Stage 4 · Deploy · into real workflows

Generative AI

Most people have judged generative AI by what a free chatbot did with a one-line prompt. Given your precedent library, your rate card and your rules, the same technology drafts the proposal, the board paper and the plain-English rewrite in your voice, and it does not invent, because it may only write from what it was given.

Talk to an AI architect See the packages

Grounded in your documentsYour voiceProposals and tendersReportingReviewed before it leaves

Copilot licensed byBe Cloud permissions checked withBe Secure

What it does

Drafting that starts from your knowledge, not the internet’s.

Generative AI writes, summarises, analyses and produces: reports, proposals, analysis, first drafts of anything. Left to itself it writes from everything it has ever read, which is why the free version sounds like everyone and is sure of things that are not true. The fix is to change what it starts from.

A grounded system is given your documents first (precedents, past proposals, policies, the numbers from your systems) and told to write from those and nothing else. It is given your tone and told what it must not do: quote a price that is not on the rate card, promise a date, cite a document it was not shown. The draft that comes back is recognisably yours, every claim traces to a source, and a person reads it before it leaves. The judgement about what to send stays where it always was.

Where it fits

Three kinds of writing that stop taking a week

Proposals

Tenders assembled from what you have already written

A consulting engineer responding to a council tender: the response is assembled from past submissions, CVs and method statements, priced from the rate card, and the people who will win it spend their time on what is new instead of what is boilerplate.

Reporting

The board pack drafted from the systems

A company of a hundred and twenty people: the monthly report is drafted from the accounting system, the CRM and the project tracker, in the board’s format, with the variances flagged. The CFO reviews it and adds the judgement.

Processing

Summaries, comparisons and plain-English rewrites

Two versions of a contract compared clause by clause. A forty-page policy rewritten at a reading level the people who follow it can manage. Safety instructions translated for a workforce that speaks several languages, checked by a person who speaks them.

How we build it

Grounded generation, built to the five rules of the architecture

Knowledge before models. The precedent library, the rate card and the style guide go in before a model is chosen. A mid-range model with the right documents beats the best model guessing.
Consequence needs a person. Nothing it writes leaves unreviewed. The design names who reviews what: a partner for a proposal, a clinician for a patient letter, the CFO for the board pack.
Integrate, don’t replace. Drafts appear where your people already work: Word, Outlook, the document system you use. Copilot is often the front door; the grounding makes it useful.
Every action leaves a record. Each draft records which sources it used and which version a person approved, so you can show where a sentence came from a year later.
No vendor is load-bearing. Your documents, rules and prompts are structured so the model behind them can change. When a better or cheaper model appears, you switch.

One proposal, request to sent

What a grounded system does with a tender that closes on Friday

1
The tender is readRequirements, weighting, format and deadline extracted into a checklist.
2
Precedents are retrievedThe three past submissions closest to this one, the relevant CVs, the current rate card.
3
A first draft is writtenAgainst the checklist, in your voice, with every reused passage marked with its source.
4
The experts editThe parts that win the work are written by people. The parts that need to be present are already done.
5
A partner approves and it goesThe final version, the sources used and the approval are recorded with the submission.

Compare

Copilot or ChatGPT as they come, a prompt library, or a Be AI build

	Copilot or ChatGPT as they come	A prompt library for your team	Be AI generative build
Writes from your documents only	Partly, if permissions allow	No	Yes
Sounds like you	After heavy editing	Sometimes	Yes
Review step before it leaves	Up to each user	Up to each user	Designed in
Sources recorded for every draft	No	No	Yes
Sensible when	You want a personal assistant for each person	The writing is low-stakes and varied	The document is recurring, valuable and yours

Questions

What people ask about Generative AI

How do we stop it making things up?

By limiting what it writes from: your documents and nothing else, every passage marked with its source, and a plain “no source” when there is none. With a person reading before anything leaves, the remaining risk is the one you carry with any junior drafter.

Who owns the output?

The business terms we specify assign the output to you, and the record of sources and human edits lets you show it. Copyright in machine-generated text is unsettled in several countries, so anything you rely on for legal protection should go past your lawyer.

Will our data train someone else’s model?

Not on the business and enterprise terms we specify, which commit the provider not to train on your inputs. Consumer accounts are different, which is why governance work takes staff off them. Where data is too sensitive for any provider, the model runs in your own tenancy.

Will our clients be able to tell?

A grounded draft edited by your people reads like your people, because it was written from their work. Where a client or a regulator expects to be told AI was used, the policy says so and the record of sources and review backs it.

Write from what you know, and only that.

Bring us the document your team dreads producing every month and we will show you what a grounded first draft looks like, and what has to be in place first.

Start your AI journey →Get my AI Opportunity Score Or call 0800 23 3444