An AI workflow and product companySingapore

Models create possibility. Humancraft turns it into a way of working.

Your workflow first. AI where it improves the work.

Faster to market. Stronger client engagement. One unified platform for your teams and clients, powered by AI agents we design, deploy and connect to your tools.

AI built to work alongside your people, to improve your efficiency and revenue, within your compliance and safety rules.

Scroll to play

01The shift

Nearly four in five bought the capability and changed nothing about how the work is done.

Only 21% of companies using generative AI have fundamentally redesigned any workflow around it.

The other 79% have bought capability and changed nothing about how the work is done.

That is the whole problem, and it is not a technology problem.

Source: McKinsey, The state of AI

Models are becoming commodities. The advantage moves to whoever can redesign the workflow, earn adoption, integrate the systems, govern the actions and prove the outcome.

02What we do

We put AI into one workflow at a time, inside the systems you already run, with a human accountable at every step that matters, and a number on the wall that says whether it worked.

One workflow.

We start with the step everyone quietly works around.

Your systems.

Your core system, your CRM, your document store and your customer channels, under your access controls.

A number.

The baseline, its calculation method, its data source and its owner, agreed before we build.

Three specifics you can verify, and hold us to.

03How we work

We sit with the people doing the work, and rebuild the step everyone works around.

Step 01

Start with the people doing the work.

Rework, waiting and re-keying are three things any COO can point to on a whiteboard within thirty seconds. We find where the work actually stalls.

Step 02

Rebuild it inside the tool they already have open.

Not a new tab, not a new login. The screen your staff touch sits where they already work, wired to the systems that hold the truth.

Step 03

Place a named human at every decision that matters.

The specific points where a person signs, what happens when they do not, and where an exception stops and finds a human instead of guessing.

Step 04

Agree the one number before we build.

The baseline, its calculation method, its data source and its owner. If we cannot agree a baseline before we build, we do not build.

Step 05

Report it monthly, whether it moved or not.

Written into the engagement as a commitment. Alongside it, the guardrails we hold steady, at least one named by compliance.

Illustration

That last step is the difference. It is the risk you have otherwise been carrying alone.

04What we actually build

Six things.

I
  1. Request received
  2. Re-keyed into a second system
  3. Draft prepared
  4. Approval chased by email
  5. Approved in the flow

The redesigned process

The new sequence of steps, written down, with the removed steps struck through.

II
In the flow

The screen your staff touch

Inside the tool they already have open. Not a new tab, not a new login.

III
CORE SYSTEM CRM DOCUMENT STORE

The connections

To your core system, your CRM and your document store, under your access controls.

IV
DecisionA named human signs
If not signedIt stops and escalates

The approval points

The decisions where a named human signs, and what happens when they do not.

V
  • Adoption
  • Task success
  • Cost
  • Exceptions
  • Overrides

The dashboard

Adoption, task success, cost, exceptions and overrides. One screen, your data, your definitions.

VI

Your customers’ experience

Personalised engagement, service that answers at once, recommendations that fit. Designed with real customers, measured after launch.

That is the whole of it. Everything between a model that can do the thing, and the thing getting done on Tuesday morning.

05The path

From a named problem to a running workflow, to the revenue it returns.

1

Stage 1 · Diagnostic

Two weeks · one workflow

Diagnostic

You end with the questions your pilot is missing answered, a baseline everyone has signed, and a ranked shortlist of what to build first. If the honest answer is that AI does not belong in this workflow, that is what the report says, and you have saved a build.

What you hold at the end

  • The workflow mapped, with where it actually stalls
  • Opportunities ranked by value and feasibility
  • A baseline KPI with its calculation method
  • A named owner, data and compliance requirements listed
2

Stage 2 · Blueprint

Blueprint

The redesigned workflow, how the human and the AI interact, and the case for building it.

What you hold at the end

  • The redesigned workflow and a working prototype
  • Human approval and escalation design
  • Integration architecture
  • The business case and the measurement plan
3

Stage 3 · Build

Build

Production implementation on Ergon and Skopos, integrated with your systems and your approved models, tested with your people.

What you hold at the end

  • A running workflow in production
  • Integrations live, under your access controls
  • The dashboard with your agreed KPIs
  • The audit trail, documentation and a trained team
4

Stage 4 · Operate

Ongoing

Operate

The number, published monthly, including when it is bad. Prompts and workflows tuned, exceptions reviewed, next use cases identified.

What you hold, every month

  • Performance against the agreed baseline
  • Exception and override review
  • Tuned prompts and workflows
  • The next use cases, ranked
5

Stage 5 · The outcome

Compounding

Revenue & Customer Experience

Revenue is earned at the top line, not only saved in operations: campaigns that know your stock, personalised engagement that converts, next-best-action in front of the RM, shopping and sales assistants that close, service agents that resolve at first contact. Your customers feel it: served faster, answered accurately, followed up on time. And it compounds: the next use case takes weeks, not quarters, because the connections, controls and reporting already exist.

The numbers this moves

  • Revenue influenced and conversion
  • Customer satisfaction and retention
  • Cost to serve and cost per transaction
  • Time returned to your team

The engagement promise

The team that maps your workflow in week one is the team that defends it in your risk review, and the team on the call when adoption dips in month four.

No pitch team that vanishes after signing, no handover, no re-explaining your business to a delivery unit you did not meet.

And when it does not work

If the number does not move, we say so in the monthly report, we tell you why, and we agree whether to fix it or stop.

06The products

Ergon makes the work move. Skopos makes it visible.

The two products behind every engagement. They are what stops your project starting from zero. They are built so that when the model changes, you keep the workflow, the controls, the audit trail and the dashboard.

Ergon · the workflow engine

Ergon

Turns business processes into coordinated, measurable human and AI workflows.

  • Nobody chases the next step by email. The work routes itself and shows who is holding it.
  • Approvals happen inside the flow, with a record, instead of in a mailbox.
  • When something falls outside the rules, it stops and finds a human instead of guessing.
  • The audit trail exists before anyone asks for it, not after.
  • The same workflow pattern gets reused on the next problem instead of rebuilt.
ErgonAI stepNamed humanIllustration
SkoposAll workflowsIllustrative data

Task success against the agreed baseline, by week

Agent actions

    Versions trackedModelPromptPolicyLive

    Note. Illustrative data: the values on this screen are examples, not client results. Source for what Skopos measures, and its status: humancraft-ai.com/skopos.html

    Skopos · the visibility and control engine

    Skopos

    Makes AI performance, risk and business value visible.

    • Someone currently builds the AI update deck by hand, from four places, with no baseline. Skopos replaces that week with a screen that is already true.
    • It answers the two questions leadership actually asks: is anyone using it, and is it working?
    • Adoption, task success, cost, quality and latency, against the agreed baseline.
    • Agent action logs, human overrides, escalations and policy exceptions.
    • Model and prompt version tracking, management and audit reporting.

    Agentic assistants · chat and voice

    Anyone can put a chat window on your site this week.

    Here is what has to be true before we put one in front of your customers.

    The chat is what your customers see. That list is what makes it safe to put in front of them.

    1. 01A defined scope of permitted actions
    2. 02Sourced answers, with attribution
    3. 03A rule for when it is not sure
    4. 04A human it can escalate to
    5. 05A log of every conversation
    6. 06A KPI dashboard
    7. 07Policy and compliance monitoring
    8. 08An audit trail
    9. 09An evaluation set re-run on every change
    Customer service agentsEmployee knowledge assistantsRelationship manager copilotsProduct and sales assistantsRetail shopping assistantsOperations and compliance assistantsVoice-enabled service agents

    Product status. Ergon and Skopos are in the building stage and entering production in late November 2026. They have no client deployment history yet, and we will not claim one until a real engagement produces it and the client approves it in writing.

    07The black box

    The question is never how the model works.

    Imagine this: On 14 March something went to a customer.

    • Who authorised it?
    • What did it read?
    • Did anyone check it?
    • Had it gone wrong before?

    Today

    3 weeks

    That question starts an investigation across four teams.

    Four teams, three weeks

    With the record

    4 minutes

    We make it an answer on one screen.

    One screen, four minutes

    Note. Illustrative example: the times are examples, not client results. Source for the example and the eight-field record: humancraft-ai.com

    Every AI action leaves the same eight-field Transparency Record, on every system we build for you.

    Your risk team learns it once, and it is in place before go-live, not assembled after an incident.

    Transparency recordOne per AI action
    1. 01Permitted scopeWhat the AI was allowed to do
    2. 02Information usedWhich sources and data it read
    3. 03InitiatorWhich system or person started the action
    4. 04OutputWhat it produced or did
    5. 05Human reviewWhether a person checked or overrode it
    6. 06Task outcomeWhether it completed successfully
    7. 07Model and configurationExactly what ran, and which version
    8. 08Policy exceptionWhether it tripped a compliance rule

    08Where we work

    One flagship per sector, not a capability menu.

    Flagship · the relationship manager’s client pack

    The pack an RM builds the night before.

    Before every client review, an RM assembles a pack: holdings, recent activity, product fit, what was promised last time, and what compliance will allow them to say. It happens the night before, and its quality depends on which RM you get.

    Praxis rebuilds it as a draft that arrives before the meeting, sourced, with the compliance-sensitive lines flagged. The RM edits instead of assembles. Every version is logged, so the pack that was actually used is recoverable a year later.

    Also delivered here

    • Client briefing and next-best-action tools
    • Customer service agents
    • Personalised digital engagement
    • Onboarding and document workflows
    • Compliance-aware content and communications
    • Agent monitoring dashboards
    • Praxis · for private banks

    Flagship · connecting marketing to inventory

    The campaign that works sells out in four days.

    Most retailers run marketing and inventory as separate systems with a meeting between them. So the campaign that works sells out in four days, and the one that fails leaves stock nobody planned for.

    We connect the two. The campaign knows what you are holding. The buying plan knows what demand is doing this week. The number we move is picked with you in the diagnostic: stockouts on promoted lines, markdown on slow stock, or margin per campaign.

    Also delivered here

    • Segmentation and campaign planning
    • AI-assisted content production and approvals
    • Shopping and service agents
    • Low-stock and overstock alerts
    • Slow-moving stock identification
    • Margin and carrying-cost dashboards

    Operational process automation

    The work that carries the P&L.

    Document and data handling, internal approvals, exception routing, quality assurance, knowledge retrieval, reporting and reconciliation. Rebuilt as human-in-the-loop workflows with the same controls and the same dashboard as everything else we ship.

    Also delivered here

    • Process discovery and redesign
    • Document and data handling
    • Internal approvals and exception routing
    • Knowledge retrieval
    • Reporting and reconciliation
    • AI-assisted testing
    • ClassEd · enterprise learning

    Every offer connects a defined user, a workflow, a business outcome, a control model and a measurement plan. We do not enter a sector with a generic AI feature.

    09What we believe

    AI succeeds when it is built alongside the workflow, alongside the human, alongside the next quarter’s plan.

    Humancraft AI builds AI to work alongside humans, for humans.

    We are headquartered in Singapore, building for banks, ambitious enterprises and growing businesses across Asia and the Middle East. Outcomes that compound. Built with the humans who use them.

    10A two-minute audit

    Score your last pilot.

    A common mistake is the siloed pilot. BCG found three in four banks stuck in siloed pilots and proofs of concept. Score yours: one point for each line that was true. If you have not run a pilot yet, use it as the checklist for your first.

    Source: BCG, For Banks, the AI Reckoning Has Arrived

    0/ 9

    Your score

    Tap each line that was true of your pilot. The score reads itself.

    11Where it starts

    Two weeks.One workflow.

    Models create possibility. Humancraft turns that possibility into a way of working.

    Start with a call. Bring one workflow and the person who owns whether it improves. If we are a fit, the two-week diagnostic leaves you holding a start baseline, a ranked shortlist of what to build first, and an honest answer about whether AI belongs there.

    Start the conversation →

    Outcomes you can seeControls you can trust