Recently, while talking to a care provider that spends about £2-3 million per year on agency staffing, we discovered that they already had a procurement company in place helping with pricing and contracting. This is smart.

However, when we asked the next logical questions:

• Which agency is least expensive for a specific role?

• Do multiple agencies cover any one site simultaneously?

• What is your actual "blended" cost after factoring in VAT?

We couldn't find anyone who could confidently answer these questions. Not due to a lack of effort, but due to a lack of speed.

We asked them to send us a stack of invoices to see if we could verify some numbers. The info provided consisted of hundreds of invoices from dozens of agencies with each agency using its own format. A few of the invoices were scans of paper documents and therefore not readable by computer software. To put it into context, if we broke each invoice down into one line for each shift worked by each employee at each rate charged, the invoices would consist of over 2000 lines of data. No one had ever put together a list of all of this information. There was no single spreadsheet containing all of this information.

Therefore, we did the unglamorous thing: took every line of data from every invoice; worker, role, site, shift, hours, rate, VAT status and then compared each item to the original invoice's total. If the items we extracted did not equal the total on the original invoice from the agency itself, we assumed that we made an error and not the agency. It was this method of comparison that uncovered all of the interesting information.

It uncovered the small errors such as misreading a digit or wrapping an employee's name around two lines of text or mixing two different VAT rates on the same invoice. It also identified an error in our methodology: initially we assumed the site from the file name of the invoice. Many agencies send their invoices via email as "document (32).pdf" and do not include the location in the file name.As a result of this assumption, we incorrectly labeled hundreds of lines of data prior to the reconciliation stage. During the reconciliation stage, we were forced to go back and extract the location information from the body of each invoice. When we finally established the accuracy of the data, this is what it revealed:

Agency spending was highly concentrated. Approximately one third of total agency spending was done with a single agency. The top three agencies accounted for approximately 60 percent of total agency spending.

Average rates were high. Shift rates for similar roles were consistently higher than what we would typically expect to see in similar geographic areas.

Sleep-in shifts were the worst. Sleep-in shifts were billed at approximately 30 percent higher than what we would consider average for similar situations.

Rates varied significantly based upon which agency billed them. Similar roles, similar types of shifts, similar sites but the rates still differed by nearly 30 percent per hour among agencies.

A few rates looked too low to add up. A handful of agency rates were low enough that, once standard employment on-costs, National Insurance, pension, holiday pay are factored in, there's barely room for a margin, let alone confidence in how the worker being supplied is actually being paid, whether via PAYE or an umbrella company, or whether it meets minimum wage obligations.

A number of shifts were billed twice. The same shift, for the same worker, billed twice. Not a one-time occurrence but a recurring problem; probably occurring long before anyone paid close enough attention to notice. On its own, this would have cost the client thousands of pounds.

There were inconsistencies regarding VAT application. Some roles were billed with VAT that should not have been billed with VAT. Conversely, some roles were billed without VAT when VAT should have been applied. The HMRC nursing agency concession applies only to registered staff delivering genuine direct medical care under the direct supervision of a qualified nurse, not the personal and support care that represents the majority of shifts in this sector. Getting this boundary wrong does not represent a minor discrepancy; it represents a potential liability that HMRC can pursue retrospectively against the provider, not the agency that miscoded it.

Nothing required anything fancy. No predictive models. No AI/ML tricks. The value derived solely from enforcing consistency upon documents that were designed to lack consistency; cross-referencing documents that are typically never referenced against one another; and refusing to accept an output unless it reconciled line-by-line back to source documentation. A procurement company reviewing rates and contracts is operating at a much higher level, rarely do they possess sufficient insight to sit inside 2000 individual shift lines; compare them against timesheets; and verify each one against its own invoice. This is a very different type of task and it is precisely the type of task that is often neglected because it is tedious, not because it lacks value.

An unfortunate reality for many providers is that they lack not analysis but a basic record of what they are actually purchasing and absent this record overcharging; multiple billing; and significant compliance risk can go unnoticed for years. If you spend significant money on agency staffing and cannot answer those questions today without spending weeks manually chasing invoices it is worth discovering what exists within your own data.

Your own data will typically reveal far more than people expect. In this case, the output was a clear, plain-English summary handed back to the provider with the finding laid out, and with clear recommendations on the questions to ask next. Whether or not they go on to become a client, they're better informed than they were before.

If you're concerned about how your agency spend is being managed and don't currently have the detail you'd need to make informed decisions, get in touch with Florence or email me directly at dan.ashenden@florence.co.uk. We'd be happy to help.