Collegis
Where to Start with AI: How Collegis Turned 21 Ideas into 7 Priorities
Ask Collegis where AI could help, and the list fills up fast. Forecast upkeep, board deck assembly, invoice matching, close checklists: almost every recurring task looks like a candidate. The harder question is which ones deserve the team’s limited time, budget, and attention first, and how anyone will know it paid off.
That was the question facing Collegis, an education services company, this summer. The team was curious about AI and already experimenting with it. What it didn’t have was a way to decide where to put real resources, or a clear measure of what success would look like once it did.
Putting a number on every idea
Collegis picked five workstreams to examine: contractor forecasting, compensation forecasting, timecard and month-end close, the monthly board reporting deck, and revenue forecasting for existing and new business. In a series of working sessions, the team walked VFP through each process step by step: what they do, how long it takes, who is involved, which systems and spreadsheets they touch, and where things tend to go wrong.
Those sessions surfaced 21 opportunities for AI to take on manual work. A list of 21 is a wish list, not a plan. To turn it into one, VFP scored every opportunity against the same three questions:
Time today
How many hours does the team spend on this each cycle or month?
Effort to build
Is this a low (1 to 2 weeks), moderate (2 to 4 weeks), or higher (4+ weeks) build?
Dependencies and risks
What has to be true first, and what could go wrong?
That gave every idea a business case in the same terms: hours returned, weighed against what it takes to get them. It also made the trade-offs visible. A check that catches math and formatting errors before the board deck goes out was scored as a fast, low-risk win. A tool that builds the new-business forecast straight from sales pipeline data moved down the list, because the pipeline data needs cleaning first. Automating it early would only automate today’s errors.
What the numbers showed
Subtracting the one-time build effort from the hours each task consumes over a year, the full list adds up to more than 1,100 estimated hours the team could get back. Most of that sits in three places:
Existing-business revenue forecasting
About 530 hours a year, much of it spent keeping a very large forecast spreadsheet in sync with the system.
Monthly board reporting
About 310 hours a year, mostly copying numbers from complex workbooks into slides and checking that they match.
Contractor forecasting
About 170 hours a year of reformatting files and matching time against invoices.
Those hours aren’t headcount savings; they’re capacity the team can shift from preparing numbers to analyzing them, running what-if scenarios, and making strategic decisions.
From 21 to 7
No finance team can take on 21 automations at once, and trying to would spread the effort so thin that none of them land. The point of the scorecard was to make focusing easy to defend.
VFP brought the scorecard back as a recommendation and asked the team two questions: which of these would make the biggest difference to your week, and is there a case for starting somewhere other than where we would? Collegis chose seven. They led with existing-business revenue forecasting, the largest pool of hours even though it isn’t the quickest build, and paired it with smaller wins in contractor and compensation forecasting so value shows up early and keeps arriving.
Those seven account for more than 700 of the 1,100+ hours, fit into the first five months starting in October 2026, and each carries its own measure of success: the hours it is expected to give back. The remaining opportunities stay on the roadmap, sequenced behind them, and the pace can pick up as Collegis learns to build its own automations.
Foundation before solutions
Three things go in place before the first automation is built:
A deliberate start
Around 10 Claude licenses for the Collegis finance team, rather than a broad rollout on day one.
A secure connection to company data
An MCP connection so Claude can work with the company’s Salesforce and Certinia data, respecting each person’s existing permissions. No one sees anything they couldn’t see before.
Clear ownership
A named finance AI owner, human review of everything AI produces, and training so the team can eventually build its own automations.
Coming in Part 2
The scorecard predicted outcomes for every use case. In Part 2, we’ll check it: what happened when the first of the seven moved from the roadmap into daily use, how many hours it actually returned, and what Collegis changed along the way.
