Agentic
Friday, 4.40 pm. An enquiry comes in via your form. Who replies, and when? If the honest answer is ‘it depends on who’s got time’, then you don’t have a lead problem. You have a process problem masquerading as a lead problem.
Thursday, 5.20 pm. A Head of Operations in Munich books a demo request for 120 seats. Who calls back, and when? If the honest answer is "whichever SDR happens to be online", you do not have a pipeline problem. You have a process problem dressed up as a pipeline problem.
Friday, 4.40 pm. A shipper asks for a rate on three fixed runs a week into the Ruhr. Who prices it, and when? If the honest answer is "the planner, once the last trailer has left", you do not have a lead problem. You have a process problem dressed up as a lead problem.
Thursday, 5.10 pm. A scale-up asks through your website for a quote on a new customer portal. Your account manager is at a client, your tech lead is mid-sprint. Who calls back, and when? If the answer is "whoever has room first", you don't have a lead problem. You have a process problem disguised as a lead problem.
Friday, 4.40 pm. A German buyer sends a specification for a special-purpose machine through your web form. Who replies, and when? If the honest answer is "once engineering has a moment", you don't have a lead problem. You have a process problem dressed up as a lead problem.
30 mins · senior consultant · no slide deck
Your inbound enquiries come in via different channels: email, web forms and LinkedIn. Three inboxes, three owners, no shared timer.
Your inbound arrives in pieces. Demo requests in HubSpot, chats in Intercom, trial sign-ups in-app. Three sources, three teams, no shared clock.
A rate query via the web form, a spot load on the freight exchange, a tender in the owner's inbox. Three doors, three owners, no shared clock.
Your inbound lands in three places: the partner directory, extra-work requests in the ticket system, an email to info@. No shared clock.
Your enquiries arrive split. RFQs to sales@, leads from your agent in Cologne, quote requests via the service desk. Nobody owns the clock.
As a result, the initial response depends on your diary rather than on your process. Whoever happens to be available determines how quickly a prospect receives a reply.
That mechanism is self-perpetuating. Latency feels like a problem that requires effort, so you solve it by putting in the effort: someone who takes a quick look in the evening.
The mechanism keeps itself alive. Slow follow-up feels like an effort problem, so you fix it with effort: a founder scrolling the Slack alerts for new sign-ups at 11 pm.
The mechanism feeds itself. A slow quote feels like a busy week, so the owner fixes it with effort: Sunday evening at the kitchen table, pricing lanes from last year's rate sheet.
The mechanism feeds itself. Slowness feels like an effort problem, so the managing partner reads info@ on Sunday evening and asks a senior developer to scope between two sprints.
The mechanism feeds itself. Slow replies feel like a lack of effort, so you fix them with effort: the project manager who reads an RFQ on Sunday evening.
Workload does not scale and varies from week to week. Whether it’s a release, a holiday or a peak period, response times increase precisely when your volume rises.
Effort does not scale. A Product Hunt launch, a new pricing page, the last week of the quarter: your response time climbs exactly when the board is watching pipeline.
Effort does not scale and varies by week. In the Q4 peak and the November tender season your response time slips, exactly when most shippers ask for a rate.
Effort does not scale. In the go-live week of your biggest project, over the summer and in December, when clients spend leftover budget, response time climbs as demand rises.
Effort doesn't scale. A factory acceptance test on site, the week after Hannover Messe, the summer shutdown: that is when response time slips, as enquiries rise.
Work it out using your own figures, over a full month. Divide the number of leads that received an initial substantive reply within four hours by the number of inbound leads you received that month. The denominator is the total number of leads received, not the number you replied to. If 40 leads came in and 17 received a reply within four hours, that’s 17 divided by 40, or 42.5 per cent. The remaining 23 are what you really need to focus on.
The second-order cost is included in that figure of 23. Multiply it by your own probability of winning the sale and your own average order size. That amount doesn’t appear in any report, because a lead that never received a reply isn’t recorded as a loss anywhere.
Your own figures based on your own inbound traffic, not a measured benchmark.
Answer them based on your own CRM export and what you’ve seen happening this week, not on a hunch.
The loop is PRAL. Perceive: it reads incoming messages from your form, your shared mailbox and your LinkedIn messages, and puts everything in a list.
The loop is PRAL. Perceive: it reads your demo form, Intercom chats and new trial accounts with their first usage, German ones too, and puts it all in one queue.
The loop is PRAL. Perceive: it reads requests from your web form, planning inbox and freight exchange and puts them in one queue, lane and loading date attached.
The loop is PRAL. Perceive: it reads your contact form, info@, leads from the partner portal and tickets tagged as new work, and puts them all in one queue.
The loop is PRAL. Perceive: it reads your web form, sales@ and your trade-fair leads, pulls the specification from the attachment and puts everything in one queue.
Reason: he asks three to five screening questions about the budget, timeline and service required, and scores the answers according to criteria that you set.
Reason: it asks about seat count, the current tool, when that contract ends and who signs, then scores the answers on axes you set, such as ICP fit and expected ACV.
Reason: it asks about lane, weekly volume, load profile and start date, and scores the answers on axes you set, such as: does this fit the reefer that stands idle on Tuesdays?
Reason: it asks about stack, deadline and fixed price versus time and materials, then scores on axes you set: does it fit your margin, will it grow into managed services?
Reason: it asks three to five questions on specification, required delivery date and capital approval, and scores the answers on axes you set.
Action: he confirms receipt and assigns the hot lead, along with the rough responses, to a team member. The warm lead enters a follow-up sequence, rather than becoming a cold lead. Learn: any correction made by your sales representative to a score is fed back into the system.
Act: it confirms the request and hands the 150-seat team with an expiring competitor contract straight to your AE, raw answers attached. The eight-seat startup goes into a nurture sequence instead of going cold. Learn: every score your AE corrects goes back into the system.
Act: it confirms receipt and puts the shipper with fixed weekly volume Nijmegen–Hanover, raw answers attached, before your owner that afternoon. The one-off request for four pallets enters a follow-up sequence instead of going cold. Learn: every score your planner corrects feeds back in.
Act: it confirms receipt. A SaaS company seeking managed services after its supplier walked away reaches the managing partner within the hour, raw answers attached. A web shop with budget for next year gets a follow-up sequence. Learn: every score your account manager corrects goes back in.
Act: it confirms receipt and hands the hot lead, a pharma plant with approved budget and a fixed validation date, to your sales engineer. The feasibility study without budget enters a follow-up sequence. Learn: every score your salesperson corrects goes back into the system.
The polarity is shown here once, and is identical throughout. High is a better fit. Low is a less suitable fit. If a lead scores 0 on service fit, the agent stops and no one follows up on it.
On the autonomy ladder, this is L2, with L3 at the routing stage. The acknowledgement of receipt and the first round of enquiries can be handled autonomously. The score remains a proposal that a human reviews. Quoting a price, sending a quotation, making a commitment: these will always remain tasks for humans. Capgemini estimated in 2025 that around 15 per cent of processes would be running at L3 or higher, rising to 25 per cent by 2028. Anyone promising more than this for this workflow is engaging in ‘agentic washing’.
Capgemini, 2025
A prospect who enters their name, company and telephone number is a data subject. The GDPR applies from the moment that information is entered, not just when a contract is signed. Four things are established before the agent goes live.
In 2026, Deloitte surveyed 3,235 leaders across 24 countries for its State of AI report. 74 per cent want agentic AI by 2027, whilst 21 per cent have mature governance frameworks in place for it. It is within this gap that these projects fail.
Deloitte, State of AI, 2026
No implementation. You record the baseline measurement, note down the three axes and their polarity, and specify who is authorised to approve an override.
Gartner predicts that by 2027, around 40 per cent of agentic projects will be scrapped due to a lack of clear business value. You can prevent this by setting a target figure in advance.
And one more thing, because it really does happen. As soon as you log latency, you can see for each lead who picked it up and when. That’s tracking people, and I’m not going to pretend it isn’t.
Therefore, agree that your sales staff will be the first to see their own logs, that the figures will be reported by team rather than by individual, and that the benefit is that overtime will be eliminated. Set out these terms before the first assessment, not afterwards.
Gartner, 2025
Your unit of measurement is the number of demo requests per week. Your tool isn’t the form, but your product log: has this applicant already logged in, and what have they interacted with? Your timer runs in minutes, because the prospect is currently using your product.
Your unit of measurement is the quote request per lane. Your tool is the request itself: origin, destination, volume, frequency, and whether it is a one-off journey or a contract volume. Your clock is the planner’s day, as your rate depends on capacity, which fluctuates on a daily basis.
Your unit of measurement is the RFQ with its technical appendix. Your tool is that appendix: do the requested specifications fall within the capabilities of your machinery? The clock is ticking in working days, as the decision is made by an investment committee rather than by a single person.
This afternoon, you can check whether you can even set up the qualification exam using a template you already have. If that doesn’t work, there’s nothing to automate yet. In that case, you’ll end up buying a system that makes guesses at questions which you yourself haven’t properly defined.
First, the boundary – and this is deliberately placed above the instructions. Do not enter the names, email addresses, telephone numbers or company names of real prospects here. Work from memory and describe leads as categories. Do this using a business subscription where you know in which region the data is processed, not a consumer account. As long as there is no data processing agreement in place, no personal data should be entered.
Je bent mijn sparringpartner voor het scopen van een lead-kwalificatie-agent. Stel mij een vraag tegelijk en wacht op mijn antwoord voor je verdergaat. Vraag mij achtereenvolgens: 1. Via welke kanalen komt mijn inbound binnen, en wie leest die nu? 2. Welke drie vragen stelt mijn beste verkoper altijd als eerste? 3. Waaraan herken ik achteraf een lead die nooit klant had kunnen worden? 4. Wat is mijn huidige doorlooptijd tot eerste inhoudelijk antwoord, en over welke noemer meet ik die? Lever daarna vier dingen op: A. Vijf kwalificatievragen in mijn eigen woorden, in de volgorde waarin een prospect ze krijgt. B. Drie scoringsassen met de polariteit expliciet. Neem letterlijk over: Hoog is beter passend. Laag is minder passend. Scoort een lead 0 op dienstfit, dan stopt de agent en gaat er geen mens op. C. De situaties waarin de agent moet stoppen en escaleren naar een mens. D. Drie testleads die ik zelf kan doorlopen: een die hoog moet scoren, een die 0 op dienstfit moet scoren, en een waarbij de agent hoort te twijfelen. Is een van mijn antwoorden te vaag om op te scoren, zeg dat dan en vraag door. Verzin geen cijfers die ik niet heb gegeven.
If the model gets stuck at question 2 or 3, your problem isn’t a technical one. It means your sales process hasn’t been formalised yet, and that’s what RENEW does first.
If there are fewer than about ten inbound leads a month, it doesn’t pay off. That’s our estimate, not a measured benchmark. At that volume, your latency is a scheduling issue, and you sort it out with a meeting, not with a system.
Nor does it form part of a sales process where the first step is a technical discussion that only the founder can conduct. And it does not form part of a process that is not documented anywhere.
It does not belong in an enterprise deal where the first call is a security review only your CTO can lead. Nor does it belong on a sales playbook that exists only in the founder's head.
Nor does it belong in a contract-logistics tender where the first step is a warehouse walk-through with your operations director. And it does not belong on a rate structure that exists only in the owner's head.
It does not belong in a tender or a framework deal with your largest client, where the first step is an architecture session with their CTO that only you can lead. Nor on a process nobody has written down.
Nor does it belong on a retrofit enquiry where the first step is a technical call only your chief design engineer can take. Or on a quoting process nobody has written down.
Don’t set it up if you don’t want to say ‘no’ to any applications. A qualification agent who never turns anyone away is simply a bottleneck with extra steps.
I’d be happy to arrange a meeting where we can work out your response time over a full month and determine whether an agent would help resolve this.
I would be glad to book a call where we work out your speed-to-lead per channel over a full month and decide whether an agent shortens your CAC payback.
I am happy to set up a conversation where we work out, over a full month, how long your rate requests wait and whether an agent solves anything.
I'd be glad to set up a call. We reconcile your response time over a full month, the way you would a project, and decide whether an agent solves anything here.
I'm happy to set up a call where we time the first reply to every RFQ from last month and decide whether an agent solves anything here.
30 mins · senior consultant · no slide deck