Why machine learning could drive value in procurement

Data analytics can be used to regain control and automate price checking. Historical purchases are categorised at a low level and these categories are then applied to future transactions. As a result, live transactional data can be automatically linked to the catalogue prices originally promised by a supplier. Machine learning enables housing providers to use these ‘rules’ to quickly check product compliance across multiple merchants.

This mechanised data ‘bridging’ immediately shows whether suppliers have hiked their rates, whether it has happened before, how many times and if the increase is above price rises agreed at tender stage. It can also predict the likelihood of certain suppliers being non-compliant on pricing going forward.

This same AI process can help housing associations to go deeper in driving value. For instance, it can automatically flag whether staff are buying from core product lists or whether they are purchasing alternative brands at higher rates. It can also support product rationalisation, cutting a common but costly practice in the sector in which different types of one product are needlessly and often unknowingly procured across a housing organisation.

Identifying patterns in data can also support collaborative attempts to enhance efficiency, particularly where groups of housing organisations pool their data and use insights to influence the supply chain at a much higher level.

For example, at Procurement for Housing (PfH) we are looking at the possibility of combining member data to influence demand forecasting and manufacturing runs in line with preferential commodity periods (for example when copper costs are low). This could then put social housing procurement teams into ‘business advisor’ roles – guiding their organisations on when to reschedule planned maintenance to take advantage of optimal commodity pricing or exchange rates.

“AI needs granular data. But social housing providers haven’t been as strong as other sectors in gathering low-level data”

But there are a number of barriers to AI adoption in social housing procurement. Many innovative businesses at the cutting edge of predictive technology (often start-ups) find it hard to break into the social housing marketplace. This is partly down to not having enough capacity or experience to successfully bid for big public procurement contracts. But it is also due to the high levels of risk that housing associations attach to changes within the tendering, sourcing or transactional process – a perception that comes from tight regulation in the sector.

Another hurdle is the quality of information held by landlords. In order to operate, AI needs granular data. But social housing providers haven’t been as strong as other sectors in gathering low-level data. A key element of machine learning is taking unstructured information (for instance product descriptions) and classifying it across a category, across multiple suppliers and across multiple organisations – providing a standardised dataset to analyse. Without detailed records in the first place, this categorisation cannot take place.

Over the next two years, as data quality improves, the procurement workforce changes and the financial benefits continue to be proven, AI will become more mainstream in social housing buying activity. It may sound daunting, but really, machine learning is just about categorising data so it can be used to make better decisions. Hopefully the barriers surrounding this technology will continue to be dismantled so more social landlords can benefit from such a rich vein of intelligence.

Steve Malone, managing director, Procurement for Housing