Ask two staff members the same question and you could get two different answers. Nine tools behind one login closed that gap, including the diagnosis that once got built live while the client waited.
/ The problem
In a service business, quality comes down to the small judgment calls staff make dozens of times a day. Is this document set up right? Is this opportunity actually a match? Why has this client applied fifty times and heard nothing back? What does the policy say about this exact situation?
Those answers lived in a few experienced heads, plus a policy document everyone meant to reread. So the answer a client got depended on who they happened to ask, and a change in policy took weeks to reach the people doing the work.
There had been an earlier version of this idea, built fast. It let whoever used it choose how much each request cost, with no password in front of it. Eventually someone pointed a script at it and ran the bill up.
/ What we did
We built one internal site with nine tools behind a single login: paste in a client's document and get the exact fixes it needs; paste in an outside opportunity and get a plain yes or no on whether it's a real match; get ready-to-use search terms when good matches run dry; build a full sourcing plan with named target companies; ask a policy question in plain English and get the answer plus the exact email to send; the same thing again for the coaching team, from their own separate rulebook; work out which job titles a client can honestly claim, given what their work permit allows; diagnose a client who has applied dozens of times with nothing to show for it; and a standing reference for the eligibility questions that come up every week.
The diagnosis tool is the one that changed how the day runs. It looks at three things at once, whether the client is aiming at the right thing, whether their document reads as a match, and whether it is pointed at the right industry, and answers in seconds. It used to be a separate app, and before that it got built live while the client sat waiting.
The policy answers aren't written into the software. A job runs each day, merges each team's source documents into one master document, and the right tool reads from that. Two teams, two rulebooks, one method, and a policy change reaches everyone the next morning without anyone touching the software.
one judgment, not three, and it comes back in seconds
/ What changed
Whoever picks up the client gives the same answer, and the answer is current.
Calls start with the diagnosis in hand, instead of fifteen minutes of building one while the client waits.
Each use costs somewhere between a fraction of a cent and about a penny, so nobody rations it.
The hole that got the earlier version abused doesn't exist here.
/ Decisions worth explaining
Those settings are locked inside the tool and can't be changed from outside. That is the structural version of the fix. A spending cap added on afterward would have left the same door open.
If the password is ever missing, the site returns an error instead of letting people in. The failure you don't want is the quiet one that opens the door.
Aim, document, and industry are decided together, so they get weighed against each other. Asked separately, they come back confident and contradicting each other.
It was tempting to just hand back a better-sounding job title. Instead it checks whether the client can honestly carry that title first, and when they can't, it returns the path to get there and the backup route, rather than a relabel that falls apart the moment someone asks a question.