QASI

We take on work of all dimensions, in any industry, from a home extension or groundworks of a few hundred thousand pounds to large projects worth several dozen million pounds. We treat all clients with the same professionalism, speed, efficiency and dedication.

The functions below are the full range of quantity surveying and commercial management work we cover. Each is handled by the specialist for that field, working with the AI instance they trained; the schedule of services describes how the nine service items combine them.

Write to us with an outline of the project or the problem and the contract form it sits under, and the specialist for that service item will reply.

QASI stands for Quality Assured System Integration.

This is not a buzzword salad, like so much of the marketing of supposed AI systems. What we have spent four years doing is ensuring that our dedicated AI software integrates the various aspects of the required workload with human competence built in and layered across disciplines.

Humans alone tend to make mistakes. AI alone tends to create a lot of filler and busywork, along with sometimes catastrophic "hallucinations". QASI used, and continues to use, human-perfected experience of real-life situations and events as training models for proprietary AI systems, to produce outstanding quality results in a fraction of the time and cost of traditional methods. The output is then verified by a human expert in the specific field of the output to ensure it is a coherent, relevant report, absent time-wasting superficial filler, irrelevancies, or contractual absences or errors.

For example, service items 3.0 Contract Review, 5.0 Dispute Resolution, 7.0 Specification Compliance and 8.0 Legal Framework are all related and may all be simultaneously applicable to a single event, while each can also operate independently or in a different combination. All of them are also potentially reliant on the often forgotten but extremely important 9.0 Document Archiving.

As anyone who has been in the construction industry for decades is aware, integrating all of these aspects of a contract or a site event can be very challenging, and missing an item in a document that was reviewed cursorily by a human months or even years before can have dire consequences. This sort of problem is almost entirely eliminated by integrating each aspect of the job, event, contract dispute or preparation for adjudication, relating the problem through several steps of human-directed AI research. But the issue is not resolved by merely issuing multiple queries to a commercially available AI.

A common issue that is now becoming evident with AI is that, while very fast, it can focus on the wrong or even irrelevant elements of a nuanced or intricate contractual or legal issue. There are also multiple documented cases of hallucination in the commercial models. We have listed a few reference documents at the end of this page, and new issues are reported daily in this field.

Relying on the standard commercially available AI models often produces reports, results or suggested action policies that, while appearing slick and professional at a superficial language level, are quite often a net negative, or detrimental to the real situation being faced.

Our network of experts has increasingly come across this issue with firms that engage AI in their responses. Instead of providing a benefit, such efforts currently make it easier to find the problems those firms are operating under. Often the work is produced with AI use undeclared and presented as if performed by the responsible human, but through our experience of working with AI-integrated systems we are very familiar with such documents and can normally tell them apart from correctly done work very quickly.

We have guarded against that issue, which becomes worse over time: recent studies of AI-augmented business have found that despite initial apparent efficiency gains, the longer-term curve is catastrophic in many cases. The managerial class traditionally responsible for oversight tends to become de-trained, and a complacency based on not knowing how the AI actually works becomes almost institutionalised.

In the UK, what exists on AI accuracy in construction is vendor marketing (95 to 99 per cent accuracy claims) and trade-press anecdote, together with our own witnessed experience that the failures are systematic rather than random: scope interpretation, specification-driven requirements, undrawn work such as conduit routing. That is the dangerous kind of failure, because a bid can be mostly right and wrong on the critical line.

From inception we have worked in such a way that our long-term approach is corrective and constructive, based on facts from hundreds of documented instances across dozens of contracts. We have thus eliminated much of the assumption and hallucination so prevalent in the standard commercial models. Our proprietary model is specific to our industry and produces nothing without expert human oversight, through systems we have developed over four years and tested in the real world.

Specific issues with the legal aspects, and references

Most of the work done on checking AI-produced output comes from the USA, because the UK market is still somewhat behind, but the issues are already prevalent in our experience. The following provides some context.

The best-documented evidence of what goes wrong when AI output is sent out unchecked, or produced by confident operators without long experience in the specific discipline, comes from the law, the discipline closest to contract review and dispute work.

Stanford's RegLab benchmarked the leading paid legal research tools on over 200 pre-registered queries and found them hallucinating in more than one answer in six (Lexis+ AI and Practical Law above 17 per cent, Westlaw's AI-assisted research above 34 per cent), concluding that lawyers would need to check every proposition and citation, which erases the promised time saving.

The consequences are now a matter of record. A tracker maintained by Damien Charlotin counts 1,240 US court filings with fabricated AI citations between April 2023 and July 2026, rising from 11 in 2023 to 524 in 2025 and 668 in the first seven months of 2026, with 97 monetary sanctions and 59 disciplinary referrals in 2025 alone, and lawyers (not only litigants in person) responsible for a growing share.

In England the Divisional Court addressed it directly in Ayinde v Haringey and Al-Haroun v Qatar National Bank (June 2025), where 18 of 45 cited authorities in one case did not exist. Dame Victoria Sharp held that such tools "are not capable of conducting reliable legal research" and that the professional duty to verify against authoritative sources is undiminished by using them; wasted-costs orders and referrals to the Bar Standards Board and the SRA followed.

The pattern in every one of these cases is identical: a fluent, confident output that was never checked by someone qualified to check it. That is the gap QASI's method is built to close. Each AI instance is trained in one field by the specialist who practises it, and nothing it produces leaves the team until that specialist has vetted, checked and queried it.

Sources

A few of the many available sources for our contention and lived experience.

Bills of quantities, cost plans and tender pricing, measured to NRM and priced against live rate libraries. Even when built from a project library these can still come up short if not tested against current and live suppliers. Global events can rapidly change the pricing of prime materials, and the contractual position in such cases needs to be clear.

Where we have tender documentation issued as a blank BoQ we can price the whole schedule, sometimes in minutes, if we are supplied with costings from the specified suppliers. But even when that is the case, we still verify things at a human level and check the priced elements, especially when these are unique to a project or fall away from industry-standard specifications, which are simultaneously checked in conjunction with the requested materials and elements of work to ensure compliance.

We also verify any weaknesses in the pricing relating to contractual issues such as changes in prices due to force majeure events, or identified risks. We are also beginning to quantify these risks on the basis of global events that fall well outside any actually issued contract documentation. Given the fluidity of current events, these are issued as suggestions or risk-mitigation possibilities based on possible outcomes, for the client to evaluate and make decisions on, rather than built-in assumptions that any AI would likely bake in to its output.

Case studies

We have priced various BoQs for clients in anything from minutes to a few days, with projected profit margins clearly identified. The results at the end of the project tallied with our projection to within less than about 4%, even in the few cases where disputes arose to create quite large divergence from the original pricing.

Take-off, re-measure and valuation of works, with every quantity traceable to the drawing, revision and clause it came from. While much of this aspect still needs to be performed with human effort, the review of take-offs from drawings is a far quicker exercise using our integrated AI system.

We can retrieve industry-standard costs quickly and provide an overview of realistic cost expectations for a project, including for as many elements as required, including design and legal representation for contract creation.

We are currently experimenting with proprietary AI-measured drawings, and expect to be able to verify the process at our strict levels of accuracy in the near future.

Case study

One instance of a large, multi-million-pound stadium refurbishment project, whose cost had been estimated by both client and designer to be within a certain budget, was put on hold once a more detailed QASI analysis was done of real-world costs. In this instance the project did not go ahead, but this was definitely a net positive for all involved. Had the QASI detailed review not taken place, the client would have run out of money before the project completed; the construction firms that would have been involved in the second half of the project would have suffered losses, as the first part of the project was funded but the second part was not; and the designer's specifications, though professionally excellent, exceeded the budget the client had allowed for.

JCT, NEC and bespoke forms read clause by clause for risk allocation, payment mechanics, notice periods and amendments that shift liability. We have done the most extensive work in this regard, especially where there are additional legal obligations, contractor's and employer's requirements, schedules, timings and other critical issues that are often overlooked by purely human reviewers, or misinterpreted or hallucinated by standard AI models.

Both before and after contract issue and signing, this is a critical step that traditionally many contractors and their clients have largely assumed, often based on documents that are a standardised version which does not necessarily reflect the details of the actual project.

We have human experts whose whole function was purely the review of every contract document, both prior to acceptance and after, in order to be aware of the dangers, expectations, penalties and specific issues that would trigger compensation events, extensions of time, non-compliance and all other aspects of a construction project. Their training of AI to be aware of the weaknesses and strengths of every contract, in its totality and in the context of all the applicable contract documentation, is possibly unique at present and certainly at the leading edge of the field: we are unaware of any other initiative, network or firm that has dedicated this kind of resource and technology to this aspect of the construction industry. A core function that is ultimately the foundation for everything that takes place in any project.

Case studies

Both in projects that went to adjudication and in those that resolved before it, the QASI review of the situation produced positive results for our clients:

• Wins, including financial compensation for events they had suffered.
• Negative outcomes at adjudication avoided, by settling the issue before it escalated.
• In two cases, a warning to our clients about the current state of their disputes from a legal standpoint, and the mitigation strategies going forward to avoid the maximum financial censure from a negative adjudication.

In all instances our clients had already entered into contract without the benefit of a QASI contractual review prior to acceptance of the contract conditions.

Identification, pricing and substantiation of variations and compensation events, with contemporaneous records assembled as the work proceeds. This is an especially strong aspect of our work.

This is especially efficient where the QASI network has been used from the very start of the project to quantify and produce all pricing, has verified all the contractual positions, and so on. In such cases the production of financially and legally sound variations is a very quick exercise. However, even when inheriting an already executed contract, this work can first be regularised and brought up to the correct standard if it had not been, evaluating each position on its merits and then implementing the regular practices that ensure the situation is streamlined going forward.

Case studies

There have been several. As most contracts have variations, and our review process includes for items often overlooked, in conjunction with the various contractual stipulations, we have been able to secure pricing of variations that had been in dispute for months, by reference to the specific legal case each fell under. In some cases we have availed ourselves of somewhat obscure contractual principles that allowed an increase in the pricing of fundamental material, such as reinforcement, due to global events that had increased the price gradually over the extended period of the contract. These being large contracts, the variations, which in some cases had not even been identified, came to several hundred thousand pounds of value.

Preparation of claims, responses and referrals for adjudication, mediation and arbitration, built on the documentary record rather than recollection. An aspect we have extensive experience with, with optimal results, thanks to our specifically tailored approach to all legal aspects of the construction industry.

One of the aspects QASI has focussed on from the start, as a sizeable portion of the work done over the last four years has dealt with various forms of dispute resolution. As explained in the QASI overview, the legal aspects of AI work are among the least reliable and most important areas, where AI can create very serious issues instead of resolving problems or simplifying the legal challenges faced.

The QASI approach has specifically ensured that open-source LLM AI has been trained on existing case law (UK and FIDIC mainly) as well as tested on actual projects where we provided assistance to our clients in a dispute. Because of the rigorous, fact-based training data that is proprietary to the QASI network, we obtain legal resolution results from our AI in the first instance that are an order of magnitude more reliable than the commercial products available on the market today. After that, verification by a human expert is performed before the answers are also stress-tested by different legal AI models and perspectives, each with its own human expert oversight. The final product is then reviewed once more by a human expert before issue. To date, no document we have issued within the legal aspect of the construction industry has been found to contain a hallucination or a misrepresentation of any legal principle.

At times this has also meant our clients discovered that their position was not as favourable as they expected, because they had entered into the contract without the benefit of a QASI review of the contractual documents they signed up to. It is therefore advantageous to have a QASI review of any contractual documentation a firm wishes to issue, as well as to use the QASI network to verify a position either before one enters into it or, where it has already been agreed, to review the actual position's weaknesses, strengths and possible conflicts, or even contractually invalid clauses, which unfortunately do appear in some traditionally created contracts.

Case studies

Intimately related to the contract review process, this is a field in which we have had a great deal of experience. Our successful outcomes range from obtaining financial compensation for our clients for events they suffered, be they contractor or employer (we have worked for both), to reducing the inevitable costs of making good, compensation, LADs or non-compliance arising from undeniable and contractually indefensible situations that arose within a client's contract prior to any involvement with a QASI-related service or person. Often our services have resulted in a peaceful and fair resolution for both parties, reducing the need for adjudication or further legal steps.

Checking of measured work, valuations and final accounts for arithmetical, contractual and specification errors before they reach the client. We also run constant quality control on our own output, and have done so from the very start, which is what has allowed us to build systems that continue to outperform the standard AI models everyone else is starting to use.

As we have explained in our overview of how QASI works, every issue of any QASI-produced work is vetted by an experienced human who has verified and checked all the AI-assisted work in accordance with their decades of experience in the field.

Case studies

Having verified the design against contract specifications and proposed supplier details, our reviews have prevented contracts that were at signing stage from being entered into, due to the discovery of incompatibilities between the three aspects of the project: design, specification and supplied product. Sometimes these issues are very subtle, having to do with the sourcing of original material, the finishing process of a product and so on, and go beyond the mere acceptance of a proposed design drawing against the manufacturer's issued processes.

Cross-checking of installed and proposed work against the specification, standards and the employer's requirements. Standard practice for our outputs and standard practice internally: our experts and their specifically trained AI are cross-trained with each other, forming a resilient network that can pass any problem through several layers of scrutiny at a speed unmatched by human operators alone, and a quality unmatched by standard AI models.

Very often, specifications are standard documents that get adapted to multiple projects, in some cases with inadequate vetting of the detailed requirements of a particular project. Such issues can be discovered and resolved prior to tender or contract documentation being issued, as well as highlighted in the case of a contractual dispute or inconsistency where the specifications were accepted as a matter of routine without the necessary detailed review having taken place.

Specifications are also important for the various compliances that need to be verified with regard to design, material compliance, supplier origin of the product, tolerances of any detailed element, and any other unique requirements. All of these aspects can be crafted with precision prior to official document acceptance, as well as verified against supplier-provided elements of the work.

Case studies

Intimately related to quality control, the QASI systems have provided reviews of specifications in conjunction with the contract documents, as-built information, O&M manuals, and all the various documentation and compliance relating to H&S concerns, the CDM Regulations and so on.

Statutory and contractual context for every commercial decision, including the Construction Act, payment regulations and limitation periods, along with any specific changes, addenda or unique contractual details relating to a particular contract.

As we have seen in the overview of how QASI operates, the legal aspect of any corporate situation is the one in which the various commercial AI models have been tested the most, and found to create the most serious kind of problems: hallucinating entire case law, misreading or misrepresenting issues, cases and legal statutes, and so on. We have tested this aspect of our AI-assisted work the most, and have always ensured that any legal advice or contractual position is vetted by the responsible trainer of the specific AI module used.

We have also ensured that the various AI modules are able to integrate with each other to find and resolve any contractual inconsistencies, legal loopholes or nested clauses that create unique legal scenarios. At each step a human expert provides oversight of the process: from the individual modules to the nested integration of their results in various configurations, the details are always verified by an experienced human operator with decades of experience in the field.

Case studies

We have provided the legal framework and current position to clients who were evaluating the potential risks of initiating adjudication based on the state of affairs they were operating under. In every instance to date, both in the cases that went ahead to adjudication and in those that did not, or conversely where the client's counterpart initiated adjudication instead, our prediction of the potential outcome based on specific risks tallied very precisely with the final outcome. The risks we identified were inevitably the very ones that became fundamental in the process; even where such risks were not fully explored or taken advantage of by either our client or their counterpart, the specified risk itself was the lever used, proving a consistency in our evaluation that, as far as we know, is unique in the industry for AI-assisted work. We are certain this is due to how carefully we have created our own proprietary process of AI training, coupled with constant expert human supervision.

Structured retention of correspondence, drawings, valuations and notices so that the record of a project can be produced in full, years later, in minutes, as can every iteration of any contested issue over the life of the project. Very large archives of emails and documentation can be searched on specific targeted topics in days instead of weeks or months.

Structured retention of all documents pertaining to a contract should be standard operating procedure on any building contract, but even when done correctly, the sheer volume can be overwhelming when researching a specific event or issue, particularly if the original staff on a project have rotated out or changed over time. Our approach allows for proper document retention from the start of a project, and allows us to sift through enormous amounts of archived data, be it emails, issued schedules, timelines, valuations, additional-work variations or agreements and so on, in a fraction of the time, and then to order the material in the correct context or chronology to present the factual state of events as they occurred.

Case studies

In almost every project where the QASI initiative was used, research of massive quantities of documentation had to be performed, sometimes on one person's single email archive totalling 30 or 40 GB of email data alone. Additionally, the review of various iterations of multiple documents would have taken several months to analyse even cursorily by a human. In some cases even AI would take a couple of weeks to analyse and sort massive data archives, as these had to be split into manageable chunks and their output separated by the specific information being sought, which at times applied to multiple things, or conversely where multiple details of an event were required to decide properly on its validity.

The effective time saved by the QASI proprietary way of dealing with such issues saved many months of billable hours for our clients and, more importantly in some cases, produced work of a quality of detail, specificity and clarity that would have been unthinkable a few years ago, even with several months of dedicated work by a handful of human employees. In one instance approximately 6 months of human work was resolved in 5 days. While this is not always a typical result, it shows the advantage of the QASI approach to construction project management.

The headline

An anonymised £50m project we did, 18 months on site, as delivered against the same scope run through the 2026 QASI process

As part of our own process of R&D we actively did each stage of the work described in detail below both with traditional quantity surveying and commercial management staff we employed and, at the same time, had the same work performed by other, equally capable staff using AI.

This was three years ago, when almost no one was using AI in any viable commercial capacity, much less in this rather expensive, but very necessary, test of performance.

The first iteration of this AI-augmented work was carried out at the time, in 2023, in parallel to the actual human work done, but was not implemented at the time. It was merely for our own R&D, to see how viable using AI in our work was or would become. The second iteration is what QASI is capable of doing today, after the updating of the various AI protocols used and extensive training of each AI model we use on dedicated hardware.

And yes, we did run the entire thing from scratch again, using the information originally received, with the same time gaps between receiving it (often, waiting for specific information from the client is a large percentage of the time taken to complete a task). Our more recent test, performed in August 2026, is the second iteration of the same work.

It also needs to be stressed that the latest version is not just faster, but more incisive, direct and cleaner in its presentation. The salient points of the project are highlighted as priority items, the order of operations is better outlined (with reasons provided if requested, but not cluttering the executive summary or operation reports from the start), and the cost savings from value engineering are also more concisely and accurately expressed.

Keep in mind that we continue to run various tests, experiments and other research alongside our day-to-day work, so the performance of our QASI systems is on a continuing trend of refinement for efficiency, speed and, above all, quality.

Below the time comparison, we have a cost comparison analysis, so you can get a sense of expected real-world spend.

Five ways of doing the same job

The two outer columns are what a team of people alone would need to reach each quality level; the three inner columns are what happened and what the AI-assisted process does. The 2026 figures are in late-2026 money; for a like-for-like comparison using the 2022 money equivalent, see the note below the table.

The scope-gap problem

Over the project, £4.80m of subcontract cost could not be recovered from anyone. The records split it into six causes. Two of them, an unpriced structural package and work falling between packages, are the kind a systematic commercial process catches at tender and order stage; on this project the design review sat with the client’s own technical team rather than with our commercial team, so the package scopes were never reconciled against the design as it developed. The other four have a ceiling that no process, human or AI, can lift, because the cause lies outside the commercial team’s control.

Method

The project is anonymised: a £50m scheme, 18 months on site, early 2022 to July 2023. The as-delivered column is the actual cost and price. The 2022–23 AI-assisted column was not modelled afterwards: it was a parallel exercise run for our own internal measure at the time, on the same information as it arrived, a few weeks behind the human-only work. The 2026 column is the same scope re-run through the current system in August 2026. The two human-only columns, the effort a team of people alone would need to reach each quality level, are estimates produced by the 2026 system. Late-2026 figures are restated at a Consumer Prices Index factor of 1.25 (ONS CPI, February 2022 to December 2026) so that purchasing power is like for like.

QASI is a group initiative that grew out of a team of freelancers working together that has commercially managed circa half a billion pounds of construction work over four years: NHS hospitals, GMP and cell-therapy manufacturing facilities, power and battery-storage plants, large residential schemes, prisons and rail, under NEC, JCT and FIDIC forms, from first cost advice through to final account and adjudication.

The process of getting ahead with AI began in 2022, when AI first became available for practical use. The group started training various AI models on specific tasks drawn from singular aspects of quantity surveying and commercial management work, and tested them against live projects.

The work was never performed solely by AI. From the start the approach was to train and teach different AI instances and different AI software on specific tasks: legal framework, quantification of costs, variation work, dispute resolution, document archiving and so on. Each AI was managed and trained by an expert in that specific section of the profession, and that expert remains responsible for it. Our representatives who trained the AI are also familiar with construction and commercial work in mainland Europe and Asia, not only the UK.

Today the QASI initiative owns its own server and hardware, running specifically trained AI instances that are unique to it. Some of these are open source AIs that have been specifically trained exclusively in their own field, so that their programming is not affected by the limitations placed on the more commercial and mainstream AI products. Over the years the team has developed its own proprietary versions on that foundation.

The result is AI-assisted work of outstanding quality, produced in a fraction of the time. But AI-produced work is never simply sent out as the result of a query. It is vetted, checked and queried extensively by the expert human who trained the instance in its specific field.

Now, having reached a level of quality control, confidence, and processing capacity that has been extensively tested and refined in real application in the field over the last four years, the team has launched a brand new initiative on that foundation. That initiative is QASI.

Why QASI is a group, not a company

QASI is a working association of independent specialists, each trading in their own right, who have worked together on the same projects for years. We have kept it that way deliberately. Each of us carries our own professional standing, insurance and accountability, and the client contracts directly with the specialists doing the work rather than with a company that assigns staff to it. There is no overhead to carry and no payroll to feed, which is part of why the pricing on the Comparative page is possible. And a loose association is the right shape for the way the work is actually done: the specialist for a given service item, the AI instance they trained, and the colleagues who cross-check the output, assembled around the problem rather than around an org chart. Should a client’s procurement rules require a single contracting entity, one of our members’ companies can act as lead for that engagement.

Why we don’t name ourselves here

Much of what we do is dispute work, and much of the rest is commercially sensitive to the clients we do it for. Our members are often engaged where a name on a public website would compromise the position they are there to protect, or would be read by a counterparty before it should be. So the site describes the method and the record, and the people introduce themselves in person. Write to us and the specialist for your service item will reply under their own name.