Skip to content
Build n Bloom
Beyond advisory · What we build

Built into the work your team already does.

We build workflows, AI agents, integrations and internal tools around a specific piece of work in your business. Each one has a clear job, limited access, tests it has to pass and a person who decides what matters. Most projects combine two or three of them, and we always start with the simplest one that does the job.

Book a free Fit Call How we work

Four things we build, often together.

Each kind of build has a clear job, a boundary on what it may do and tests it has to pass before you rely on it.

Workflows

A defined path that moves work from one stage to the next. Rules handle the predictable steps. AI reads, compares, sorts or drafts where that helps.

Illustrative examples
  • An enquiry opened as a job with its details checked.
  • A supplier bill matched to its order.
  • A weekly report assembled from three systems.

Stays humanAnything consequential, unusual or customer-facing.

Acceptance provesRepresentative cases reach the right destination, and missing information stops the case visibly.

Agents

AI with one job, a fixed set of sources and tools, and limits on what it can do without a person. We build an agent only where the steps can't be pinned down in advance. If rules would do, we use rules.

Illustrative examples
  • Drafting a quote from the price list and past jobs.
  • Preparing a reply to a customer from your policies.
  • Summarising a call into CRM notes for checking.

Stays humanApproving what goes out, what it costs and what's promised.

Acceptance provesIt works from permitted sources only, shows what it relied on, passes the agreed cases and routes uncertainty to a person.

Integrations

Connections between the systems you already use, so information moves once and stays consistent.

Illustrative examples
  • Accepted quotes creating jobs and draft invoices.
  • Form submissions landing in the CRM with duplicates caught.
  • Status updates posted where the team already works.

Stays humanWhich systems connect, what may be written to them and who approves changes.

Acceptance provesOnly agreed fields are read or written, and a failed connection is handled safely.

Internal tools

Small purpose-built apps, views and dashboards for a job your current software doesn't quite fit.

Illustrative examples
  • A workshop schedule that shows jobs, materials and dates in one place.
  • A review queue for drafted replies.
  • A simple dashboard for the numbers the owner checks every Monday.

Stays humanHow the team uses it and what it should show.

Acceptance provesThe people who use it can do the agreed tasks on representative work.

Illustrative examples, not client work.

How the pieces fit in one piece of work.

A fictional joinery workshop wants to quote faster without the owner checking every number at night. Here's how one build might connect the parts.

KJ-EX-02Kelvane Joinery · enquiry to workshop · fictional
  1. Workflow + integration

    Enquiry in

    A website enquiry opens a record in the CRM. Missing measurements are flagged and a request for them is prepared.

    Decides: The office coordinator sends the request.

  2. Agent

    Draft quote

    A draft is prepared from the current price list and similar past jobs, with each line linked to its source.

    Decides: No one yet. It's a draft.

  3. Human decision

    Estimator decides

    The estimator corrects and approves. Quotes above an agreed amount go to the owner.

    Decides: The estimator, then the owner where agreed.

  4. Integration

    Job and invoice draft

    The accepted quote creates a job on the job board and a deposit invoice draft in the accounting system.

    Decides: The bookkeeper sends the invoice.

  5. Internal tool

    Workshop view

    The job appears on the workshop schedule with materials and dates.

    Decides: The workshop lead sets the order of work.

Fictional example. Kelvane Joinery is invented. It is not a client, a result or an installed system, and nothing here runs.

Keep your tools. Connect the work.

We start with the systems your team already trusts and connect only what the work needs.

  • Microsoft 365
  • Google Drive
  • Slack
  • Notion
  • HubSpot
  • Salesforce
  • Xero
  • QuickBooks
  • Airtable
  • GitHub

Connections depend on permissions, API availability and the scope agreed for your project.

Agree what “working” means before we build.

We write the tests before the build, from your real work, and run them before you accept and again during stabilisation. Acceptance proves the installed system behaves as agreed on those cases. It doesn't promise a business outcome or replace your team's judgement.

  1. Representative work passes

    Agreed test cases, drawn from real work, move through each step and reach the right place with their details intact.

  2. The awkward cases behave

    Missing information, an unusual request or a failed connection stops the case visibly. The system doesn't dress up a missing source, field or approval as a finished answer.

  3. The person who decides stays in control

    Anything consequential, hard to reverse or involving money waits for a named person, who can inspect, correct, reject or approve before it is saved or sent.

  4. You can see what happened

    Inputs, sources, what AI contributed, what people changed and the final state can be traced.

  5. The old way still works

    Until the new system has proven itself on real work, your current process stays available, and the handover records how to switch back.

Built with the people who'll use it.

We involve the people who do the work from the start, because they know where it really gets stuck.

The new system runs beside the current way of working first, with them reviewing what matters. Before handover we train them on representative cases and write down who owns the system, how problems are reported and how to go back to the old way if needed. Then you accept it in writing, and thirty calendar days of in-scope stabilisation follow.

The aim is a team confident enough to improve the system, and to build small workflows themselves, with help where they need it.

A build has to earn its place.

We'll recommend something smaller when it would do the job

  • A clearer process, with no new software.
  • Better use of a tool you already pay for.
  • A standard product configured for the work.
  • Plain rules instead of AI, when the steps are predictable enough.
  • Waiting until the work is steady enough to build around.

We'll suggest starting somewhere else when

  • The first piece of work would be high-risk if it went wrong. Start contained; earn the bigger one.
  • The information the system would need isn't yours to use, or isn't reliable enough yet.
  • The time your team would spend checking and correcting it would outweigh what it saves.

Questions about what we build.

What's the difference between a workflow, an agent and an automation?

An automation follows fixed rules. A workflow is the whole path a piece of work takes, which may include automations, AI steps and human decisions. An agent is an AI component with one job, limited tools and a person who reviews what matters. We use the simplest of the three that does the job.

Do you build custom software or set up tools we already pay for?

Both. We start with what you have and build only what's missing. Many projects are mostly configuration and connection, with a small custom piece where it matters.

Which tools can you connect?

Common examples include Microsoft 365, Google Drive, Slack, Notion, HubSpot, Salesforce, Xero, QuickBooks, Airtable and GitHub. Each connection depends on permissions, API availability and the agreed scope.

Do AI agents act on their own?

Only within limits you agree. Anything consequential, such as sending to a customer, committing money or changing an important record, goes to a named person first. Early on, a person reviews more; the limits widen only as the agreed tests keep passing on real work.

Can you use the AI subscriptions we already have?

Often, yes. Which model and provider process your data is agreed and disclosed for each installation.

What if it doesn't work?

If core work fails its agreed tests, we correct and retest it. If it still fails, the service guarantee sets out when you can stop and get fees back.

Read the service guarantee

Bring the work. We'll help you decide what to build.

Talk through one part of how your business runs with a founder. You don't need a technical brief. You keep a short reviewed provisional Fit Note with the next useful decision.

Book a free Fit Call

25 minutes · no fee · no confidential files