Skip to content
What it is Who it's for How it works Inventory engine Order book Supplier intelligence Scoring engine The learning period AI & predictive models Architecture Technical methods Supplier agents Dashboard Contact Request a demo

Retail inventory & procurement intelligence

AI-powered inventory.Automated procurement.

IntelliFind continuously learns retailer demand, monitors supplier markets, and autonomously optimises when, what, and where retailers should buy.

Explore how it works Contact IntelliFind
—
SKUs under continuous monitoring in a pilot catalogue
—
Supplier sources parsed into structured records
—
Reorder evaluations run per hour
Engine state --:--:--
—below benchmark
—resting orders
—market orders today

The problem

Procurement decisions are made faster than people can make them well.

A mid-sized retailer carries thousands of SKUs across dozens of suppliers. Every one of them has its own price, lead time, minimum order quantity and reliability record — and all of those move. Buyers reconcile spreadsheets, supplier emails and gut feel, then place the order that is easiest rather than the order that is best.

The cost shows up twice: capital frozen in stock that isn't selling, and margin lost to shelves that are empty when demand arrives.

01

Demand moves, benchmarks don't

Reorder points are set once and rarely revisited, so they lag seasonality, promotions and category shifts.

02

Supplier data lives on websites

Prices, stock indicators and delivery estimates are public but unstructured — nobody has time to check them daily.

03

Chasing eats the team

Missing deliveries, refunds and reimbursement claims consume the hours that should go into buying decisions.

What is IntelliFind

One engine between your sales data and your suppliers.

IntelliFind sits on top of the retailer's inventory and sales data, builds a continuously updated picture of the supplier market, and turns the two into procurement decisions. Six capabilities, working as a single loop.

Inventory intelligence

Continuously monitors retailer inventory levels and the rate at which each product is selling.

Demand prediction

Models estimate expected demand per product and flag when the underlying sales pattern has changed.

Supplier intelligence

Gathers supplier information directly from supplier websites and converts it into structured records.

Procurement optimisation

Scores every available purchasing opportunity and identifies the most suitable supplier for the order.

Automated ordering

When an inventory threshold is reached, the engine can submit and route the order without waiting for a buyer.

Autonomous supplier communication

AI agents handle operational issues — missing orders, reimbursements, refunds and policy questions.

IntelliFind operates at the retailer inventory and procurement level. Sales data is used in aggregate to understand product demand — the platform does not profile or order on behalf of individual shoppers.

Who it's for

Built for the people who own the stock.

Retailers

  • Reduce stockouts on products that are actually selling
  • Reduce capital tied up in excess inventory
  • Improve purchasing decisions with comparable supplier data
  • Automate repetitive procurement work
  • Monitor supplier pricing and reliability continuously
  • Improve inventory turnover
  • Respond dynamically as demand changes

Retail operations teams

For the teams who carry the day-to-day responsibility for:

  • Inventory
  • Procurement
  • Supplier management
  • Purchasing
  • Order management

IntelliFind is designed to sit alongside the team as a decision layer — it proposes, routes and records, at whatever level of autonomy the team configures.

Larger retail organisations

The architecture is designed to scale across large product catalogues, many suppliers and multiple retail locations, with per-location inventory state and shared supplier intelligence.

Scaling: design target

Multi-site behaviour is part of the platform's intended architecture rather than a capability proven at scale today.

How it works

Eight stages, running continuously.

The engine never stops at step eight. Every confirmed order and every day of sales returns to the front of the loop as new evidence. Open any stage for the underlying method.

The retailer provides historical sales and inventory data during the initial trial and training period. IntelliFind ingests transaction-level sales, stock movements, purchase history and — where available — past supplier orders and delivery dates.

This is also where the data contract is agreed: which fields are available, at what frequency they refresh, and how products map to the supplier catalogues IntelliFind will monitor.

IntelliFind analyses sales rates, inventory behaviour, product demand and other relevant signals to produce a per-product demand estimate with an associated uncertainty, rather than a single flat average.

Products with sparse history borrow strength from their category and product attributes, so a new SKU starts with a defensible prior instead of nothing.

The engine continuously queries retailer inventory and collects supplier information. On the retailer side that means current stock on hand, recent sales velocity and open orders. On the supplier side it means price, availability indicators, delivery estimates and order constraints, refreshed on a schedule per source.

Available supplier and product opportunities are scored on price, reliability, delivery time, quantity availability and other factors. The score describes how attractive a specific purchase is for this retailer, right now, at the quantity actually required.

The same product from the same supplier can score very differently at 20 units and at 500, because bulk terms, minimum order quantities and availability all enter the calculation.

The inventory engine determines whether each product is above benchmark, in the benchmark region, or at critical inventory. This part of the system is deterministic: the models supply the demand estimate, and clearly stated rules turn that estimate into a threshold and an action.

The engine submits and routes the appropriate procurement order — a resting order held against a score threshold in the benchmark region, or a market order when inventory is critical. Routing, approval requirements and spend limits are configured by the retailer.

Confirmed orders and updated sales and inventory data feed back into the system. What was promised is compared with what arrived: actual delivery time against estimate, quantity delivered against ordered, price paid against price advertised.

That comparison is what makes supplier reliability a measured quantity rather than an opinion.

Demand estimates, benchmarks, scoring weights and seasonal models update as new information arrives. A sustained change in sales rate moves the benchmark; a supplier that misses three delivery windows moves its own reliability term. The loop then restarts at monitoring.

Retail inventory management engine

Two thresholds decide everything.

The engine continuously queries retailer inventory and compares current stock against calculated thresholds. Where the level sits determines not just whether to order, but what kind of order to place.

Critical inventory levelthreshold 1

The inventory level required to cover expected demand during the shortest available replenishment period, assuming the current sales rate remains approximately constant.

Falling below it means the product is likely to run out before any supplier can resupply it. When that happens, a market order is triggered: IntelliFind identifies the supplier with the lowest applicable IntelliFind cost score for the required quantity and routes the order immediately.

Inventory benchmarkthreshold 2

A higher, healthier level. A simplified representation:

Benchmark demand ≈ current sales rate × (1 + time to longest relevant delivery)

In the benchmark region IntelliFind can place a limited (resting) order attached to a supplier/product score threshold. If conditions improve, the order rests. If inventory keeps falling toward critical, it escalates to a market order.

This formula is a conceptual representation intended to make the logic legible. The deployed model may incorporate additional variables — demand variance, supplier reliability, shelf-life, order cadence and cost of capital among them.

Inventory simulation — single SKUhealthy
—benchmark level
—critical level
—days of cover remaining
—engine action
Healthy inventory Benchmark zone — resting order Critical zone — market order Projected depletion at current sales rate

Retailer order book

Every procurement decision, on one book.

IntelliFind maintains a retailer-specific order book recording resting orders, market orders and confirmations — with the supplier, product, quantity, price, expected delivery, status and the scoring information behind the decision.

  • Order bookResting & market orders
  • Confirmed orderSupplier acknowledges
  • Inventory updateIncoming stock recorded
  • Engine queryThresholds re-evaluated
  • New decisionRest, escalate or hold
Order book — illustrative extractdemo data
RefProductSupplierTypeQtyUnitScoreExpectedStatus
OB-4471USB-C 65W chargerNorthwell ComponentsResting240£8.4081.4—Awaiting score ≥ 84
OB-4470iPhone 14 display assy.Vantor Parts LtdMarket60£31.1088.224 SepConfirmed
OB-4468Braided lightning cableHarlow DirectMarket500£1.9290.623 SepIn transit
OB-4462Tempered glass 6.1"Pacegate SupplyResting1,000£0.6176.9—Escalation armed
OB-4455Bluetooth earbuds (gen 3)Northwell ComponentsMarket120£14.7583.119 SepDelivered — reconciled
OB-4451Fast-charge power bankRidgeway TradeMarket80£11.2072.417 SepShort delivery — agent opened

Once an order is confirmed it updates the retailer's inventory state, which is then queried by the inventory engine on its next pass — so a confirmation immediately changes what the engine does next.

Supplier intelligence

No supplier integration required.

IntelliFind does not need access to a supplier's inventory system. It discovers and continuously collects supplier information by intelligently monitoring supplier websites, then transforms what it finds into structured data the procurement engine can act on.

https://northwell-components.example/catalogue/usb-c-65w

Northwell Components

Product: USB-C 65W GaN charger, UK 3-pin, 2-port

Trade price £8.40 per unit ex VAT. £7.65 on orders of 500+

Stock status: In stock — ships same working day

Delivery: 2–3 working days (UK mainland)

Minimum order: 50 units

Returns accepted within 30 days, unopened, buyer pays carriage.

Shortages or damage must be reported within 72 hours for reimbursement.

Trade desk: trade@northwell-components.example

Structured supplier record

      

    


    

      
What gets collected

        
    
          
  • Product information
  • Price
  • Availability indicators
  • Delivery estimates
    • 
          
  • Minimum order quantities
  • Bulk purchasing requirements
  • Supplier policies
    • 
          
  • Returns policies
  • Refund policies
  • Reimbursement policies
  • Contact information
    • 
        

      

      
How it becomes usable

        
Raw pages are parsed into typed fields, normalised across suppliers (units, currencies, lead-time formats), matched to the retailer's own SKUs, and versioned so a price or policy change is recorded as an event rather than silently overwritten.

      

      
What it is not

        
It is not a live feed from a supplier's ERP. Availability indicators are read as the supplier publishes them, and IntelliFind treats them as evidence with a confidence and an age — not as guaranteed stock.

      

    

  







  

    

      
Product & supplier scoring engine

      
A single number for "how good is this purchase?"

      
The IntelliFind score expresses the attractiveness of one purchasing opportunity — this product, from this supplier, at this quantity, today. Move the weights and watch the ranking change: the engine does the same thing when the retailer's priorities or the market do.

    


    

      

        
Scoring weights

        

        

        

        

        

        

        
Weights are normalised to 100%. In deployment these are calibrated during the learning period rather than set by hand.

      

      

        
Opportunity: USB-C 65W charger — 240 unitsranked live

        

        
IntelliFind is an optimisation system designed to learn which purchasing decisions produce better outcomes. It does not guarantee that the selected supplier is optimal.

      

    


    

      
Price
How competitive is the unit price at the quantity actually required, landed cost included?

      
Supplier reliability
How consistently does this supplier fulfil orders in full, on time, at the price quoted?

      
Delivery time
How quickly can the product arrive — and how tightly does the supplier hold to its own estimate?

      
Quantity availability
Can the supplier actually provide the required quantity now, or only part of it?

      
Bulk purchasing
How do the terms behave as order size grows, and where do the break points sit?

      
Historical performance
How has this supplier–product combination performed for this retailer previously?

    


    

      
Training against business outcomesimplementation option

      
Over time the scoring model can be trained against outcomes the retailer actually cares about rather than proxies alone:

      
    
        
  • Revenue
  • Profitability / P&L
  • Inventory efficiency
  • Product availability
    • 
        
  • Customer satisfaction
  • Speed of replenishment back to sale
    • 
      

      
The objective is to learn which purchasing decisions produced better results for this retailer, and to shift the weights accordingly — not to claim a globally optimal answer.

    

  







  

    

      

        
Estimated during the trialcalibration

        
    
          
  • Product sales rates
  • Demand distributions
  • Inventory requirements
    • 
          
  • Supplier performance
  • Delivery behaviour
  • Product demand patterns
    • 
          
  • Initial scoring weights
  • Benchmark parameters
    • 
        

      

      

        
The learning period

        
The engine earns its thresholds before it uses them.

        
During the initial trial, IntelliFind learns from the retailer's historical and incoming data. Nothing is ordered on assumptions carried in from another business.

        
The scoring model's weights and biases are evaluated and calibrated before deployment, and benchmark parameters are checked against what actually happened in the historical window: would this threshold have prevented that stockout, and would it have left capital sitting in stock that didn't move?

        
The purpose is straightforward — to make sure the deployed model is calibrated to this retailer's operating environment, with its lead times, its seasonality and its suppliers, rather than to a generic one.

      

    

  







  

    

      
AI & predictive models

      
Models advise. The engine decides.

      
IntelliFind runs AI agents and predictive models alongside a deterministic inventory engine. The models supply demand estimates, seasonality and anomaly signals; the threshold and routing logic that spends money stays explicit, inspectable and rule-based.

    


    

      
Demand
Update demand models
As new sales data arrives, models reassess previous assumptions about how fast a product is selling and how much that estimate should be trusted.

      
Evidence
Test hypotheses
Competing explanations for a change in sales are evaluated, and parameters move only when the observations provide enough evidence to justify it.

      
Cold start
Predict new product demand
For products with little history, predictive models estimate an initial expected sales rate from product attributes and category behaviour.

      
Seasonality
Detect seasonality
Once enough history exists, recurring patterns are identified and folded into forecasts, so benchmarks rise before a season rather than after it.

      
Weighting
Reweight models
New evidence shifts the weight given to each model, so forecasts adapt as the retailer's sales environment changes.

      
Boundary
Where the LLM stops
Language models read supplier pages and draft supplier communications. They do not set thresholds, choose suppliers or release spend on their own.

    

  






  

    

      
Technical architecture

      
Select a layer to open it.

      
Data enters from the retailer, becomes a demand estimate, meets supplier intelligence at the scoring stage, and leaves as an order — which returns as inventory and closes the loop.

    

    

      

      
    

  







  

    

      
Technical methods

      
What's running underneath.

      
Where an implementation is fixed it's stated plainly. Where it isn't yet, it's marked as an implementation option rather than dressed up as something already deployed.

    

    

      

        
Per-SKU sales rate is estimated as a distribution, not a point. Low-volume and intermittent products are poorly served by simple moving averages, so candidate approaches include exponential smoothing families for stable sellers, intermittent-demand methods such as Croston-type estimators for slow movers, and gradient-boosted regressors over engineered features (recent velocity, price, promotion flags, category, weekday) where enough history exists. implementation option

      

        
Historical observations are decomposed into level, trend and seasonal components so that a rise in sales can be attributed rather than merely observed. This matters directly: a trend change should move the benchmark permanently, a seasonal peak should move it temporarily, and noise should move it not at all.

      

        
Each product carries a prior over its sales rate — from its own history, or from its category when it is new. Each new day of sales updates that prior into a posterior, so the estimate and its uncertainty move together. Wide uncertainty produces more conservative thresholds; as evidence accumulates, the engine is able to hold less safety stock for the same service level.

      

        
Before changing a model assumption, the system asks whether the observed change is distinguishable from normal variation. Sequential tests and change-point detection are used to compare "the rate is unchanged" against "the rate has shifted", with thresholds set so that routine noise does not trigger a permanent benchmark change. implementation option

      

        
Choosing a purchase is a constrained problem, not a sort by price. Quantity required, minimum order quantities, bulk break points, budget limits, supplier capacity and delivery windows all bound the feasible set; the scoring function then ranks what remains. Where an order can be split across suppliers, this becomes a small allocation problem rather than a single choice. implementation option

      

        
Heterogeneous attributes — currency prices, day counts, percentage fill rates, boolean availability — are normalised onto a common scale, then combined under learned weights into a bounded score. Attributes are normalised within the relevant comparison set so that a score means "relative to the alternatives available for this order", which is the only comparison a buyer actually faces.

      

        
Every procurement decision produces an observable outcome: it arrived or it didn't, it sold through or it sat, it cost what was quoted or more. Those outcomes can be used as a reward signal to adjust scoring weights over time. Because the retailer only observes the outcome of the supplier it chose, off-policy correction and deliberate exploration budgets are needed to avoid locking onto an incumbent. implementation option

      

        
Residual monitoring across sales, inventory, supplier pricing and delivery performance flags observations that the current models did not expect — a sudden price move on a supplier page, stock that falls faster than any forecast, a delivery window that slips repeatedly. Anomalies are raised as alerts and can suppress automatic ordering on the affected product until reviewed.

      

        
Recurring annual and weekly patterns are estimated once enough history is available, and applied as multiplicative factors on the forward demand estimate. Because the benchmark depends on the current sales rate and the longest relevant lead time, a seasonal uplift correctly raises the benchmark ahead of the season by roughly the lead time — which is when the order actually has to be placed.

    

  







  

    

      
Autonomous supplier support agents

      
The chasing, handled.

      
When an order doesn't behave, an agent picks it up: reads the supplier's own published policy, finds the right procedure and contact, drafts the communication, and records the issue against the order.

    


    

      

        
01
Order placedRouted to supplier, logged on the order book

        
02
Order expectedDelivery window tracked against the supplier's estimate

        
03
Order not receivedWindow closes with no confirmed receipt

        
04
Agent investigates supplier policyRetrieves the stored policy record for this supplier

        
05
Agent identifies the procedureReimbursement or refund route, and the reporting deadline

        
06
Agent identifies the contactCorrect desk or address from supplier information

        
07
Agent drafts the communicationOrder reference, quantities, dates and the policy clause

        
08
Draft held for reviewNothing is sent until a person approves it

        
09
Issue recorded against the orderStatus, correspondence and outcome stored on the order book

      

      

        
Autonomy level

        

          
          
          
        

        
Agents produce a draft and stop. Every message leaves the platform only after a person sends it.


        

        
Agents also handle:

        
    
          
  • Missing orders
  • Refund requests
  • Reimbursement requests
  • Delivery issues
  • Policy questions
  • Order discrepancies
    • 
        

      

    

  







  

    

      
The full loop

      
A system that is never finished learning.

    

    

      
      

        
Stage detail01 / 10

        Sales data
        
Transactions arrive from the retailer's systems and become the ground truth for everything downstream.

        
Select any stage on the diagram. The loop advances on its own — each pass is a chance for the engine to be less wrong than it was.

      

    

  






  

    

      
Dashboard preview

      
What the buying team actually looks at.

      
An illustrative view of the IntelliFind console. All figures shown are demo data from a fictional retailer.

    


    

      

        
        
OverviewInventoryOrder bookSuppliersForecastsIssues

        
Harbour Lane Retail · 3 sites · synced 2m ago

      

      

        

          
Total inventory value£418,720+2.4%

          
Products below benchmark137+19

          
Critical products12+4

          
Orders placed (7d)84+11

          
Expected deliveries29next 5d

          
Inventory coverage18.6 d-1.9

        


        

          

            
AI demand forecast — all categoriesunits/day · 30d actual, 14d forecast

            
          

          

            
Supplier scoresweighted, live

            

          

        


        

          

            
Inventory coverage by categorydays of cover vs benchmark

            
          

          

            
Sales-rate changes7d vs 28d baseline

            

              

                
Product7d rateChangeModel

                
                  
Tempered glass 6.1"41.2/d+34%Benchmark raised

                  
USB-C 65W charger14.0/d+6%Within variance

                  
Bluetooth earbuds g38.7/d-22%Change-point flagged

                  
iPhone 14 display5.1/d+12%Seasonal uplift

                  
Fast-charge power bank3.4/d-8%Within variance

                
              

            

          

        


        

          

            
Active orders12 resting · 17 market

            

              

                
RefTypeSupplierQtyScoreETAStatus

                
                  
OB-4471RestingNorthwell24081.4—Awaiting 84.0

                  
OB-4470MarketVantor6088.224 SepConfirmed

                  
OB-4468MarketHarlow50090.623 SepIn transit

                  
OB-4462RestingPacegate1,00076.9—Escalation armed

                
              

            

          

          

            
Supplier issues & alertsagent queue

            
OB-4451 non-delivery — reimbursement drafted
Ridgeway Trade · awaiting approval
2h

            
Price increase detected +8.2% — USB-C 65W
Pacegate Supply · score revised 79.1 → 74.6
5h

            
Delivery window slipped twice this month
Ridgeway Trade · reliability term lowered
1d

            
Refund received — OB-4399
Harlow Direct · £214.80 credited
2d

            
Seasonal model updated — screen protectors
Benchmark +31% through Nov
3d

          

        

      

    

  







  

    
In one pass
Problem to loop.

    
    
      
  • ProblemInventory, demand, supplier data and chasing all move faster than a team can track
    • 
      
  • IntelliFindConnects retailer sales and inventory data to continuously collected supplier intelligence
    • 
      
  • IntelligenceModels learn demand patterns and update their predictions as evidence arrives
    • 
      
  • OptimisationOpportunities evaluated on price, reliability, delivery, availability and more
    • 
      
  • AutomationThresholds and orders managed according to the retailer's configured rules
    • 
      
  • LearningConfirmed orders and new sales data adapt the models and the benchmarks
    • 
    

  







  

    

      
Contact

      
Talk to IntelliFind.

      
For retailers, suppliers and investors — we're happy to walk through the engine in detail, including the parts that are still design rather than deployment.

    

    

      

        
Ethan Taankemtaank@gmail.com

        
Matthew Johnsenmwojohnsen@gmail.com

        
        
Prefer a walkthrough? Mention your catalogue size and roughly how many suppliers you buy from, and we'll tailor the demo to your data.

      

      

        
Request a demowe reply by email