In this case study, we focus on a single agent with proven value: Peter, the proposal expert. Writing proposals is a time-consuming task in advisory services. In fact, it is often seen as unpaid work; the outcome is uncertain and it typically requires input from senior staff to stand a good chance of success. Peter has therefore been designed to use available data sources as context (such as previously won proposals, lost proposals, email correspondence, price lists and consultant availability) and use this to draft a compelling proposal for future clients.

Figure 1: Peter draft a proposal

Once the proposal is ready, Peter can share it with a second agent, Sylvia, who immediately checks the document against internal administrative guidelines. This forms a final agent-to-agent (A2A) check before the proposal is reviewed and completed by a human consultant.

Figure 2: Sylvia, the administrative agent, checks the proposal before it is sent.

As a final step, an agent can export the work to an organisational template, complete with the correct formatting, font and logo. This is a crucial step, as working with agents without specific templates results in significant formatting work for the consultant involved, such as copying, pasting and reformatting in Word.

Figure 3: Using templates in AIAIAI

The business case for Peter is straightforward. Our client writes around 50 proposals a year, spending an average of 60 hours per proposal, which equates to about 3,000 hours on proposals annually. Peter and Sylvia are not yet perfect, but they already handle around 80% of the proposal drafting today, with the remaining work consisting mainly of expert content reviews. This saves roughly 2,400 hours per year. At AIAIAI, we expect this 80% automation rate to increase gradually as the model's context becomes richer through feedback on previous work, and as model quality continues to improve.