Floowed

The Decision Engine

What actually decides the case.

A decision is rarely a checkbox. Underneath one sits the evidence it was given, the figures worked out from that evidence, the gates it has to clear, the score it is graded on, and a different outcome for every combination. All of it yours to set.

Start freeBook a demo →
  1. 01Gather
  2. 02Interpret
  3. 03Chase
  4. 04Decide
  5. 05Act
  6. The middle three repeat until the case is decision ready. One case, below, told the way it runs.
01Gather
02Interpret
03Chase
04Decide
05Act

GatherFrom your systems, from external sources, and from documents in whatever state they arrive.

InterpretWorks out what it means, not just what it says. A photographed statement, a bureau response, a field in your CRM, an answer somebody typed: all read the same way, and checked against each other.

ChaseAsks whoever still holds a piece of it, then stays with it. Reads the reply, works out whether it actually answers the question, and goes back again if it does not.

DecideRuns every gate your rules require. Anything still open sends the case back to be chased. When they all clear it comes back with a recommendation, and you choose what it may settle on its own.

ActWhat the decision sets off happens here. A record written back, a message sent, an endpoint called: wherever the outcome needs to land, that is where the case ends.

One supplier onboarding case · keep scrolling

Your systems
  • ERPPO request, 3 lines
  • CRMSupplier contact
External sources
  • RegistryCompany record
  • ScreeningCompany and directors
Documents
  • PDFBusiness permit
  • PhotoBank statements, 6 months
  • ScanAudited financials
  • PDFCompany registration
  • MissingTax clearance certificate
Bank statements, 6 months · photographedRead
Account nameMatches the name on the permit✓
Period coveredMarch to August, six months✓
Closing balanceRecalculated from the lines, agrees✓
Registration numberMatches the invoice✓
Tax clearance certificateExpired in March!

Read the same way as the ERP field and the registry record, then checked against each other

To the supplier · email
Your tax clearance certificate expired in March. Send a current one and we will carry on from where we are.
Attached. Sorry, the renewal came through last week.
Reply read · certificate current · gate closed
To Finance · Slack
Budget confirmed for this supplier at the requested volume?
Confirmed.
Reply read · confirmed
To Legal · email
Contract attached for review before the PO goes out.
Waiting · asked again Thursday
✓Permit is currentClear
✓Registration matches the name on the invoiceClear
✓Financials cover six monthsClear
✓Tax clearance has not expiredClear
✓Company and director screenClear
✓Bank account belongs to the company on the permitClear
✓Finance confirmed the budgetClear
✓Legal has seen the contractClear
Recommended: approve. 8 of 8 gates clear.Above $250,000 your team decides
✓PO raised in your ERPWritten back
✓Supplier record created, evidence attachedWritten back
✓Supplier told, Finance and Legal toldSent
✓Procurement endpoint calledCalled

Case closed · every check and every reply kept with it

The decision was never the slow part. Getting to it was.

Why they’re still manual

Two problems, and most tools only solve one.

You need both solved or the decision stays on somebody’s desk. Most of the market has taken the first one.

Problem

The evidence problem

Plumbing. Genuinely hard, and a lot of products solve it.

  • It sits in more than one placeA system, a sheet, an inbox, a folder. Almost everything that decides something assumes it all arrives on one screen, already structured.
  • At least one of those places isn’t a systemA spreadsheet somebody maintains. A document that arrives as a photo. An answer that only exists in somebody’s head until you ask. Where most tools stop and most of the work starts.

Problem

The logic problem

Not plumbing. This is the half almost nothing solves.

  • The rule isn’t really a ruleIt has thresholds, exceptions, and a band where it depends. Nobody wrote it down because the person who knows it has always just known it, which is fine until they are on leave or they leave.
  • The answer isn’t yes or noIt is how much, on what terms, with which conditions, or which of four grades. A tool that only routes an approval cannot hold that, so a person holds it instead.

Floowed is both halvesThe gathering and the chasing solve the first. The rules solve the second. Either one on its own leaves the decision exactly where it started, on somebody’s desk.

Inside one decision

Every gate, weight and band is yours to set.

Those situations are simple to describe. What decides them is not. This is one decision in full, and the same five layers sit under all of them.

  • Data elements

    The facts, pulled from wherever they live. Invoice dates and amounts from accounts, the filed accounts as a document, open disputes from the CRM, the registry record.

    • Days overdue, per invoice
    • Turnover, last filed year
    • Open disputes, count and value

  • Synthetics

    Facts the business does not hold anywhere, computed from the ones it does. This is usually where it stops being possible in a spreadsheet.

    • Average days to pay, rolling 12 months
    • Trend in order value, quarter on quarter
    • Disputes as a share of orders

  • Hard gates

    Pass, fail, or unknown, on any element or synthetic, at thresholds you set. A failed gate can stop everything regardless of how good the score is.

    • Nothing more than 60 days overdue
    • No unresolved dispute above 5,000
    • Registry status active

  • Scorecard

    Weighted attributes producing a score and a grade. You set the weights, the bands, and what each band is allowed to do. Bring your own model if you have one.

    • Payment behaviour · 40%
    • Financial strength · 35%
    • Relationship length · 25%

  • Outcomes

    Not one verdict. Every combination of grade and gate gets its own, and each can carry conditions rather than a flat yes or no.

    • A or B · approve to the requested limit
    • C · approve at half, review in 90 days
    • D or any failed gate · refer to a person
Trade credit · a worked exampleDecided
  1. 01Data elements
    • Days overdue, per invoice12 · 30 · 0
    • Turnover, last filed year2.4M
    • Open disputes, count and value1 · 3,200
  2. 02Synthetics
    • Average days to pay, rolling 12 months34
    • Trend in order value, quarter on quarter+8%
    • Disputes as a share of orders2.1%
  3. 03Hard gates
    • Nothing more than 60 days overduePass
    • No unresolved dispute above 5,000Pass
    • Registry status activePass
  4. 04Scorecard81 · Grade B
    • Payment behaviour · 40%34
    • Financial strength · 35%29
    • Relationship length · 25%18
  5. 05Outcomes
    • A or B · approve to the requested limitThis one
    • C · approve at half, review in 90 days
    • D or any failed gate · refer to a person
Approve to the requested limit. Grade B, 3 of 3 gates.Recommended, your team decides

In the product

What a decided case looks like.

Every gate that was checked, what it was checked against, and where each number came from. Open any line and the evidence sits underneath it.

Trade credit · C-2417Decided · v7
Approve to the requested limit. Grade B, score 81.Recommended
  • Nothing more than 60 days overduePass

    Oldest open invoice 30 days · from accounts, 14 invoices read

  • No unresolved dispute above 5,000Pass

    One open dispute, 3,200 · from the CRM, ticket #4471

  • Registry status activePass

    Active, filed accounts to March · from the registry

Scorecard 81 · Payment 34 · Financial 29 · Relationship 18Decided on the version live at the time

Getting one live

Describe it. Check it. Test it. Then publish.

Usually an afternoon for the first one, and considerably less for the next, because the integrations are already there.

  1. 01

    Start from the library

    Most decisions a business makes have been set up here before. Pick the one you recognise and it arrives with the checks that decision usually needs, shaped by everyone who has run it, so you are changing what does not fit rather than starting at nothing.

  2. 02

    Or describe your own in the Decision Builder

    Tell it the decision you want made, the way you’d brief a new joiner: which systems hold what, which sheet has the rest, who has to confirm, and what should happen at the end. It drafts the decision from that, including the gates it thinks you meant.

  3. 03

    Read it back before it runs

    The draft comes back as a flow you can read, not code: every step laid out in order, every gate saying what it checks and where it looks, every weight a number you can move, every band yours. Add the exception your business has, delete the one it invented.

  4. 04

    Test it on decisions you’ve already made

    Run it against cases where you already know the answer, and compare. The disagreements are the useful part: usually the rule is too tight, occasionally the original call was wrong.

  5. 05

    Publish, and keep every version

    Only a person with permission can publish. Every version is kept with who changed what and when, so a decision made in March is still explainable in November against March’s rules, not today’s.

On the record

Every case keeps its own file.

Not a log you have to reconstruct. The case itself holds what happened, so months later you can open it and see exactly how it was decided, against the rules that were live at the time.

  • Every gate checked, which passed, which didn’t, and the grade it scored
  • The evidence behind each one, and the figures worked out along the way
  • Who was asked to confirm something, when, and what they said
  • The version that was live when the case ran
  • Every chase that went out, and how long the case waited

Where we stop

Not every decision belongs on Floowed.

Breadth with no edges reads as no product. So here is the edge.

What it isn’t for

Three shapes we turn away, and will keep turning away.

  • Decisions with no rules behind themTwo experienced people read the same file, reach opposite conclusions, and both are right. If it can’t be written down, it can’t be run.
  • Decisions that happen twice a yearSetting it up costs more than deciding it by hand. Come back when it’s weekly.
  • NegotiationsIf the answer is agreed with the other side rather than derived from the evidence, there is nothing here to automate.

What it doesn’t replace

Two things we have no intention of taking off you.

  • Your workflow automationZapier, n8n, and the automation already sitting inside your CRM or ERP move the work, and they carry on doing exactly that. Floowed sits above them and makes the call none of them was built to make.
  • Your system of recordIt stays yours. Floowed reads from it, decides, and writes the outcome back through an API that works in both directions. Your team carries on working where they already work.

Go deeper

In credit and risk

The same engine, on a credit policy

Where the engine has run longest: bureau data, bank statements, spreading and scorecards, with the customers and the screens behind it.

The hard inputs

The documents nobody else can read

Scanned, photographed and handwritten. The one place we’d rather show you than tell you.

Start with one already built.

Write it in the Decision Builder and run it against cases you’ve already decided, or tell us the decision and we’ll set up the first one with you.

Start freeBook a demo