Skip to content

The work

Agents running in production.

Four real deployments within industrial client operations. Here is what each one was built to solve, how it works and what it delivered.

Four real deployments inside client operations. Here is what each one was built to solve, how it works and what it delivered.

Every deployment below ran on the client's own infrastructure, under their existing security and access controls. Their data never left their environment.

Manufacturing

Production planning and scheduling

Sequenced production schedules across a multi-plant precast operation, with rapid replanning when the day goes wrong.

SpeedCost
87% Less planning time
20 to 30% Setup savings

The problem

Production planning across multiple plants and their sub-teams was a manual, time- intensive process made harder by its reliance on the expertise of a few key individuals. Suboptimal element batching meant setup changes were more frequent than they needed to be and each one added avoidable downtime. Any disruption compounded these losses, and the dependence on a few people made each one harder to absorb. If the people who owned the plan were absent, no one was left to build or rebuild it. If plant teams were short-staffed or a material delivery arrived late, the plan had to be reworked and the rest of the schedule shifted with it. In every case, replanning was slow, and the lost time compounded an already inefficient process: it had already become a cost by the time a new plan existed.

How it works

The agent generates sequenced schedules with delivery ordering, optimizing element batching and setup time as it builds them. When inputs change during the day the agent replans rapidly, ensuring the schedule remains current. Because planners see the whole schedule rather than their own slice of it, they can plan across plants and teams, batching work more effectively, anticipating knock-on effects and balancing load where it is needed.

The value case

The agent's value lies in the time it gives back. Plans that once took significant effort to build and rebuild are now produced quickly, removing the opportunity cost lost to rework. Setup changes are fewer and less costly, and delivery dates are protected with genuine buffer rather than optimism. Planning output is also standardized: the knowledge of the best planners is commoditized and applied across every plant and plan, so quality no longer varies with who builds the plan or who is available on the day.

Manufacturing

Material mix optimization

Shortening the qualification cycle between a newly available input material and a mix that can actually be sold.

SpeedWasteCost
80% Shorter recipe cycles
12% Average carbon dioxide reduction

The problem

A new supplementary material can be one of the most effective ways to lower the carbon of a product, by replacing part of its most carbon-intensive ingredient. However, getting it into a saleable mix was slow. Its composition had to be evaluated, candidate combinations tested against the governing European standards, and material properties proven over a 28-day testing period. Because each lab-bound cycle took so long, only a fraction of the possible combinations was ever tested and there was no time to systematically explore how each one traded cost against embodied carbon. A material that could lower the carbon of a product waited behind the testing queue.

How it works

The agent generates candidate recipes from the full catalog of available materials and predicts the key properties of each, so a far wider range of combinations can be assessed before anything reaches the lab. It analyzes substitutions and optimizes for minimal embodied carbon, minimal cost or a stated balance of the two, depending on the goal of the work at hand. Standards compliance is built into the search as a constraint, rather than checked once a candidate has been chosen.

The value case

Speed to market is where this pays off: a shorter route from a new material becoming available to a mix that can be sold. With the most promising candidates identified before lab testing, lower-carbon materials reach the market sooner and ahead of competitors. The same work speeds up supplier decisions and gives a systematic view of how cost, mix properties and embodied carbon trade off against each other.

Manufacturing

Material strength optimization

Predicting product quality from mix and process data so a plant can tune its operational setpoints instead of over-specifying material.

CostWaste
75% Cycle reduction

The problem

Hitting a strength target reliably meant manually decoding mix design parameters and adjusting cement and water ratios through trial and error, guided by the designer's own experience. With results dependent on the judgment of whoever designed the mix, quality was inconsistent and performance varied between batches. The safe response to that variation was to add more cement than the specification required, keeping every batch comfortably above target. That margin carried a cost, both in the material itself and in the energy used to produce it. This cost was paid on every cubic meter produced.

How it works

The agent predicts product quality from the mix design and production data, then optimizes the operational setpoints that determine it across water, cement, aggregates and admixtures. With strength predicted in advance, setpoints can be tuned to meet the target precisely and held consistently from one batch to the next. Every adjustment respects the exposure-class constraints the finished product has to meet.

The value case

Less material and less energy per cubic meter, by removing the trial-and-error loop and the habitual over-specification that compensated for it. Per-unit savings were modeled with the client and are commercially sensitive, so they are not published here.

Mining

Drill and blast

Automated image analysis, quality assessment and reporting for an international gold and copper producer.

SafetySpeed
85% Time reduction
95% Accuracy

The problem

In mining operations, fragmentation analysis measures how well a blast has broken the rock and informs engineers whether the next blast design needs to change. The analysis was entirely manual. Engineers photographed the blasted rock with an object of known size in the frame for scale, sized the fragments in specialist desktop software, then compiled the report by hand. Each analysis took hours, throttling blast optimization. With results dependent on individual interpretation, they differed between operators and sites making comparison across the portfolio unreliable.

How it works

The agent takes the blast images, sizes the fragments and calculates the key size metrics, evaluates blast quality against expectation and generates operational recommendations for the next blast design. It compiles the report allowing the engineer's role to shift from assembling the analysis to reviewing and signing it off. Every image is analyzed the same way, giving results that can be compared directly across operators and sites.

The value case

An hours-long workflow reduced to minutes. The analysis holds consistent across operators and sites ensuring that blast optimization decisions rest on comparable numbers.

Professional services

Contract and payment review

Extracting commercial terms, benchmarking rate cards and identifying price or payment-term leakage across fragmented agreements.

SpeedCost
85% Less review time
$3.2m+ Price leakage found per cycle

The problem

Commercial terms are spread across contracts signed by different teams over many years. Rates, payment terms and price-adjustment clauses are recorded in different formats, making it difficult to check whether invoices match what was agreed.

Manual reviews are time-consuming and usually cover only a sample of agreements. Incorrect rates, missed adjustments and payments that do not follow agreed terms can go undetected for several billing cycles.

How it works

The agent extracts rates, payment terms and price-adjustment clauses from contracts into a structured, searchable format. It compares supplier rate cards, identifies potential discrepancies and monitors the commodity prices and exchange rates linked to each agreement.

Reviewers receive a prioritized list of issues to investigate, rather than having to read every contract.

The value case

Review time fell by around 85%, allowing the team to check every agreement in each review cycle rather than relying on a sample.

Each cycle identified over $3.2 million in price leakage, giving the team specific discrepancies to investigate and address. The extracted terms remain available for subsequent reviews, reducing repeat work.

Logistics

Logistics pricing review

Detecting and quantifying pricing anomalies, then creating executive-ready outputs for investigation and action.

CostRisk
$4.2m+ Exposure found per review cycle

The problem

Logistics pricing combines route-specific rates, surcharges, fuel adjustments, additional service charges and customer-specific agreements. These change frequently and are difficult to check across large volumes of transactions.

An individual invoice may look correct while concealing a wider pricing problem: a route priced below cost, an expired surcharge still being applied or a temporary discount that was never withdrawn. Without systematic checks, these issues can continue for months before their effect on margins becomes clear.

How it works

The agent checks each shipment's pricing against the applicable rates and commercial terms. It also compares pricing across similar routes, customers and periods to identify inconsistencies that individual invoice checks may miss.

For each issue, it estimates the financial impact, identifies how long it has been occurring and assesses how much may be recoverable. It prepares the findings for the team responsible for investigation and corrective action.

The value case

Each review cycle identified more than $4.2 million in financial impact.

The findings show where the issue occurred, its estimated impact and who needs to investigate it. Reviewing all transactions, rather than a sample, helps teams address existing discrepancies and check whether they recur.

Finance

Group forecasting

Retrieving and reconciling source data, then producing Base, Downside and Upside scenarios across multiple entities.

SpeedRisk
90% Less preparation time

The problem

The finance team had to collect and combine figures from several businesses to prepare the group forecast. Each business used its own ledger and reported its figures differently, so the team spent significant time gathering data, checking versions and resolving discrepancies.

By the time the figures were ready, there was little time left to explore different scenarios or respond to what the forecast showed.

How it works

The agent gathers financial data from each business and reconciles the figures into a consistent basis for the group forecast.

It then produces three scenarios, Base, Downside and Upside, for each business and the group as a whole. When assumptions change, the team can update the forecasts without repeating all the preparation work.

The value case

Forecast preparation time fell by around 90%, giving the finance team more time to analyze the numbers and act on the findings.

The team could routinely compare three scenarios rather than rely on a single forecast. A consistent approach to reconciling the figures also reduced the time spent resolving the same discrepancies each reporting cycle.

Start with one process.

Have a project, transformation or operational challenge in mind? Tell us what getting it right would unlock, and we will do the rest.

See one built on your own data.

Tell us the work you would start with. We will build an agent on it and show you before anything is committed.

Request a demo How we deploy