PracticeWorking with AI: Delegation & Oversight
Automate it or keep it human
The artifact
Proposal: Automate competitive research briefs From: Tomas (Ops) To: you (COO) Date: Nov 12
The problem
Sofia produces our competitive brief every two weeks — a 6–8 page document covering competitor pricing changes, feature launches, positioning shifts, and notable customer wins/losses, assembled from vendor sites, review platforms, job postings, press, and our own win/loss notes.
It takes her about four hours per brief. She is our only competitive analyst and she is the bottleneck on three other things: the quarterly pricing review, sales battlecards, and the analyst-relations calendar. All three are late.
The proposal
Run the brief through an agent workflow instead. I prototyped it last week: it pulls the same public sources, structures them against our existing brief template, and produces a draft. Wall-clock time: about 20 minutes.
I've attached the prototype output next to Sofia's most recent brief. Honestly, they're comparable — the agent version is arguably better organised, and it caught a pricing-page change on Meridian that Sofia's brief missed.
The numbers
- 4 hours → 20 minutes per brief
- 26 briefs/year
- ~100 analyst-hours/year returned
- Sofia gets her three stalled workstreams back
Risk
Low. Output is public-source research, not customer data. Sofia reviews each brief before it goes to sales. We'd keep her in the loop.
Recommendation
Switch to the agent workflow starting with the Nov 26 brief. Sofia reviews and signs off; we keep the template unchanged so nothing downstream breaks.
Sofia's reply, forwarded to you by Tomas:
The prototype output is genuinely good and I don't want to be the person defending my own job here. Two things though.
The four hours isn't drafting. Drafting is maybe 45 minutes. The rest is checking — vendor pricing pages get cached weirdly and show stale numbers, review-site posts are often about the wrong product tier, and job postings get over-read (a "Senior Pricing Manager" opening doesn't mean a pricing change is coming, though it did that one time). I've caught something material in roughly every brief I've written — last month it was Northbeam's "new" tier that turned out to be a relabelled old one, which we'd have put in a battlecard.
On Meridian: the agent did catch that pricing change, and that's real. I missed it.
I can review the agent draft. I'm not sure how long a real review takes versus writing it myself, because I've never done it that way round.
Your task
Decide: adopt the agent workflow as proposed, adopt a modified version, or decline. Commit to one and defend it, including what your choice costs. Aim for 300–500 words.
150 more characters before you can submit