At a Glance
The Problem
Approximately 70 tenders arrive at Wavehill every month. Every one of them must be sifted through, then a decision made on whether or not to bid for the work.
In order to make that decision effectively a consultant needs to understand the nature of the bid, the scale of the work and the wider market context. Extracting that information and making judgement calls at each stage of the process caused a significant administrative burden. That is where the time went.
Capturing the tender.
Every document had to be read by the administrative team who extracted specific data fields and entered these into a database where the rest of the business would find them. A read plus a piece of manual entry, on 70 tenders a month.
Back officeVoting on the project.
The consultants then opened the document, with only the title to give an initial understanding of the project, reading through all the details and give their decision to bid or not to bid.
ConsultantsReading it when it was not theirs.
A tender went to a wide audience, so several consultants would each spend a few minutes working through a document only to establish it was not for them. Whether the right person then heard about it relied on somebody passing it on.
ConsultantsKeeping it moving.
The back-office team had to manually chase people on tenders without responses.
Back officeFiling & Recording the outcome.
Once a decision was reached, the administrative team came back to it: filed the document in the right place, then updated the record accordingly. Another administrative task on every tender.
Back officeThe cost was not the length of any one step. It was 840 tenders each going through six administrative steps, 216 hours a year that could be automated.
What We Set Out To Do
Four intended outcomes:
- Have AI handle the data extraction, the system handle the entry and channel post. Capture should be a single action, not a sequence of them.
- Give consultants enough to judge relevance at a glance, so they can see if the nature of the bid and scale of the project is worth looking into.
- Route every tender to the people it concerns, reliably, rather than depending on someone choosing to pass it on.
- Make the chasing automatic, so nothing goes quiet by accident.
Underneath all four: build it inside the tools Wavehill already used. Most automation projects fail on adoption rather than technology. If the team has to learn a new system to get the benefit, plenty of them will not.
The Process
Discovery and process mapping. We mapped the tender process as it actually ran rather than as it was described. Every step, who touched it, what moved and where the time went.
Prototype in a test environment. Built against sample tenders and deliberate edge cases before anything touched live data.
User acceptance testing with the people who run the process. Hands-on sessions with the people doing the work daily rather than the people signing the budget. Their feedback changed the build.
Phased production go-live. Released in stages over a few weeks rather than switching everything on at once, starting with the data gathering automation. Each stage was proven in live running before the next one shipped.
Iterating in the open. On feedback from the team the approval model changed to better track approvals with conditions. The design changed while it was running, deliberately.
How It Works Now
1 Step
The team puts the tender document in a folder.
That is the entire manual process.
Everything after that runs on its own. AI reads the document and extracts the data fields, so nobody has to enter them anywhere. It writes a short summary of the project and scores the tender against the criteria Wavehill set. A Power Automate flow posts the tender, tagging the consultants who fit the sector and location.
Those consultants vote in the thread as they did before. AI reads the replies and determines if they are an approval, rejection or comment. Once a tender has been approved or rejected, a second flow moves the file to the right folder, updates the record and posts a confirmation. Anything still undecided is chased on a schedule tied to its own deadline. Nobody has to remember any part of it.
The Results
216 hours a year theoretical time saved.
216
hrs/year automated
1
manual step left
840
tenders a year
The data entry step is gone. The back office no longer reads the document to extract the data and enter that data into the SharePoint list. That whole section of the process is cut out. They input the document and everything after it happens on its own. The same goes at the other end: the file move and the status update after a decision are no longer theirs to do.
Consultants can quickly determine if a tender was desirable. The summary and the score give them enough to judge relevance at a glance. The full document is still there to give the details when they need it.
Tenders reach the right people by default. Routing is no longer a favour somebody does. It happens on every tender, to the people whose sector and location fit.
Chasing became a system. Reminders go out on a fixed cadence against every open tender, to the people who owe a decision. No manual reminders.
Consistent data. The same fields are always captured. The criteria are now applied identically every time, so one tender can be compared against another with confidence. Consistent data is worth having in its own right.
Adoption was immediate because nothing changed. The build sits inside the tools they already used. No new interface, no new logins, no new habits.
In the administrative teams’ own words, “this is not something that gets put on a shelf.”
Where to Start
If a document arrives and three people have to touch it before anyone decides anything, that is usually the first thing worth fixing.
No single step looks expensive. Multiplied by everything that arrives in a year, it usually is. Book a discovery call and we will map where your time is going.
Book a Free Discovery Call