After launching a suite of artificial intelligence (AI) services to help clients with their digital transformation projects, Moore Kingston Smith turned that expertise in house, in a push to enhance and streamline internal workflows. The result is a bespoke, generative AI (Gen-AI) platform called AssureRight, developed to unify and streamline the firm’s internal processes, making day-to-day work more efficient, holistic and easier to manage. 

Moore Kingston Smith Director of Innovation Jared Goodrich says that the firm decided to develop the platform itself because its needs would not be met by off-the-shelf products. While impressed by the immense potential of Gen-AI, he says the firm was frustrated by a lack of progress in the software market to integrate that power into real-world workflows. 

A Gen-AI platform provides the features and tools to allow users to build their own workflows in a simpler, safer way. It does this by ring fencing the models and automating the prompts, inputs, parameters and safeguards, so that users can use it without worrying about these considerations. This automation also allows for the processing of a much larger number of requests.

“AssureRight is designed to enhance quality and productivity, while ensuring rigorous control over data privacy and security,” Goodrich explains. “Its main role is to empower our subject matter experts by providing them with tools to create, customise and share their AI workflows throughout the firm, or within their immediate department or team. The platform produces high-quality, standardised results, while its inherent transparency enables users to easily review outputs and verify them against source documents.”

AssureRight’s ambitious scope called for advanced infrastructure. That required the firm to harness the power of a sophisticated, large-scale Gen-AI model.

Ongoing assessment

In terms of its selection criteria, Moore Kingston Smith wanted a model with as low a hallucination rate as possible, a large context window, strong mathematical capabilities and reasonable pricing. It eventually chose Gemini 2.5 Pro. However, Goodrich stresses that AssureRight was developed with a model-agnostic approach.

“Gen-AI models are challenging each other and becoming ever-more sophisticated on a constant basis,” he says. “If OpenAI were to release a model that we felt was better, we would pivot very quickly. It’s important to carry out ongoing assessments of available models to ensure you are using the one most applicable to your business needs.”

As a fun way to stretch Gemini’s legs, Goodrich presented it and a rival model with two different versions of Mary Shelley’s Frankenstein. The second version had a single, changed word buried somewhere in the middle. Gemini picked it out very quickly, while the competitor struggled.

“Now, that wasn’t a control experiment in the strictest sense and therefore not decisive for selection,” Goodrich says. “It’s quite likely that Gemini had already ingested the novel as part of its training, which may have polluted the test. However, I fed in two versions of another book of around 116,000 words, published after Gemini’s latest training date, and again it found the single, changed word straight away.”

In Goodrich’s assessment, competition in the AI-models market is only set to intensify. “That will make things interesting, because OpenAI has been the frontrunner for so long.”

Human input

Goodrich stresses that creating the user interface is one of the most challenging aspects of building a software platform on a Gen-AI model. The interface’s effectiveness hinges entirely on that of the workflow operations behind it. Therefore, it must be designed with input not just from the technical team, but partners and other subject specialists.

“They are the experts,” Goodrich says. “So they must have direct input on the platform’s underlying processes to ensure they’re correct and aligned with how you need to work as a firm. Lots of software companies won’t be able to fulfil that ask by themselves. However, accounting and auditing firms have all the relevant expertise to fill in the blanks, because they understand how the platform’s use cases should inform its processes and the ensuing user interactions.”

From that perspective, Moore Kingston Smith had an inbuilt advantage. “My colleague Becky Shields saw around 10 years ago that building tech systems in house would allow for better, more bespoke solutions tailored to our needs,” Goodrich notes. “For example, we built our own, internal AI risk engine to analyse general ledger transactions. So, we’re fortunate for our development team to have dual expertise in accountancy and software development. That massively accelerated our development pace. We understand what we’re building, why we’re building it and who we’re building it for. And having users who can give us instant feedback makes it fast for us to iterate.”

Wild West

Firms of different sizes will need to take different approaches to finding the AI solutions that are right for them, Goodrich points out.

Small firms may be entirely reliant on third-party vendors. As such, they will need someone within the firm who is constantly monitoring which models are out there, what they can do and what sorts of data-privacy requirements accompany them.

“It’s the Wild West out there with AI products at the moment,” Goodrich warns. “A major concern for inexperienced buyers is ‘AI washing’, whereby vendors misleadingly play up AI content in their products that won’t make a huge, operational difference.” 

Those exaggerated claims will naturally come with inflated charges, Goodrich explains. You could also quickly land in a situation where you’re paying many vendors, each of which is addressing a very narrow problem. “You’ll end up with a patchwork of solutions that isn’t very holistic – and the costs can mount up significantly. Some of the vendors I’ve been talking to are trying to charge insane fees for very narrow problems. You’ll also have to carefully screen prospective products for data privacy concerns.” 

In the mid-tier – the upper end of which is Moore Kingston Smith’s berth – firms must consider a ‘build vs buy’ approach that is likely to blend in-house and external expertise. Goodrich knows several firms that are partnering with universities and/or software houses to share knowledge and experience as a means of coming up with appropriate solutions. “There are ways of making it work with partners if you don’t have in-house expertise,” he says. “But ideally, you need someone leading your project who will have a foot in both camps.”