How do police training providers set their prices?

Police training providers don't usually open up about pricing. This piece walks through the five factors that drive what training actually costs: synchronous vs asynchronous delivery, off-the-shelf vs bespoke, the cost of running the business, user vs organisation licensing, and the years of expertise baked into the design. A separate section covers what generative AI is changing about each, with a recent example where we built a complete draft programme from a 200-page piece of statutory guidance in hours using a bespoke eight-agent team.

Most training providers won't talk publicly about how their pricing is built. This is that conversation. Five factors determine what a police training product or programme costs, and getting any of them wrong in a brief produces a procurement that won't survive scrutiny. Generative AI is now shifting several of those factors, which is covered after the five.

Factor 1: Asynchronous or live?

Asynchronous, online-only training is build-it-once-deploy-it-many-times. The development cost is large; the delivery cost per learner is small. Live training, in classroom or online, builds in a recurring cost: facilitator time, every cohort, every session.

This is the first thing to clarify in a specification. If a brief asks for a 'flexible blended programme' without naming where the live and asynchronous components sit, the pricing conversation has to start from scratch. Designers can build either; the cost shape is fundamentally different.

Factor 2: Off-the-shelf or bespoke?

The supermarket analogy works. Waitrose carries several times more product variation than Aldi or Lidl. Aldi and Lidl's operating model is built around limited choice, and that's how they keep prices low. Waitrose customers are paying, partly, for the range.

The same applies to police training. If a programme has to be custom-built to a specific force's requirements, the cost reflects the design, build, and IP transfer. If something off-the-shelf meets most of the specification, the price is lower because the creator absorbed the development cost in advance and is recovering it across multiple buyers.

The trade-off is real. Bespoke gives you exactly what you asked for and the IP at the end. Off-the-shelf gives you something proven, at a lower price, in less time, but you don't own it.

A surprising number of briefs ask for bespoke when off-the-shelf would have done the job. That's a procurement conversation worth having early. Generative AI is starting to shift this calculation in a substantial way; see the AI section below.

Factor 3: The cost of doing business

When I was a police officer, I had no idea what running a business cost. I had worked on a neighbour's potato farm, in a village shop, cleaning loos in a mountain hotel, and pulling pints by the Thames. None of that prepared me for running a consultancy. I'd guess most senior police buyers are in roughly the same position.

Costs of doing business break into three categories.

Fixed direct costs of the product. For a digital training product, this is the narrator, the subject matter experts, the software licensing, the image licensing. These costs are incurred once during development. They scale with the size and complexity of the product, not the number of learners.

Indirect costs. Salaries, NI, pensions, accountant, accreditations (Cyber Essentials, ISO 27001), software for the business, office. None of these increase per learner. They are the cost of being a real, compliant, going concern. At a specialist consultancy, the biggest line in this column is people.

Cost of goods sold. These rise with the number of learners. User licensing for the LMS if provided. Delegate management and administration. Support. If a programme has 100 learners or 3,000, the COGS varies in roughly direct proportion, which is why some pricing models look very different at scale.

Factor 4: User-based vs organisation-based licensing

Three ways a digital training product can be priced and hosted.

User-based. The buyer pays per learner per period. Costs scale with volume. The provider takes responsibility for delegate management. Best for forces that want minimal internal admin and predictable per-learner costs.

Organisation-based. The buyer licenses the product into its own LMS for a fixed period (often 12 months) and a maximum learner count. The provider hands over the keys; the force takes on delegate management. This works well when the force has strong internal L&D operations and wants control.

The right model depends on what the force is set up to handle internally, not just on which looks cheapest in a spreadsheet.

IP-based. An organisation buys the ownership of the training materials and intellectual property itself - this involves additional design cost but if often the most effective for bespoke courses that are built from the ground up on the operational context and environment of the organisation. The work gets done and then belongs to the organisation to deliver in-house, update in the future, and train their own trainers to deliver.

This is actually my favourite way to work - the training can be more successful, the ownership of the material is more meaningful to the client, and the working relationship is deeper - training internal teams to then take ownership and take a programme forward once they've had support on bringing the thing into existence is essential.

Factor 5: The years behind the work

A useful analogy, possibly apocryphal. Picasso is in a Paris café. A would-be art collector spots him doodling on a napkin. He puts the napkin in his pocket and prepares to leave. She approaches and asks for the napkin. Picasso names a sum of several thousand francs. She protests: 'but you only spent five minutes.' The reply: 'no, I didn't. That sketch took me sixty-seven years.'

The point is about what's actually being paid for. When a digital training programme is built quickly by a specialist, the speed isn't a discount signal. It reflects the years of accumulated expertise that allow the work to happen quickly. Police training that genuinely works is usually written by people who have spent fifteen years working out what makes it stick. Pricing that reflects only the keyboard hours misses what the buyer is actually getting.

This isn't a claim that all providers price fairly. It's a claim that 'quick to build' and 'cheap because quick' are different things, and a buyer who conflates them can often end up commissioning the wrong work.

What generative AI changes (and what it doesn't)

Generative AI is shifting some of the underlying maths in three specific ways.

It changes the economics of bespoke. What used to take six months and a team of designers can now take days. A recent example from our own work: I built a complete draft training programme based on a 200-page piece of statutory guidance in a matter of hours. The work was done by a bespoke eight-agent team, with agents specialised by section of the guidance, working in parallel on different parts of the programme. The output covered learning objectives, structure, content scoping, and sequencing for a programme that would normally take weeks of design time to produce a first cut of.

But also - and back to the 'years behind the work' point above - it only took a matter of hours, because we've dedicated thousands of hours of staying close to the frontier of generative ai and business applications in L&D since late November 2022 when models first hit the commercial market.

It makes quality assurance and human experience the new differentiator. Anyone can generate a first draft. Whether that draft accurately reflects the source material, and whether it's structured according to evidence-based training design rather than the default patterns LLMs produce on their own, is where the expertise now lives. The eight-agent draft was passed through two layers of QA before it reached human review. The first QA layer checked guidance fidelity: does the draft accurately reflect what the statutory guidance actually says, and is anything material omitted? The second QA layer checked training design quality: is the structure evidence-based, sequenced appropriately for adult learners, and free of common L&D bad practice? And then was I able to review the outputs for quality from numerous perspectives because I have worked actively in the subject matter area. The power of the tools is immense - and they will only get better from here - but as we are all appreciating the potential for training slop is now infinite; judgement, taste, discernment and expertise are now the differentiator. Not to mention a highly fluent and evolved understanding of security and data privacy.

It changes the handover. In my recent experiment, the finished drafts were exported directly to a presentation-grade tool through MCP. MCP, the Model Context Protocol, is a standard that lets AI tools connect to other software directly. Instead of producing prose that someone then has to manually port into a presentation tool, the AI handed the finished material to the tool directly, ready for human review and refinement. The hours saved are real, but the bigger gain is that the work doesn't get diluted in translation between tools.

What AI doesn't change is the Picasso point. AI accelerates the keyboard hours. It doesn't replace the judgement that decided what counts as an accurate read of the guidance, what counts as evidence-based design, and what bad L&D practice looks like. The accumulated expertise still has to live somewhere. In our case, it lives in the prompts, the QA layers, and the people who built them and take a clear-eyed view of the outputs.

What the buyer is paying for is increasingly the design of the workflow rather than the keyboard hours. The shift is real.

How to use this in a procurement

Four practical points for a brief.

Name where in the synchronous to asynchronous spectrum the programme sits, and what the live component has to do.

Decide whether you need bespoke or whether off-the-shelf would do the job. Bespoke is rarely cheaper, and is sometimes necessary. With generative AI in the workflow, bespoke is now faster and cheaper than it used to be — though not free, and not without quality control.

Ask the provider to show how their pricing breaks down across the three cost categories. If the answer is opaque, the brief will be too.

Ask how generative AI is being used in design and development, and what the validation approach is. A provider who can't answer this clearly, or who claims AI doesn't change anything, hasn't engaged with how the work is now done - and with how this can benefit the buyer. Deep contextual knowledge about security approaches is a non-negotiable here.