Reducing procurement time by 70% through an AI workflow tool

How we enabled a global business to cut procurement time across multiple geographies by 70% - whilst gaining more insights to their suppliers than ever before and creating a competitive bidding process.

Background

As an multi-national business with more than 5,000 employees and operating for over 45 years - the “formal” procurement process was embedded in an existing ERP with a long term technology partner - however individual teams were often resorting to a mix of email and Excel spreadsheets to handle many sidecar processes - with upwards of 80% of effort going into these unofficial tasks, largely done to quickly green-light procurement and stay competitive - where updates to the existing ERP would be costly and pegged at multi-year timelines.

Unfortunately, these off the books processes had on occasion led to human error - allowing unexpected spend to go through along with not capturing the correct supplier data and sign-offs required.

The team in charge of global procurement approached us to see if there was any way we could speed up the procurement process without transitioning away from their existing ERP.

Growing demands for more data

In the last 10 years, given their incumbent market position - but understanding that the world of the future is built on a backbone of technology, the organisation’s leadership has mandated a shift to being a data-led organisation.

This also has trickled down into the procurement process, both in the form of data for process improvement (logistics timeframes, competitive pricing/open bids, etc) as well for long-term concerns around ESG impacting the supply chain and choice of suppliers in order to meet ESG related commitments.

Finding a foothold

From our early doors discussions with the global procurement team - the main two areas of the procurement process that were time intensive and stood out as key candidates to be shifted to automation with AI were:

• Collection of ESG data: At present there were an average of 20 emails back and forth with an existing supplier as long with team specific spreadsheets created depending on the team in order to collect supplier level data for reporting purposes.
• Open tenders for bidding: Often new suppliers would be found using ad-hoc research methods, referrals or even inbound contact by the supplier - however there were no governance, trails or proper mapping of item to cost by supplier - thus teams were often not choosing the most competitive supplier choice.

Building a two-pronged vision

Our prototypical workflow tool was designed to handle these two concerns, whilst also layering on top of the existing ERP as a source of truth, as opposed to circumventing it.

As the company would procure many different classes of goods and services, the platform had to be dynamic enough to handle existing and new use cases without requiring additional IT involvement - in other words, users had to be able to select which data should be collected for both ESG and open bidding.

Fortunately, with the advances in LLMs - we envisioned a solution that would utilise generative AI in order to create dynamic intake forms and offer suggestive pathways to the internal user - making it almost as flexible as a spreadsheet, without the time, risks or overheads induced through the existing spreadsheet and email workflow model.

The system would take the format of a public facing supplier portal backed by an internal management tool - with full security measures such as single sign on internally and two factor authentication on the supplier side.

From vision to prototype

We agreed a baseline scope with the procurement team - aiming to achieve coverage of 80% of previously ad hoc processes - with a 12 week build time, highlighting the importance of a minimum viable product and garnering maximum results with the smallest possible effort.

In this case, 80% of all goods/services purchased across the organisation could be grouped into as little as 50 supplier types out of a total of 1,000+ distinct supplier types.

After this, we ran through existing records - bucketing supplier types into distinct groups - and looking at the number of attributes required to be collected both for a bid and ESG. Attributes which were present across more than 1 were classed as common, and their relationship based on supplier type mapped out for our suggestive engine.

The system was designed to be used conversationally, with a simple prompt such as “I want to procure n tons of wheat to our Antwerp depot, with a max open time of 90 days and maximum spend of €x,xxx per ton” being translated into a post on the supplier portal, all form fields to be captured and allowing the supplier to submit supplier level ESG data points - with a fully fledged customisation and double-check process post prompt.

Shifting gears to production

Once the prototype was built, the procurement team were happy to involve outside stakeholders in order to build out a production grade system - with all the security measures, audibility, scalability required for a full cross-org roll out - with an estimated 6-8 month timeline, switching from our fixed price prototype to blended model (part fixed, part time and materials) for production - enabling them to get a full understanding of costs along with complete alignment on outcomes.

At this point, their internal IT was fully involved and we set-up a shared project alongside them - along with utilising their existing cloud provider.

From a user adoption standpoint, we wanted to embed training directly in the system - for both supplier and internal users - so ensured that each process had a step-by-step user experience, allowing the user’s to focus on the particular piece of data to be entered.

In all, the partnership was successful - with early doors adoption rates within 3 months of launch at over 50% of the team, leading to a reduction in cost of 13% on average due to open bidding - with an expectation to capture almost all non ERP level procurement processes in the system within a 24 month timeframe.