— Funding & accelerator program design

Accelerator and funding
programs, designed.

We design the selection, the cohort structure and the funding mechanics for accelerators, grant windows and investment-readiness tracks — plus the evidence trail underneath them.

Built on the screening design the World Bank published: 2,000 firms interviewed before a single seat was allocated.

Book a 30-minute design call See how 2,000 firms were screened

Under about fifty applicants, a scoring model is theatre. We will say so.

The SPARK Innovation Hub sign-up screen beside a panel reading Your next opportunity starts with who you meet
A curation decision, running live. The line under the button reads “membership is reviewed — every profile is a real founder, NGO or expert.” We built that platform and still host it.
0Decisions made on a criterion the applicants never saw
3Arms a cohort can be split into, so a funder can measure the programme rather than report it
1Scorecard, applied identically to every applicant and visible to your panel
YoursThe screener, the scorecard and the scoring rubric transfer at close

What the engagement gives you. Volume — applicants screened, firms funded, startups accelerated — sits in section 9 with the people who did it, where it evidences capability instead of opening with our own output.

— How selection goes wrong

Nobody argues about the money.
They argue about who got it.

Three failures we have watched from inside the room. Each one produces a cohort. Only one of them produces a cohort you can explain.

  1. First past the post

    The window opens, four hundred applications arrive in forty-eight hours, and the cohort is whoever had the fastest connection and the emptiest week. It looks like demand. It is a measure of who was already well-connected enough to be watching, which is usually the opposite of the mandate.

  2. The panel that already knows people

    A room of experienced, well-intentioned people choosing from memory and impression. It produces a decent cohort and no record. Six months later a rejected applicant, a board member or an auditor asks why one firm was in and another was out, and the honest answer is that nobody wrote it down.

  3. Criteria written after the shortlist

    The rubric appears in the final report, reverse-engineered from the firms that were chosen. Every criterion fits perfectly, because each one was picked to fit. This is the version that passes a light review and fails a real one, and it is far more common than anyone says out loud.

All three share a root cause. Selection was treated as an administrative step before the programme, when it is the single design decision that determines what the programme can possibly achieve.

— What we built

Two thousand firms,
one identical question set.

On the World Bank's Jordan programme the selection ran before the intervention was designed, at a scale that made an impression-based process impossible. The numbers below are in the published report.

  1. Build the frame before the window opens

    Micro and small businesses across Jordan were listed first, with a specific effort to reach refugee-owned firms that conventional business registries do not capture. You cannot select fairly from a list that already excludes the people in the mandate.

  2. Interview, then screen — in that order

    Two thousand firms were interviewed. A screener application went to all of them and 1,946 completed it. Interviewing first is slower and it is what makes the screener answerable by the people it is aimed at.

  3. One criterion that decided the shape of everything

    Whether the firm would actually participate in an online B2B platform. A single, checkable, declared-in-advance question — and the programme was built around the firms that said yes rather than the firms that looked best on paper.

  4. Assign in batches, and keep the assignment

    The survey ran in batches through 2018 and 2019, and each completed batch was assigned to treatment and control groups. Assignment happened as the programme ran rather than being reconstructed for the report at the end.

  • 2,000Firms interviewed before selection
  • 1,946Completed the screener
  • ~2,100Businesses in the baseline survey
  • 400Selected for hands-on support

Every figure from the World Bank's published report — Figure 1 and section II. Twenty-one per cent of the applicants were funded, and that ratio was known before the window opened rather than discovered at the end.

The Directory page on the SPARK Innovation Hub with tiles for Users, Posts, Pages, Groups, Events and Blogs
Selection does not end at the intake form. On the SPARK Innovation Hub the directory is what lets a funder find the organisations it screened, years later, without asking anyone.
Around twenty women and one child standing behind two tables covered with handmade soap, crocheted goods and preserves
One of the firms that came through that screen. The Syrian Jasmine Workshop went on to take an international order for 50 kg of handmade soap through the platform — a cooperative that a first-past-the-post window would very likely have missed.
— The one thing worth arguing about

A cohort can answer
a question, or just
receive a service.

Almost every accelerator selects a cohort and supports it. That produces beneficiaries and a completion rate. It cannot tell a funder whether the support did anything, because there is nobody to compare the cohort to.

Splitting the intake into arms costs almost nothing at design time and is impossible to add afterwards.

Three-arm assignment from one screened poolA single screened pool splits into three arms. Treatment one receives the platform only. Treatment two receives the platform plus hands-on design support. The control arm receives neither, and is what makes the other two measurable. All three are drawn from the same screener, on the same criteria, at the same time.1,946screened on one criterion setT1 · the platformListed, searchable, able to receive a quotation requestarm oneT2 · the platform and hands-on supportDesign mentoring, product guidance, marketing supportarm twoC · controlNeither. The reason the first two can be read at allarm three
The assignment the Jedad programme actually used — T1, T2 and a control, drawn in the report's own Figure 1 as an automated randomisation step running on each completed survey batch.

What each arm buys the funder

The control arm is the one that gets cut from every budget, and it is the only one that turns a delivery report into evidence. Without it, a programme can report that 400 firms were supported and nothing at all about whether the support mattered.

Two treatment arms rather than one answers the question a second funding round actually turns on: was it the platform, the hands-on support, or only the combination?

And where this one fell short

The arms were built. The impact evaluation was never funded — the report says so directly, noting the follow-up survey would happen “should additional resources become available.”

So the structure exists in a published document and the analysis does not. We would rather you read that here than find it yourself. It is also the argument for costing the evaluation into the programme budget at design time, which is the conversation we now open with.

— Live on this page

Weight the criteria.
Watch the cohort change.

Seven product categories from the Jedad programme, with the outcomes each one actually produced. Move the weights and the ranking re-sorts. This is the argument for fixing weights before a window opens, in the only form that survives contact with a sceptic.

Every value is from Table 2 of the World Bank's published report. Scores are each category's share of the best performer on that measure, weighted as you set it. Nothing here is modelled.

  1. 1Wood products77
  2. 2Stuffed crochet toys63
  3. 3Prayer rugs44
  4. 4Photography services17
  5. 5Scented candles12
  6. 6Pottery8
  7. 7Leather goods8

Weighted evenly, wood products lead. Pull every weight to zero except the B2B pipeline and prayer rugs take first place on a single JD 100,000 line.

Which is the point. A programme that had funded on pipeline alone would have backed a category that trained seven producers, and a programme that had funded on training volume would have missed the one that sold the most. The weights are the strategy.

— Cohort structure

One cohort, three
depths of support.

A flat cohort spends the same on the firm that needed a document and the firm that needed six months. Tiering the support inside a single selected cohort is how a fixed budget reaches further without lowering the bar at intake.

A page from the Arabic design guide for wood producers, showing three inlay stone patterns drawn as line diagrams beside explanatory text
Tier one. A page of the wood-products guide. Nine of these were produced, and 150 firms received them without attending anything.
A full room of business owners, most of them women, seated facing a presenter and two screens reading test fast, test often in Arabic
Tier two. A capacity-building workshop in Amman. 200 firms went through the group sessions.
A small group seated around a low table while a mentor holds up a woven textile for the maker sitting opposite her
Tier three. One-to-one, product in hand. 50 firms received a series of these.

Guides and reports · 150 firms

The cheapest tier, and the one that scales.

Written recommendations and video walkthroughs produced by the design team, sent to firms that never sat in a session. Nine product-category guides were produced on the Jedad programme and this is who received them.

Right for firms that already trade and need a specific correction rather than a relationship.

Group workshops · 200 firms

Where a method gets transferred.

Capacity-building sessions covering the concept of design, knowing your customer, prototyping and testing. Delivered in a room, at cohort scale, with the work on the walls.

Right for a cohort that shares a problem and will learn faster from each other than from a document.

One-to-one · 50 firms

Expensive, and the tier that moves products.

A series of sessions with the design and marketing team, product in hand. Participants brought revised designs back to the next session, which is what separates mentoring from advice.

Right for the firms whose product is close enough to market that a redesign changes the outcome.

Control

No support. A design decision.

Selected on the same criteria, at the same time, and deliberately not supported in this round. It is the arm that makes the other three legible, and it is the one a steering committee will ask you to remove.

Right for any programme expecting to make a case for a second round of funding.

The tier sizes above are the Jedad programme's, from section IV of the published report. They are not a template — the ratio between them is the budget decision, and it should be argued about before intake rather than settled by whoever shouts in month four.

— How we work

Criteria first.
Applications after.

The order below is the whole method. Once a single application has been read, no scoring rubric written afterwards is fully credible, however good it is.

Weeks 1–2Agree the questionWhat the funder has to be able to say at the end, in one sentence. Every criterion below is derived from it, and anything that is not gets cut.
Weeks 2–3Build the frameWho is reachable, who is systematically missing from the registries, and what it takes to reach them. On the Jordan programme this meant listing refugee-owned firms that no conventional database held.
Weeks 3–4Publish the rubricCriteria, weights and tier sizes, fixed and shared before the window opens. This is the step that makes every later decision defensible, and it is the step most programmes skip.
Weeks 4–8Screen and assignThe screener runs, scoring runs against the published rubric, and assignment to tiers and arms happens in batches as they complete rather than in one pass at the end.
Delivery, then afterRun, measure, hand overOutcomes recorded against the arms as the programme runs. At close, the screener, the rubric, the scores and the assignment record are yours — including the rejections.

A rubric written after the first application is read is a description of a decision already made.

— Who designs it

Two people, and
the reason it is them.

Programme design at this level is a small-team discipline. These are the two who do it here, and the credentials are checkable.

Portrait to come1:1 · team photography

Majd Shams

Innovation & startup development

Twelve years in innovation management, with an MBA and an MSc in the field. Majd has taken more than 140 startups through incubation and acceleration programmes, and specialises in design thinking, business model innovation and investment readiness. He is a principal architect of the programme design methodology this page describes.

  • 140+ startups
  • Investment readiness
  • Design thinking
Portrait to come1:1 · team photography

Tareq Malas

Co-founder & COO · AI and strategy

Six years of senior consulting at McKinsey and a PhD in AI and data science. Tareq is responsible for making sure an engagement is grounded in data rather than opinion — which on a selection design means the arm structure, the scoring model and whether the thing can actually be measured at the end.

  • McKinsey
  • PhD, AI & data science
  • Measurement design

Accelerator partners we work alongside: Flat6Labs, iPark, BeVisioneers and the Mercedes-Benz Fellowship. We publish no figures against those relationships, because none has been agreed with them for publication — the named work with figures on this page is the World Bank programme.

— When to call someone else

Four times this is
the wrong spend.

Said here rather than on the call, so you can rule us out in two minutes.

  • You have fewer than about fifty applicants. At that size a panel reading every application carefully will beat a scoring model, and the model exists mostly to be shown to somebody. Run the panel and write down the reasons.
  • The cohort is already chosen. If the firms are known and the process is a formality, we would be building a document that dresses a decision. That is a real service and somebody sells it. We do not.
  • You are deploying capital, not designing a programme. Due diligence, term sheets and portfolio construction are an investment discipline. We design the selection and the cohort structure around a fund; we do not make the investment decision.
  • The evaluation is the deliverable. If what you need is a randomised controlled trial with a published paper at the end, that is an economist's job and there are teams who do it properly. We build the programme it can be run on, and we work alongside them.
Umbrella500 went beyond the brief. They understood our mission, moved fast, and delivered a platform that brings our community together like never before.
Samer HaffarSPARK

That platform now carries 500+ organisations and the engagement became a three-year partnership. Selection design is the part of it nobody photographs, and it is the reason the directory is worth searching.

— Questions

The ones that decide it.

Our funder will not pay for a control group. What then?

Then you do not have one, and we design the strongest programme available without it — which usually means a staged rollout, so the firms funded in round two act as an imperfect comparison for round one at no extra cost.

We will put the cost of a control arm in front of you once, with what it buys, and then build what you decide. It is a real trade and it is yours to make.

Who owns the scoring model at the end?

You do — the screener, the weights, the rubric, the scores and the assignment record, including every rejected application. That last part matters: a selection process you cannot re-run is a process you cannot defend, and the rejections are where the questions come from.

Can you run the selection, or only design it?

Both. On the Jordan programme we were part of the team that listed firms, interviewed them and ran the screener at two-thousand scale. If you have delivery capacity in-house we would rather design it, train your team and step back.

How do you keep a screener answerable by the people it targets?

By interviewing before writing it. The Jordan screener followed two thousand interviews, which is why 1,946 of them came back completed. A screener drafted in a capital city and mailed to a governorate has a completion rate that quietly re-selects your cohort for you.

What if the applicants are nothing like the proposal assumed?

That is common and it is recoverable if the rubric is published, because you can show exactly what changed and why. Adjust the weights in the open, record the decision and the date, and tell the applicants. What is not recoverable is a quiet re-weighting after the scores are in.

Do you work in Arabic?

Arabic is where we start. The team is in Amman, the Jordan screener ran in Arabic, and the platform the selected cohort worked in was built in Arabic because the international alternatives offered no Arabic support to the people expected to use them.

How do we start?

A thirty-minute call about the mandate, the money and the window. If there is a fit, the first deliverable is the question the funder has to be able to answer at the end — everything else is derived from that one sentence.

— The brief

Tell us what the window
has to achieve.

Send the call for applications if it is drafted, or the mandate and the budget if it is not. We answer in a day, usually less.

Or write directly: [email protected]

Opens your mail app with the details filled in. Nothing is stored on this page.