How to Plan a B2B Product Launch Backwards From the Pipeline Target

On 11 April 2026, at IndoSales Bengaluru, I launched three Claude skills in front of 271 sales people: "Three Claude Skills, built over two weekends, launched live on stage". My note afterwards: "The real signal was not the applause. It was what hit my DMs over the next 48 hours." (LinkedIn, 15 Apr 2026, people unnamed.) A launch is judged by the conversations it starts.

As VP Global Performance Marketing at Locus: $4.2M pipeline in H1 CY22. At FieldAssist, the FAi launch GTM (gtmexpert.com, last updated Jan 2026).

The usual failure is the order of work. The date, the assets and the calendar get planned first, and the commercial number arrives last, as a slide. Nobody can then say whether the plan can reach it.

Working backwards fixes the order, if it reaches two things launch plans often skip: what the number is worth, and how many accounts can supply it. A sum that stops at leads is a budget request, not a test.

Five moves, in this order. The sums are plain multiplication, so a spreadsheet can do them. The tools save drafting, and the judgment in each move is yours.

The event planner skill supplies no conversion rate or benchmark, and with web tools it looks up public holidays. The launch commander in CRAFT GTM starts from your launch type, sector choice and launch date. It cannot read your funnel or your calendar: your targets sit in its goals table and do not size the tasks. The PR and launch pack builds a claims register before any copy.

Step 1. Start above pipeline, then count the accounts

About this example. Gnani.ai is a real company. Gnani.ai has not reviewed this guide. Facts about Gnani.ai come from public sources, linked alongside with the date I checked them. The launch, the product capabilities, the buyer and customer types, the dates and every number marked illustrative are inputs I chose to show how the method works; they are not Gnani.ai's data or plans, and the tool outputs are exercise results, not Gnani.ai's real strategy. The "I" in these steps is an exercise role. This is a worked example, not a client result.

What is public: at the India AI Impact Summit 2026, Gnani.ai launched a voice model that Outlook Business names Inya VoiceOS. On 31 March 2026 it raised a $10 million Series B, the first close of a larger round, for global expansion. Company-reported: more than 200 enterprise customers, with banking and financial services a large share of deployments.

On 6 October 2026, in a release from Dubai and Bengaluru, it took its partner programme to the Middle East. Company-reported: three regional partners, "multiple banking and insurance customers in the last few months" in the UAE, "fine-tuning available for regional languages including Arabic", and deployment on a regulated client's own infrastructure. The release gives no customer count or conversion figure, and does not name Saudi Arabia.

The exercise, all of it illustrative: if you ran go-to-market at a company like Gnani.ai, how would you launch a voice agent for loan collections and customer service to banks and lenders in the UAE and Saudi Arabia? Treat the UAE as a small base and Saudi Arabia as new. No rate is published for either, so every rate below is a first test.

Start above pipeline. The illustrative goal is 3 signed deals at $150,000 of first-year value, $450,000 in all. At an illustrative 15% win rate that needs 20 opportunities, or $3.0 million of pipeline in the two quarters after launch. At an illustrative 1 in 2 from accepted lead to opportunity, that is 40 accepted leads.

Define the stages and have sales sign them. First stage: a named person at a target account agrees a dated next step. Accepted: sales agrees in writing that the account fits and will work it. Opportunity: an account with a sponsor and a scoped pilot, valued at the first-year contract the pilot would become, not the pilot fee.

Count who can supply that, each account once. Wikipedia's UAE list shows 23 national banks by my count. Its Saudi list gives 39 licensed banks: 11 local, 4 digital and 24 foreign branches. Redo this from each regulator's register. Leave out the branches, add an illustrative 40 sizeable lenders, and the list is about 80 accounts.

So 40 accepted leads is half the list and 20 opportunities a quarter of it, in six months, through bank vendor reviews. The buyer, a Head of Collections, exists once per institution, and four roundtables of 15 would seat about three in four of the list. My reading: the illustrative $3.0 million is a stretch. I keep it as the ceiling and take the low case in step 2 to finance. With a list of 40, the target would fail here, before any budget is spent.

Step 2. Cost the routes, print a low case, stage the spend

Events owe half here: 20 accepted leads, an assumption to agree with sales. My reading: a bank often lets a new supplier in through someone it has already approved, so test the partner share first.

Input to the event planner skill, all illustrative. Target 40 accepted leads, events' share 50%. Two roundtables, Dubai and Riyadh: 15 guests each, 100% relevant, 50% to first stage, 50% accepted, $20,000 each. One regional fintech show, visited with meetings booked: 40 contacts, 60% relevant, 30% and 40%, $25,000. Two webinars: 120 registrants each, 40% relevant, 10% and 30%, $3,000 each. Its script returned the first table alongside, ending "Gap: 6.7 accepted leads short of the share events must deliver".

The skill asks how to close the gap rather than closing it: more roundtables, a larger share for outbound and partners, or a lower target. The decision is yours.

My decision, not the tool's: add roundtables in two stages. Commit the April pair and hold an illustrative $40,000 for a June pair. Rerun with four roundtables, the script returned an illustrative 20.8 accepted leads for $111,000. That clears 20 by only 0.8, and the show's 2.9 count only if they can become opportunities by about 23 September. Without them, events stop at 17.9. Choose a show by the end of June, or move its illustrative $25,000. And 15 guests means seated, so invite more.

The other 20 accepted leads need a plan too. I ran named-account outbound and two partners through the same script, in the second table. Together that is an illustrative 40.8 accepted leads for $146,000, about 20 opportunities. These routes show $1,750 per accepted lead against $5,347 for events only because salaries sit outside the sum. Tag each lead with its account and report distinct accounts, since the same banks will appear in several rows.

One line is not a forecast. Cut the first stage, accepted and opportunity rates by a third and the plan falls to the low case in the last table: about 6 opportunities and $0.9 million for the same money. My suggested rule: show finance both lines, and approve an illustrative $106,000 now, not $146,000.

Set an illustrative cost ceiling too. Say finance will spend an illustrative $150,000 for that $450,000: $3,750 per accepted lead. The plan costs about $3,600 and passes. The low case with the pair held costs about $7,200 and fails. Cutting the show, the dearest line at $8,681, still leaves about $6,000 per accepted lead ($81,000 over 13.5), so that case stays over the ceiling. Salaries, in-region hosting, Arabic testing, counsel and a local seller sit outside the $146,000. Price those too.

Now the test for the held money. Accepted leads are a poor early test: with 30 guests at a true 25%, 8 or more come up only about half the time. Test the faster stage. At the pipeline review on 6 May, release the illustrative $40,000 only if at least 12 of the 30 guests, from different accounts, have reached first stage. On my sums, illustrative, a true 50% rate passes about 9 times in 10, and the low case's 1 in 3 about 1 time in 4. Below 12, the money stays held and the forecast moves to the low case. Do not shift the shortfall to outbound on the same accounts; ask a partner for new ones.

This is the point where I would want a second pair of eyes: the target fails the account count, or the low case exceeds the finance ceiling. Bring me your account count, low case, staged spend and launch readiness gates, and we will work it through in 45 minutes.

Step 3. Check the date before the plan hardens

The first illustrative launch idea was mid February 2027. The skill opened a published forecast: Ramadan 2027 in the UAE is "expected to begin on February 8, 2027", with Eid Al Fitr likely on 9 or 10 March. It asked whether to move the launch after Eid, for example to Tuesday 23 March 2027, or keep February with online formats only.

I moved the illustrative launch to 23 March and dated the plan in the table alongside. MATCH means the weekday is right and says nothing about holidays, and the dates depend on the moon sighting, so confirm the list for each market.

That forecast covers only the UAE. The Saudi holidays alongside are SAMA's schedule, as Argaam reported it.

Step 4. Clear readiness, then draft the tasks and correct them

Readiness comes before any invitation. A bank's first questions are where the call audio sits and how the agent performs on its own dialect. The release reports resident deployment and Arabic fine-tuning; neither answers for a given bank. Clear the gate in the first table by Monday 15 March, the day before invitations. If a row fails, the April pair moves.

The gate needs local facts. These are my additions, for your counsel to apply. UAE standards allow telephone collection contact only "during the hours of 9 AM to 8 PM". Saudi rules cap calls at ten every thirty days per financing product, with notice that the call is recorded. A UAE bank's customer Confidential Data "must not be shared outside the UAE without Central Bank approval and obtaining prior written consent from the customer". For hybrid and public cloud, Saudi controls say "in principle only cloud services should be used that are located in Saudi Arabia". A February 2026 central bank note on AI is, as Hadef & Partners summarised it on 8 April 2026, not legally binding, and says "Customers should be informed when they are interacting with AI systems". Read the note itself.

Input to the CRAFT GTM launch commander, all illustrative: product "An illustrative voice AI agent for loan collections and customer support, in Arabic and English, for banks and lenders, built on an agentic AI platform", launch date 2027-03-23, launch type market expansion, segments "Retail banks in the UAE, Banks and consumer lenders in Saudi Arabia", channels email, LinkedIn, webinar, events, PR and partners, team of two to five, moderate budget, industry fintech, your sector AI native, and goals "$3.0M pipeline in two quarters, 40 sales accepted leads, 20 opportunities, 3 reference customers, with the Head of Collections as the buyer".

What came back: five phases, a buying group, risks and a checklist. Keep those as a first draft; a task list built from steps 1 to 3 would do the same job. Its goals table reads "$3.0M pipeline in two quarters" from the CRM and says to "inspect qualified opportunities, exclude closed revenue".

Then fix the task order, in the second table. Due dates are "suggestions spread across each phase". The pilot task allows about 8 weeks from 12 January to a signed bank pilot, through a vendor risk review. My reading: that does not fit a bank. Here pilots follow the April roundtables, with first signatures aimed at September, illustratively. A signed pilot has no results yet, so change the goal from 3 reference customers to 3 signed pilots in the window. The illustrative $450,000 arrives after the window, if those pilots convert.

The commander also sets a "Launch event" on 23 March, which step 2 neither dates nor budgets. Invitations three to four weeks ahead would fall inside the Ramadan forecast, so drop it. The April roundtables are the launch event, with invitations from 16 to 23 March.

That leaves a harder question. 23 March now has no customer, no event and no result, the failure this guide began with. My call: keep it only as a plain note the invitations can point to, and save the real announcement for the first bank that agrees to be named. If invitations need no public page, skip the note.

Step 5. Let the release say only what is checked

Input to the PR and launch pack: the core release and the launch sequence for the illustrative launch on 23 March 2027, with the launch plan and the event plan as context, the headline idea "The first Arabic voice agent built for Gulf banks", an illustrative pilot result (a lender cut call handling time by 30%), and "use our public scale figures".

What came back: a claims register before any copy, in the table alongside. The "first" claim was removed, "Unverified: no source opened in this session states it in this scope". Arabic stays flagged: the release supports fine-tuning as the company's statement, not this agent on Gulf calls. The illustrative 30% result was held as "User to confirm", with "customer approval needed" and no baseline or period.

The register is the asset to keep, not the draft. Strike any drafted quote words nobody said. Step 4 puts no Gulf pilot before launch, so the 30% line comes out, and this illustrative release carries no Gulf result.

1. Write the revenue goal, win rate and stages, and have sales sign them.
2. Count your target accounts, and test the target against the count.
3. Run the sums on your rates, print a low case and set the test for held money.
4. Check the launch month against holidays in every market.
5. Clear hosting, language, the review pack and capacity before any invitation.
6. Draft the tasks, fix the order, and run your claims through a register before anyone drafts copy.

1. Starting from the date, not the pipeline target.
2. Never counting the accounts behind the target.
3. Reading one line of home-market rates as a forecast.
4. Spending before hosting, language and vendor review answers exist.
5. Putting a "first" or a pilot result in a release without a source, a baseline and approval.

• How Much Sales Should a Product-Led Company Add? Scoring EPIC Motions Per Segment, Then Sizing and Anchoring Them
• How I Check Product-Market Fit, Churn and ICP Gaps Before Scaling Paid Demand Generation
• All 24 guides