The pre-RFP playbook, for the person who finds public-sector work. Using AI for the four jobs of that seat: finding the deal early, reading the field, knowing the buyer before the call, and answering the solicitation. A person decides on every page.
Switchboard · withswitchboard.com · Book a call: withswitchboard.com/book-a-call
A working guide for the person who finds public-sector work, on using AI for the four jobs that make up that seat: finding the deal early, reading the field, knowing the buyer before the call, and answering the solicitation. Four working prompts. A person decides on every page.
From Switchboard. We build custom software around how a team already works, with AI where the reading is and a person at every decision.
How AI changes the four jobs of the person who finds public-sector work by hand. Written for the person who does the finding, whatever the firm sells.
What a reader that never tires does to the field, how one record replaces four browser tabs on a buyer, why fit and stage are two scores, and what stays a person's call.
Run your numbers first. Mark where your firm enters. Four sections hand you a prompt for your own AI to run on one of your own documents tonight.
Reading is a machine's job now. Deciding is still yours. That rule holds the whole page together. Every prompt below reads something and stops, one document at a time.
When this page says a reading system, it means custom software built around how your team already finds work, with AI on the reading and a person at every decision: the kind of software section 09 describes.
Five numbers from your own record. The output is what moving half your late pursuits early would be worth, in your own numbers.
The difference between the two, a year, at what one win pays you.
On the defaults: 30 pursuits a year, 70% first seen at the solicitation, a 35% win rate when early and 10% when late, one win worth $250K. That is 5.3 wins a year today, 7.9 if half the late pursuits had started early, and a $656K difference. The interactive page runs your own numbers.
You know the six steps a public project runs, and you know the winner is usually settled at the third, when the design consultant writes the scope, long before a solicitation exists. What most firms do not have is a way to be at step one for every agency they sell to. Portal alerts and bid trackers start at step five, when every competitor sees the same document the same morning.
The blue station is where the winner is usually settled. Everything to its left is public paper a reading system can read on a schedule. Everything in the salmon box arrives the same morning for everyone.
Your turn. At which step does your firm usually first hear about a project?
None of this is hidden. A capital plan is a public document, a board agenda is published before the meeting. The constraint is attention: the documents are free, dated, public, and far too many for one person to read. Reading at that volume, on a schedule, without tiring, is the job a reading system does better than your best person.
Industry research in this seat is reading: plans, budgets, agendas and awards, across every agency you sell to. Four numbers say how much reading that is. The buyers are counted, by the government, and there are ninety thousand of them. Each adopts a plan every year, and each plan moves every project one year closer to a solicitation, in public, with a number beside it.
Ninety thousand plans is a reading job no team staffs, and fifty-two thousand special-purpose governments is where most of the equipment, engineering and construction money sits. Reading one plan takes an evening. Reading every plan you sell into, every year, is not a job a person does. It is a job for a reader that does not tire, and that reader writes down more than a person has time to: one record per project, the first year a dollar appears, the funded year, the published value with its date, and the page it came from. Where the document does not say, the record says Unknown.
Your own number is the one you ran in section 01. The prompt below reads one plan of yours tonight.
The call goes badly when the agency mentions a consultant you have never heard of, or a project you thought was new and they approved two years ago. Everything they said was public. It was in four documents you did not have open at once.
So a reading system built for this seat keeps one record per project and stacks every public mention onto it: the plan line, the board minute that approved the study, the news item, the last award for this kind of work. Every fact carries the page it came from. When the same project appears under three names in three documents, it proposes the match and a person confirms it before two records become one. You walk in knowing when it first appeared, who is designing it, and what they bought last time.
The record a reading system built for you would keep. Nothing here was hidden. It was spread across four documents, and one record with its sources is the difference between four browser tabs and one page.
Every opportunity asks two questions. Does this fit us? How far along is it? Most lists answer both with one number, and the collapse hides the best opportunities: an early strong fit reads as low relevance because it is vague and years away; a live weak fit reads as urgent because it has a deadline. Combined into one score, the thing built to find work sooner buries the work it found sooner.
So a reading system keeps two scores, and that is a design decision you make before anything is built. Fit is scored against criteria your firm wrote down, with weights, and it does not change because a deadline appeared. Stage is where the project sits on the six steps, and it changes every time a new document mentions it. A strong fit that is early is the one worth finding. A strong fit that is late, everybody already sees.
Watched, not chased. It folds into the wider net.
One score ranks by deadline, so the wrong bid is on top and the best one is last. Two scores put the plant expansion where a person can decide about it two years early.
The worry under every conversation about this is simple: will software decide what we pursue? No. Software reads the agenda, pulls the fields, matches the same project across three documents, scores it against criteria you wrote, and keeps the record with its sources. Then it stops. A person decides go or no-go, enters what the firm would earn, chooses the reason for a no, and picks up the phone to the agency.
One test decides where the line sits on any step: is the machine working inside what the documents already contain, or guessing outside them? A funded year the plan states, one project matched across three documents, a score against criteria you wrote: inside, and the record is the check. Which agency moves first, what a board does with a budget it has never faced: outside, where no document can catch a wrong answer. Read inside. Decide outside. And never tune the criteria so tightly around this year's thinking that the reading stops surfacing what you did not expect.
A no is soft, and the reason decides what comes back. Some reasons are final: not our kind of work. Some hold until the documents come out: no capacity this quarter. Some hold until the project moves again: deferred. Write the reasons down once, attach a return rule to each, and a record that returns arrives with a note saying why.
And one number is never the machine's to fill in: what the firm would earn. A person enters it, and the record stores the amount, dated, never a percentage of the project, so a revised project value does not silently rewrite it. The pipeline sums those amounts and nothing else.
Everything so far is capture. When the agency publishes, the job runs on a clock, and the reading changes with it: now it is one document read against another, the same way every time. Three addenda in four weeks is ordinary. Each can move a date or revise a section your draft already answered, and a person misses one once in a career, which is once too often.
This is also where the tools most firms already own begin, and they are good at organising the response. The seat this page is written for is the earlier half, where the whole function is one or two people and the reading is the bottleneck.
Reading an addendum against your draft is the same job as reading a plan: extraction, comparison, a note on everything that changed, and a person deciding what to do about it.
Every firm that sells to public agencies sits somewhere on this curve, and no stage is a judgement on the firm. Most firms we talk to are on the first two. Mark the one that sounds like your Monday, and the answer lands underneath, which is never "buy everything".
Software ranking a list against criteria nobody wrote down is noise. Software ranking it against your criteria, with your amounts and your dates, is a forecast. Stage four is only reachable from written criteria and your own amounts.
That is what the window is worth while nothing changes. Adjust the inputs and this follows.
The four prompts on this page read one document at a time, which is as far as a chat window goes. Your territory is hundreds of documents a week. Reading every plan, agenda, budget approval and funding announcement for every agency you sell to, every week, and holding one record per project across all of them, is a reading system's job. Schedules, readers, a score against your own criteria, a queue, a record that shows its sources, a meeting view that assembles itself overnight. A person still makes every call.
Switchboard builds that kind of system as custom software, around how a team already works. The way in is the AI Jumpstart: fixed scope, fixed price, one workflow, and in this seat that workflow is almost always the finding. Two weeks of yours, four steps of ours, and your current way of working runs untouched the whole time.
Which sources get checked, where the signal is lost, and where the judgement stays with a person. Drawn with the people who do it.
The CRM, the portals, the spreadsheet, and what a reading system would need from each. Nothing in production is touched.
Your criteria, a ranked list with a source on every row, and a person deciding. If it does not hold up there, it will not hold up in production.
Phased from your reactions to the prototype, priced per phase. Yours to run with us or without us.
Everything above is yours. The map, the audit, the prototype, the roadmap, whether or not you build with us. And the criteria you wrote down are what all of it reads against.
Thirty minutes. No pitch, no deck. A conversation about which agencies you sell to, where you enter today, and what reading their plans on a schedule would change. Book a call: withswitchboard.com/book-a-call
This playbook also lives as an interactive page, with a calculator for your own numbers and four prompts to paste into your own AI. withswitchboard.com/playbook/pre-rfp
We built the every-week version for a group of manufacturers that sell to public agencies. The case study shows what it reads and what a person decides; the prototype page shows the software itself, module by module.
456 public sources read on a schedule, one ranked queue, and a person at every decision. Built for a group of three manufacturers, against how they already found work.
Read the case study