How we work

One partner, from idea to everyday use.

Six stages, a decision at every gate. Buy one phase or the whole journey.

01 The lifecycle

Six stages. A decision at every gate.

Each stage ends with something you receive and the evidence to move on.

Find

Stages 01–02

  1. Stage 1 of 6

    Discover & prioritise

    You receive

    • Workflow interviews with the people who do the work
    • Opportunity shortlist ranked by value, feasibility and risk
    • Data and document audit

    Gate 1 · to move on

    A named user, a funded problem, a measurable outcome.

  2. Stage 2 of 6

    Prepare & design

    You receive

    • Ingestion and cleaning plan for the documents and data in scope
    • Metadata, code lists and access rules
    • Clickable prototype and solution architecture

    Gate 2 · to move on

    Usable data, agreed permissions, a testable design.

    Human review

Build

Stages 03–04

  1. Stage 3 of 6

    Prototype & evaluate

    You receive

    • A working prototype on representative documents and data
    • An evaluation set: agreed test questions with expected answers
    • Failure analysis against agreed quality, latency and cost thresholds

    Gate 3 · to move on

    Quality, usefulness, latency and cost meet the agreed thresholds.

    Human review

  2. Stage 4 of 6

    Build & integrate

    You receive

    • Production application with access controls and audit trail
    • Data pipelines with quality checks and scheduled refreshes
    • Retrieval workflows with fallbacks; APIs and integrations

    Gate 4 · to move on

    User acceptance, security checks, operational readiness.

    Human review

Run

Stages 05–06

  1. Stage 5 of 6

    Deploy & enable

    You receive

    • Deployment to the agreed environment, cloud or private
    • Monitoring of usage, quality and cost
    • Documentation, training and handover to a named owner

    Gate 5 · to move on

    A responsible owner, a support arrangement, an adoption plan.

  2. Stage 6 of 6

    Operate & improve

    You receive

    • Data refreshes and quality monitoring
    • Cost optimisation and model updates as providers change
    • Workflow refinements from what people actually do

    Standing review

    A regular review of usage, reliability and outcomes.

    Human review

02 Ways to work with us

Buy a focused phase or the whole journey.

Defined scope, clear output. Priced after we review your information.

  • Engagement 01

    AI Opportunity Sprint

    A fixed-scope diagnostic that ends with a prioritised roadmap and a buildable first concept.

    Covers Discover · Prepare

    Best for
    Teams that want evidence before committing budget.
    You get
    • Ranked opportunity shortlist, workflow map and data audit
    • Prototype brief and a recommended next step
  • Engagement 02

    Pilot Build

    One workflow, one user group, agreed acceptance criteria.

    Covers Prepare · Prototype

    Best for
    Teams with a clear workflow who want proof with real users first.
    You get
    • Working prototype on your data, with evaluation set and failure analysis
    • Go / no-go recommendation for production
  • Engagement 03

    Production Build

    Milestone-based delivery of the complete product, integrated, deployed and handed over.

    Covers Build · Deploy

    Best for
    Teams with a validated pilot or a well-defined internal tool.
    You get
    • Production application, pipelines, integrations and deployment
    • Documentation, training and handover to a named owner
  • Engagement 04

    Managed Improvement

    A recurring support and optimisation agreement after launch.

    Covers Improve

    Best for
    Any product in daily use. Models, data and needs change.
    You get
    • Monitoring, quality reports, data and model updates
    • Workflow refinements and regular reviews

03 Responsible delivery

What we agree before we build.

Every proposal writes down what “working” means.

Agreed in writing · every proposal

  1. Source attribution

    Every answer from documents or data shows its sources.

  2. Human approval points

    A person reviews before anything is published, sent or acted on.

  3. Access controls

    Who sees which documents, datasets and answers, enforced in the product.

  4. Test questions and acceptance criteria

    Agreed test questions with expected answers; thresholds for quality, latency and cost.

  5. Fallback behaviour

    “I could not find that” is a designed outcome.

  6. Measurement beyond speed

    Task completion and correction effort, measured alongside time saved.

  7. Operational ownership

    A named owner on your side, a support arrangement and an adoption plan.

  8. Hosting and model-provider use

    Where it runs, which models it calls, what data they see. Private hosting available.

  9. Costs, source-code rights and licences

    Recurring costs, rights and licences, with client-specific development itemised.

04 Reuse or build

Reuse what fits. Build what your workflow needs.

Factly products where they fit, custom where the workflow requires it. The proposal says which.

Reuse

Factly products and components, where they fit

Faster, and already tested in production.

Build

Custom software, where the workflow requires it

Focused software around the way your team already works, from enquiries to tasks and records.

  • Client-specific development itemised separately; source-code rights defined in the proposal.
  • We tell you when an off-the-shelf tool is the better answer.

An example we built for ourselves

05 Buyer questions

Questions buyers ask us.

Anything specific to your systems? Ask us.

Can you work with our existing systems?

Usually, yes: through APIs, database connections, document stores and export formats, confirmed during discovery.

What information do we need to provide?

To start: the problem, who would use the tool and how the work is done today. For a sprint or pilot: representative documents and data, the access rules and a few real questions. It does not need to be clean first.

Can we start with a small pilot?

Yes. The Pilot Build is one workflow, one user group and acceptance criteria agreed in advance. Not sure which workflow? Start with the AI Opportunity Sprint.

How do you evaluate AI outputs?

We agree test questions, expected answers and thresholds for quality, latency and cost before we build. Prototypes run against that set and you see the failure analysis. In operation we measure task completion and correction effort.

Can we deploy privately?

Yes, after a feasibility check for your environment. Hosting and model-provider use are defined in each proposal. GoPie can be self-hosted and call a model of your choice, including a local one.

Who owns the software?

Source-code rights are defined in the proposal. Client-specific development is itemised separately from platform subscriptions, licences and third-party costs.

Do you offer training and support?

Yes. Training and handover are part of Deploy & enable; Managed Improvement covers refreshes, monitoring, model updates and regular reviews after launch.

Will you reuse Factly products or build something custom?

Both, and the proposal says which. We reuse Factly products (GoPie for analytics, the retrieval patterns behind Tagore AI) where they fit, and build custom where your workflow requires it.

What if our problem does not need AI?

Then we will say so. Many operational problems are solved by a clear workflow tool and clean data. We add AI where it earns its place.

Do you work in health, agriculture, climate or other high-stakes domains?

Yes, within research, reporting and operational workflows. Where a product would inform high-stakes decisions, we scope domain expertise and validation with you.

Sources (3)

  1. [1] Dataful: home · dataful.in (opens external site) accessed 2026-11-15
  2. [2] factly/gopie (AGPL-3.0) · github.com (opens external site) accessed 2026-11-15
  3. [3] factly/kavach (MIT) · github.com (opens external site) accessed 2026-11-15

Start small or start whole

Tell us what you want to build.

We will suggest where to start.