Promotion follows visibility. This audit takes a day and is frequently the single most revealing thing a manager does.
• Publish promotion criteria and worked examples. Not a ladder document nobody reads — actual anonymised examples of work that met each bar.
• Assign a mentor and a sponsor at onboarding, separately. Naming both roles makes the sponsorship explicit rather than accidental.
• Structure your interviews and keep the score sheets. Then look at the score sheets quarterly for patterns by interviewer. Some of what you find will be uncomfortable.
• Instrument your own funnel. Applications, screens, onsites, offers, acceptances, eighteen-month retention. Most organisations discover their problem is not where they assumed.

Almost every organisation we work with believes its problem is at the top of the funnel. Almost every one that instruments the funnel discovers it is at eighteen months.

Terrence Ade, Head of Pipelines, BTMP

Why none of this is charity

It is necessary for the survivability of technology itself that every culture be afforded a part to play in this great build. That is a value statement, and it is also a practical one.

Narrow teams produce predictable, expensive failures: products that miss obvious market gaps, systems that are inaccessible to a large fraction of their users, and models trained on data whose limitations nobody in the room was positioned to notice. Each of those is a commercial cost, and each recurs because the same perspective gap recurs.

The argument for programmes like BTMP does not rest on generosity. It rests on the observation that the current allocation of opportunity is producing worse technology than it needs to, and that this is fixable by unglamorous, measurable means: structured mentoring, explicit sponsorship, removing binding financial constraints, and auditing the processes that quietly sort people.

None of that requires believing anything in particular about anyone. It requires measuring, publishing and changing what does not work — including, when the numbers say so, our own programme.

Frequently asked questions