--- title: "How we closed $4,320,000 of AI deployment work for an implementation consultancy in six months. | Stone Haven Capital Group" description: "24 signed engagements at an average of $180,000 each. Six months. An AI implementation consultancy. No new partners, no conference booth, no waiting for a model vendor to pass along a name." canonical: "https://stonehaven.capital/showcase/ai-implementation-consultancy" last-updated: "2026-08-22" --- > 24 signed engagements at an average of $180,000 each. Six months. An AI implementation consultancy. No new partners, no conference booth, no waiting for a model vendor to pass along a name. Stone Haven Capital Group Schedule a Call Case study · AI Implementation Consultancy # How we closed $4,320,000 of AI deployment work for an implementation consultancy in six months. 24 signed engagements at an average of $180,000 each. Six months. An AI implementation consultancy. No new partners, no conference booth, no waiting for a model vendor to pass along a name. - $4,320,000 generated in 6 months, from 24 AI deployment engagements at an average of $180,000. - 621,943 emails to 124,389 companies, producing 1,121 replies at 0.9%. - 269 interested conversations became 86 held meetings and 24 signed engagements. - What did most of the work: we kept every cold email off the consultancy's own domain. - What went wrong: the sequence ran longer than the buying cycle. 01 ## Did they get a good result? Here it is with nothing sanded off. In the six months before we arrived, comparable new business came to $1,394,000 and arrived through vendor introductions and reputation. Six months later the firm had signed 24 engagements worth $4,320,000, a 3.1x increase, with $12,212,000 of qualified work sitting behind that. What happenedThe number Closed engagement revenue$4,320,000 across 24 signed engagements Average engagement value$180,000 Growth against the prior six months$1,394,000 before, a 3.1x increase after Qualified pipeline behind the closes$12,212,000 across 86 opportunities Earlier-stage interested pipeline$25,824,000 across 269 conversations Emails sent621,943 over six months Leads contacted124,389, of which 123,145 were new Replies1,121, a 0.9% reply rate, or 3,811 and 3.1% counting out of office Positive replies269, 24% of all replies Bounce rate1.4%, 8,707 bounces, with inbox placement at 93.0% Six months of sending, as the platform recorded it. 02 ## Did you get them in front of the right people? This is the only question in the document worth arguing about. 621,943 emails prove nothing by themselves, because most of any market bought model access and is perfectly content never shipping anything with it. 24 engagements reached signature carrying $4,320,000, an average of $180,000 each. Behind them sit 86 qualified opportunities worth $12,212,000 and 269 earlier conversations worth $25,824,000. Targeting shows up in the conversion rather than in the send count. 1,121 replies produced 269 positive ones, 86 of those reached qualified stage and 24 signed. Close to a third of every positive reply became a scoped opportunity, which does not happen in a market where nothing is stalled. One month of booking: twenty conversations created, fourteen of them held. 03 ## What worked, and why ### We kept every cold email off the consultancy's own domain. The firm's own domain carries delivery, statements of work and everything existing clients rely on, and none of that was going anywhere near cold volume. We stood up 157 dedicated sending domains carrying 472 inboxes, each authenticated on SPF, DKIM, DMARC and MX before anything went out. Warmup ran on all 472 for the life of the programme, and only part of the estate carried campaign volume at any point. Whatever was held back stayed idle and came into rotation as the domains already running picked up age. Each inbox was capped at 20 sends a day. Across six months the estate carried 621,943 sends and finished at a 1.4% bounce rate, 8,707 bounces in total, with inbox placement holding at 93.0%. Monitoring logged 0 errors and 8 alerts over the whole run. At this volume those figures are the programme. Let them drift and the domains go, and everything standing on them goes at the same time. Every sending domain authenticated, and the client's own domain absent from all of it. The domain that talks to paying clients never carried a single cold email. All of the risk sat on infrastructure that was built to be thrown away. ### We built the list around firms that already bought the technology and then stalled. The people who sign are VP Engineering, Chief Data Officers and heads of automation at companies between $200M and $2B in revenue, concentrated in financial services, insurance and logistics. That profile framed the list. What narrowed it was public evidence that a model programme exists and has not reached anybody outside the building. The exclusions did as much work as the inclusions. Companies with no announced AI activity came out, and so did anyone whose own product is a model or a platform, because that group builds this internally and will never buy it. Between them those two cuts removed well over half the original pool before a line of copy existed. Buying model access is a purchase. Shipping is an event. Firmographics tell you who could be stuck. The public signals tell you who is stuck right now, with budget already committed to proving otherwise. The first pull, before any deployment signal was applied. The same market once the deployment signals and exclusions went on. ### We opened with a question and nothing else. The first email runs two lines. No greeting, no credentials, no paragraph explaining who is writing. It asks one thing about the state of their deployment and stops there. An engineering leader reads it in four seconds and either knows the answer or does not. Every send went out as plain text, spun across every line, so no two messages leaving the estate were identical. The sequence as it ran, and the two-line opener that carried it. Holding back the credibility clause is what makes that work. A technical buyer treats an unearned claim as a reason to stop reading, and a sender introducing themselves has made one before the ask arrives. A question they can answer from memory carries no claim at all, and answering it is cheaper than deleting it. ### We ran six touches on widening gaps and passed only aligned replies to the delivery leads. Six touches across the sequence, gaps that widen as it runs, four variations on the opener and two on each follow-up, all A/B tested. A qualification step sat in front of the replies so the people who scope and price the work only saw conversations with a real deployment behind them. Senior delivery time is the binding constraint in a firm selling fixed scope. Counting out of office responses, 3,811 replies came back at a 3.1% rate. 269 of them were genuinely positive and 24 became signed engagements. The filter is what kept the remainder away from people who had billable work in front of them. ### We read the results weekly and switched off whatever was losing. Angles were judged on the quality of the reply rather than the count of them. Anything losing came off within days and the winners absorbed the freed volume. At a 0.9% reply rate, the gap between an angle converting a quarter of its replies and one converting none is the entire result. One conversation, from the first touch through to a booked slot. 04 ## The six angles we test, in every market Which message a market will answer is not knowable in advance, and anybody claiming otherwise is guessing confidently. The market decides, so we ask it directly. The opener ships six variations at once, one per angle, and reply quality names the winner. These six travel across sectors because they are built on how a decision gets made rather than on what is being sold. 05 ## 1. The teardown Point at one specific thing that is visibly half-built and describe it plainly. It proves somebody looked, and it makes replying cheap because they already know the answer. You announced the internal assistant in March and it still sits behind a staff login. Is that a model problem or an integration problem? 06 ## 2. The status quo challenge Name the default course of action and ask whether it is really holding. Lands hardest when the default is expensive and everybody has quietly stopped believing in it. Most teams we see are on their second internal AI hire before anything reaches a user. Is that the plan on your side as well? 07 ## 3. The benchmark Place them against comparable firms on one measurable thing. Not knowing where you sit is uncomfortable, and the ask is only to see the number. We looked at how long it takes insurers your size to get a retrieval system past legal review. Want to see where yours would land? 08 ## 4. The short window Attach the ask to a window that closes on its own. Deadlines invented by the sender get ignored. Ones the buyer already lives inside do not. If this is meant to be in production before the budget year turns, the build has to start within a few weeks. Is that the timeline you are working to? 09 ## 5. The direct pitch State exactly what is being sold, what it costs in time and what exists at the end. Some markets reward the visible absence of a manoeuvre. We deploy internal agents and retrieval systems on fixed scope in six months. If that is something you are trying to get done this year, worth a conversation. 10 ## 6. The referral sideways Give them permission to hand you to somebody else. The answer takes one line, and the name it returns arrives with an internal endorsement already attached. If production deployment sits with someone else on your side, happy to be pointed their way instead. 11 ## How we use them All six go out together across a split list. Within a few weeks the reply data has already named the one or two the market wants, and the rest come off before they spend any more contacts. Which one wins is unpredictable every single time. That one of them will win is not. Running six at once turns that from a bet into arithmetic. The same six port to LinkedIn on shorter wording with the underlying angle unchanged. 12 ## What did not work, and what we did about it Three things. We would rather put them in front of you than let you find them on your own. ### The sequence ran longer than the buying cycle. Six touches over five weeks, aimed at a market that makes this particular decision in about ten days. The last two touches landed after the buyer had already picked a direction, which means they arrived as noise from a vendor who had evidently not been paying attention. What we changed: the deployment signals and the hard exclusions went on, which took the pool from 376,566 down to 144,638. After verification and deduplication, 124,389 contacts were loaded into the sequencer, and the sequence was compressed so every touch landed while the decision was still open. ### Two variables were tested at once. In the same week we changed the subject line and the opening angle. Reply volume moved and there was no way to say which change moved it. Three weeks of data was discarded because none of it could be read. What we changed: one variable at a time from that point, each held long enough to produce a readable result before anything else was touched. ### The sender identity mismatched the offer. Emails went out under a generalist consulting sender name. The audience here is engineering leadership, and engineering leadership wants to hear from somebody who has shipped a system into production. The name on the email answered that question badly before the first line was read. What we changed: sending moved to a technical sender with delivery history behind the name, and the opener stopped explaining who that was, because the signature was already doing it. The opener as it first went out, under a sender name the audience did not recognise. A programme this size does not run clean, and any document reading as though it did has had this section deleted. The corrections are the part that repeats on your engagement, which makes them the part worth showing. 13 ## Why this works for AI implementation specifically Every consultancy in this market describes the same problem. The work is good, the clients who have had it delivered stay, and new work arrives through vendor introductions and reputation. That holds until the partners are busy, and then reputation becomes the ceiling. Three features of this market do most of the work: - The stall is public. Announcements, job postings and conference talks say a model programme exists months before anything reaches a user, and the silence afterwards says it never did. Both halves of that are searchable. - The engagement carries the cost. At an average of $180,000 on fixed scope, the whole programme pays for itself on one or two signatures. Very few channels have that arithmetic. - The money is already spent. Model access is bought, budget is committed and nothing is in production. Nobody needs convincing that the category matters, which removes the expensive half of the conversation. 14 ## Why would this work for your business? Possibly it does not, and learning that in week one is far cheaper than learning it in month four. The table is the fastest way to check. This works ifThis does not work if New work arrives through vendor introductions and reputation, and the partners are the bottleneckYour delivery calendar is already booked past where you can staff it There are thousands of companies you could name that you have no route intoYour entire market is a few dozen accounts you already know by name A single engagement is worth six figures on fixed scopeYour economics cannot absorb a programme that pays back on a small number of signatures Something public tells you a company has bought the technology and not shipped itNothing separates a company that needs you this year from one that never will Somebody can qualify replies before they reach the people who deliverEvery reply has to land on a delivery lead's desk Land in the left column and what transfers is the method. Nothing about this client was unusual. A consultancy with real delivery capability and no way of reaching the firms that need it beyond the people who already knew them. All of it came out of process, and the process does not change for you. 15 ## One more thing worth understanding An AI pilot that quietly succeeds is a worse position for the buyer than one that fails outright. That reads backwards, and it is the thing this entire offer is built on. A pilot that fails gets closed and the budget is released. A pilot that works gets praised in a review deck and then sits behind an internal login, because a working demo and a production system are separated by integration, security review, evaluation and somebody who owns the pager at three in the morning. The organisation now believes the problem is solved. Nobody funds solving it twice. So the buyer we reach is rarely somebody who failed. It is somebody holding a success nobody can use, with the money already gone and the credibility already claimed in public. That shapes the offer completely. Fixed scope, a defined end date, and one system running in production when it closes. What is being sold is the second half of a project the buyer already started and cannot be seen to restart. 16 ## Before and after Before After Source of new workvendor introductions, inbound and reputation an outbound engine running alongside all three Comparable new business in six months$1,394,000 $4,320,000, a 3.1x increase Signed engagements from outboundnone 24, at an average of $180,000 Qualified pipeline behind themnothing tracked $12,212,000 across 86 opportunities Outbound volumenone 621,943 emails to 124,389 leads 17 ## If you want to know whether your market has this in it A short call covers it. Describe the firms you want as clients and what a signed engagement is worth when it lands. We come back with how many of those companies are actually reachable, the conversation volume that is realistic against that pool, and a straight answer on whether outbound is the right instrument here. If the answer is no, that is what you will get on the call. Neither of us gains anything from six months of a programme that was never going to work. ## Want to know whether your market has this in it? The first conversation is short. You tell us who your buyers are and what one is worth to you. We tell you how many we can actually reach, what the meeting volume looks like, and whether outbound is the right lever for you at all. If we think it is not, we will say so. Book a consultation call → ## Read another one [Aerospace & Defense Tier-One Supplier](https://stonehaven.capital/showcase/aerospace-defense-tier-one-supplier/)[How we built a $31M+ active program pipeline for a tier-one aerospace and defense supplier.](https://stonehaven.capital/showcase/aerospace-defense-tier-one-supplier/)[Boutique Capital Advisory Firm](https://stonehaven.capital/showcase/boutique-capital-advisory-firm/)[How we built a $39.4M capital mandates pipeline for a boutique capital advisory firm.](https://stonehaven.capital/showcase/boutique-capital-advisory-firm/)[All case studies](https://stonehaven.capital/case-studies/)